As robots advance in terms of dexterity and other physical capabilities, it becomes more likely that humans may find themselves working alongside them. If that happens, how will robots’ emotional capabilities need to advance for them to successfully work with people?
In a recent study, researchers trained collaborative robots to read human emotions by not only accounting for facial expressions, but also contextual factors in the interactions as well. Through experiments with 40 volunteers, the researchers then evaluated how a robot’s ability to read human emotions and adjust its behavior in turn impacted a human’s perception of the robot and its capabilities as the two collaborated on tasks. The results—which show that the emotional capabilities of robots only go so far with humans—were published 18 May inIEEE Robotics and Automation Letters.
Seung Chan Hong led the study as part of his undergraduate thesis while studying at Monash University, in Melbourne, Australia. He notes that, while there has been a lot of hype in the advancing physical abilities of robots, this is only one piece of the puzzle. “We need to also innovate when it comes to them actually interacting with humans, not just their physical capabilities,” he says.
This prompted him to dig deeper into the emotional aspects of human-robot interactions. First, Hong and his co-authors decided to train a robot to read human emotions using a vision language model (VLM), which is similar to large language models (LLMs) such as ChatGPT, but which can also take visual inputs.
Training VLMs for Human Emotion Recognition
To evaluate their VLM, which used Gemini 2.5, the researchers had volunteers watch videos of robots handing over objects to humans—with varying degrees of success—and describe the emotions the humans were expressing. Importantly, the volunteers labeling these videos were able to take into account more context in these interactions, rather than reporting solely on the facial expressions of the humans in the video. For example, a person pausing to think with a furrowed brow may simply be concentrating on their task at hand and not necessarily be angry. Contextual factors such as drumming their fingers, pursing their lips, or other behaviors can point to the real cause of a person’s furrowed brow.
The researchers then compared their VLM to a conventional AI system that relies on standard facial analysis and object tracking that is used in human-robot interactions. They found that the VLM outperformed the traditional approach. On a scale from 0 (no similarity in meaning to the emotion identified by the human volunteers) to 1 (a perfect match in meaning), the conventional AI system achieved a score of 0.77. In comparison, the VLM achieved a score of 0.86.
Hong says, “I think [the VLM] was able to align with what human observers were seeing a lot better, because it wasn’t just looking at the person’s face for a brief amount of time, but seeing the whole scene—where the person was and what they were doing, and how they were interacting with the robot.”
In a second experiment, the research team asked 40 volunteers to interact with a robot using their VLM—but purposefully programmed the robot to make an error. The robot then had to offer either an emotionally adaptive apology that accounted for the human’s perceived response to the mistake or a pre-scripted spoken apology.
Participants overwhelmingly preferred the emotionally adaptive response, with 31 out of 40 people favoring this approach over a boilerplate apology.
However, their survey responses underscored how this emotional adaptivity was far less important than the robot’s functionality. After collaborating with a robot that failed in its task, many participants ranked their trust in the robot as lower, regardless of how it apologized for its mistake. “A personalized apology acts as a social lubricant, but it cannot repair the trust lost by the robot failing its physical task,” Hong says.
Interestingly, the VLM classified the emotions of its human partners similarly to human volunteers who observed an interaction from a third-party perspective. But when the VLM’s assessments were measured against humans’ self-reported emotions during the second experiment—the most accurate descriptions of their true emotions—its ability to accurately predict emotions dropped significantly.
“While the VLM is a good observer of outward social cues, it isn’t a mind reader,” Hong says. “It matched third-person human observers well, but it didn’t always align with the users‘ internal, self-reported feelings.”
Together, these results show that robots are not perfect at reading human emotions. So while people might appreciate their efforts, they still ultimately will want competent co-workers.
This story was updated on 15 June 2026 to correct where the research was conducted and clarify that the researchers evaluated the performance of a pre-trained model.
Yen-Ling Kuo always wanted to understand how things worked. When she was growing up in Taiwan, reading the story of Michael Faraday in elementary school piqued her curiosity about the natural world. During that time, she was introduced to Logo, a computer program with a turtle cursor to help children learn basic coding through hands-on experimentation.
It was Kuo’s introduction to programming logic.
Yen-Ling Kuo
Employer
University of Virginia in Charlottesville
Title
Assistant professor of computer science
Member grade
Member
Alma maters
National Taiwan University; MIT
In high school she learned the capacity computers held. She could write programs that completed tasks independently, she realized.
“Once I discovered how powerful computers could be,” she says, “I knew I wanted to focus on using them to solve real-world problems.”
Kuo, an IEEE member, never lost her interest in the “how” behind processes and tools. Her curiosity, combined with a stint working at a Silicon Valley company, led her to focus on innovations that live at the intersection of cognitive and computer sciences.
Kuo’s winning paper, “Diff-DAgger: Uncertainty Estimation with Diffusion Policy for Robotic Manipulation,” demonstrates a novel method to help robots better identify and estimate uncertainty when faced with scenarios on which they’ve not been trained. The method reduces the amount of human supervision, improves a robot’s rate of successful task completion, and opens up a path to introduce more complex models with bigger data demands into interactive robot learning.
She says her research will help people working in the robotics and automation fields more efficiently collect the data needed for effective model training.
Silicon Valley’s impact
Kuo earned bachelor’s and master’s degrees in computer science at the National Taiwan University, in Taipei, in 2009 and 2012. As she was nearing completion of her master’s degree, she did what many computer science graduates do: She pursued a summer internship at a tech company.
She spent the summer of 2011 at Google’s campus in Kirkland, Wash., working on the company’s comparison ads project.
As she was considering pursuing a Ph.D., a call from Google changed her plans. The company offered her a full-time role as a software engineer.
“I viewed the job offer as a positive development,” she says. “I believe it can never hurt your future research career to get some real-world experience under your belt.”
She was hired in 2012 and helped build techniques that incorporate computer vision and natural language processing to improve the customer shopping search experience. She led the company’s Shop the Look initiative, a predecessor to Google’s current AI-powered shopping experience. The project connected social media content with search results, something the company had struggled to do in the past.
Kuo and her team were tasked with building a connection between the natural language people use to describe an item and an image that matches the searcher’s intent. It was at a time when the neural network—using deep learning models to power Google products—was gaining momentum at the company. Integrating neural network tools into her work was a requirement—which raised questions for Kuo.
“I was applying the neural network tools,” she says. “But I didn’t have 100 percent certainty about how they actually worked.”
She considered how she could become more knowledgeable about deep learning models. It was a full-circle moment. She decided that after nearly four years at Google, it was time to earn a Ph.D. in computer science. She returned to MIT in 2016.
When the two met, Katz asked Kuo why she wanted to pursue a doctorate degree. She explained her interest in understanding how neural networks work and in using that knowledge to connect the physical world with human language.
He suggested she attend a summer course at MIT’s Center for Brains, Minds, and Machines, a research initiative that ran from 2013 through 2025. CBMM’s objective was to bring together computer scientists, cognitive scientists, and neuroscientists to understand how human intelligence works. The goal was to use the resulting insights to establish an engineering practice to build artificial intelligence systems.
For Kuo, it was a chance to better understand human intelligence and identify ways it could be replicated in machines.
“It was an opportunity for me to interact with other scientists and gain insight into how people learn, understand, and figure things out in the world,” she says. “I saw it as a very useful and inspiring way to incorporate those ideas into my own research work.”
During her Ph.D. studies, she was a research assistant at CSAIL. The experience helped shape her doctoral research, which focused on building AI systems that apply past learning to new situations. She developed machine learning models to support the efforts, including language understanding and social interactions.
She completed her Ph.D. in computer science in 2022 with a minor in cognitive science.
After graduation, she continued her work and collaboration at CSAIL, particularly on projects that involved the “theory of mind” concept.
Theory of mind spurs innovation
Theory of mind isn’t new, having originated with primatologists studying chimpanzees in the late 1970s. The theory recognizes that others have their own thoughts, beliefs, and perspectives. It’s a skill that allows humans to infer someone’s mental state and predict their behavior without verbal communication.
“It’s like when college roommates are moving into their dorm. They may not talk too much, but they work together naturally to coordinate their activities and accomplish goals,” Kuo says. “They can infer and mentally interpret each other’s behaviors and signals to make decisions and complete tasks without words.”
She brought her theory of mind research to the University of Virginia when she joined as an assistant professor in 2023.
Kuo conducts her research in UVA Engineering’s multidisciplinary cyberphysical Link Lab. Her broad focus is on developing computational models that help robots interpret both direct data and silent signals, from language and movements to a person’s gaze. If successful, it could give robots the same sort of physical and theory of mind reasoning capabilities that power physical and social interactions among humans.
“There are no computational frameworks yet available that will translate this kind of understanding into a robot efficiently,” she says.
She adds that the process to get there begins with improving how robots learn to perform tasks.
The evolution of robot learning
Historically, one way robots learned was to mimic humans. A researcher would manually guide a robot through a task, like cutting an apple, and it would repeat the movements. The robot was successful until the environment changed, such as when its hand was in a different position or the apple was at a different angle. The robot was then faced with a situation for which it hadn’t been trained. Without any data available to help it correct course, the robot would start making small errors that eventually led to a full system crash.
This diagram describes how the robotic gripper’s visual perception and tactile sensing prevents a potato chip from breaking.Xuhui Kang, Yen-Ling Kuo, et al.
To solve the problem, researchers developed the dataset aggregation (DAgger) method. As a robot performed a task, a researcher was on standby to provide real-time corrections during unexpected scenarios. The correction data was continuously added to the robot’s model, teaching it how to recover from mistakes.
To reduce the human monitoring effort, robot-gated DAgger was created to enable bots to query humans when the machines became uncertain.
The most popular approach to make the query decision is to train multiple models to consider when determining a course of action. If the models all agree, the robot proceeds. If they don’t agree, the robot is likely to get stuck and ask for help.
Although the multiple model approach was widely adopted, it has limitations. Practically speaking, as models become more complex, it is hard or impossible to train multiple copies. A more fundamental issue is that disagreement among models doesn’t always imply uncertainty; it could just mean there are different ways to accomplish a task.
The Diff-DAgger solution
That is the gap Kuo’s research team closed with the novel Diff-DAgger research. The approach builds on diffusion policy, a technique that helps robots account for different ways a task can be performed.
The new method repurposes diffusion loss, the signal a robot uses to improve its model during training, as a real-time confidence check. During task execution, the robot computes the signal and compares it against values from its training data using a statistical test. The signal spikes when the robot faces an unfamiliar situation and is uncertain how to proceed. The signal stays silent when the robot’s current action is close to what it learned before.
The spike represents the robot’s ability to self-diagnose and predict an imminent failure. Human intervention is triggered only when the signal spikes. No spike means the robot can be left to complete its decision-making process on its own.
Kuo’s team achieved significant results: Failure prediction rates were improved by 39 percent. Task completion rates were increased by 20 percent, and tasks were completed nearly eight times faster.
Her research at UVA gained attention from the National Science Foundation, which honored her last year with a Career Award, the foundation’s flagship grant for early-career researchers. The five-year US $665,000 grant supports her research that builds computational models for human-robot interactions through theory of mind reasoning.
She also received the Toyota Research Institute’s Young Faculty Researcher Award to teach cars to reason about interactions on the road and with the driver.
As service robots and self-driving vehicles become more available, such works are likely to make interactions between humans and robots more intuitive and useful.
Kuo ultimately wants to build more robust robots that are able to integrate into a social space with humans by engaging with us through grounded interactions, she says.
“It was a natural segue to transition from student to a full IEEE member,” she says. Today she is an active volunteer with the IEEE Robotics and Automation Society, a reviewer for submitted papers, and a presenter and panelist at conferences.
She says one of the best parts of attending conferences is having the opportunity to engage with students. She also enjoys participating as a panelist at luncheons, she says, because it gives her one-on-one time with student attendees. She can share her knowledge and offer insights as they prepare to embark on their career.
Her goal in the coming years, she says, is to broaden her involvement with IEEE initiatives and branch out to other technical committees. Sharing knowledge and learning from others is essential to anyone’s career growth, she says, and “IEEE offers a great opportunity for both.”
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
We present MotionDisco, a framework that discovers contact-rich, long-horizon humanoid loco-manipulation motions from scratch, without relying on teleoperation or motion retargeting from human demonstrations.
Some of the discovered behaviors are a little nutso:
It turns out running a workout class is a decent way to stress-test whole-body range of motion. Coordinating fluid movement across every joint at once—timing, velocity, balance compensation—is one of the harder control problems in humanoid robotics.
At AIRoA, we’re bringing robots into real homes. Check out our exclusive first video to see how they work in our development hub and real-life household settings! The project aims to develop home robots that can assist people with everyday tasks and become long-term companions in daily life. In this video, we demonstrate Toyota’s Human Support Robot (HSR) deployed in real homes, where it assists residents with everyday tasks such as tidying rooms and fetching objects.
MIDAS Hand is a fully open-source, tactile-sensor-integrated dexterous robotic hand platform for manipulation, teleoperation, and robot learning research. MIDAS stands for Modular low-Impedance Direct-drive Anthropomorphic Sensing Hand.
This video presents a novel flight maneuver for a flying bipedal robot. During forward flight, the robot performs aerial braking by swinging its legs to adjust the orientation of foot-mounted thrusters.
In this time-lapse, controllers on the ground are repositioning Dextre, our robotic handyman currently installed at the end of the Canadarm2. They used Dextre to unload equipment from the unpressurised Dragon trunk. Such a beautiful choreography to watch with Earth in the background!
This video demonstrates how AI Sapiens learns and performs humanoid motions from video-based motion capture using only a smartphone camera, without professional motion capture equipment. ROBOTIS plans to release an open-source motion generation and learning pipeline for AI Sapiens, enabling users to generate humanoid motions from video and bring them to the real robot.
Tumblenauts are a swarm of minimalist, bacteria-inspired robots designed for collaborative inspection of pressurized microgravity habitats such as the International Space Station. Unlike current intra-vehicular robots that rely on complex actuator-dense mechanisms for precise motion, the Tumblenauts use a stochastic run-and-tumble locomotion and collective cooperation inspired by bacterial colonies.
LUMOS Robotics Founder and CEO Yu Chao officially introduces Project EDGE—inviting global builders, universities, robotics labs, and creative technologists to explore the future of humanoid robotics together. To supercharge the global developer community, we are providing 100 complimentary LUMOS NIX robots to selected global partners.
How do you progress from early childhood computational thinking to advanced high school robotics? Sphero’s product offerings are intentionally scaffolded to scale for students by building critical skills and concepts at every grade level.
An examination of how socially assistive wellness robots could support the seven dimensions of senior wellness, and how a framework can measure their autonomy.
What Attendees will Learn
Why the senior care crisis exceeds incremental automation. Demographic pressure, workforce shortages, and a daily wellness-programming gap all strain traditional care models.
What defines a wellness robot as a category. The seven ICAA wellness dimensions and eight properties separate these robots from companion and medical devices.
How autonomy can be measured with CRAS. This six-level scale, modeled on the SAEJ3016 driving standard, evaluates four care dimensions.
What maps the road to full autonomy. The paper examines technical capabilities, clinical evidence, and a three-phase roadmap toward the early 2030s.
Throughout the exhibition hall at the 2026 IEEE International Conference on Robotics (ICRA), in Vienna, one demonstration seemed to attract a disproportionate amount of attention.
Two robotic hands were making a balloon dog. Slowly and deliberately, the robot twisted a long balloon into loops, bends, and joints without popping it. Visitors stopped, watched, and often returned with colleagues to watch again.
AGILINK’s balloon dog demonstration draws a crowd at ICRA 2026.AGILINK
At first glance, the demonstration appeared almost playful. Among roboticists, however, balloon twisting is widely recognized as an unusually difficult manipulation task.
A balloon is lightweight, highly deformable, slippery, and extremely sensitive to force. Every twist changes its geometry and internal pressure, turning a seemingly simple activity into a continuously changing physical interaction problem.
Humans navigate those changes almost intuitively. While making a balloon animal, people rarely think consciously about force regulation, slip prevention, or contact stability. They simply adjust.
For robots, those adjustments remain remarkably difficult. The challenge is not merely moving fingers to the right positions. The harder part is maintaining stable interaction while the object itself is changing.
Highlights from AGILINK’s ICRA 2026 demonstrations, including visuotactile sensing, in-hand manipulation, balloon-animal shaping, and other contact-rich tasks enabled by the company’s latest OmniHand platform.AGILINK
That distinction helps explain why the balloon dog drew so much attention in Vienna. What appeared to be a dexterity demonstration was, in many ways, a demonstration about contact itself.
As robotic manipulation continues to advance, a growing number of researchers are arriving at a similar conclusion: many of the hardest problems in robotics begin only after contact occurs.
Motion and Contact Intelligence for Robot Manipulation
Balloon twisting combines two challenges that robotics has traditionally struggled to solve simultaneously: long-horizon task execution and contact-rich manipulation.
The first concerns motion.
A balloon dog is not created through a single grasp or twist. It emerges through a carefully ordered sequence of manipulations, each setting the conditions for what follows. A small rotational error introduced early may appear insignificant at first, yet several steps later it can prevent the final structure from forming altogether.
In that sense, balloon twisting is a long-horizon task. Success depends not only on performing individual actions correctly, but also on preserving the future feasibility of the entire manipulation process.
To address this challenge, AGILINK began by collecting demonstrations from professional balloon artists. Human actions were mapped onto robotic hands to establish an initial manipulation policy. But successful demonstrations alone were insufficient.
In practice, some of the most valuable learning occurred when execution began to drift toward failure. Whenever instability emerged, human operators intervened and corrected the manipulation in real time. Those interventions were recorded and incorporated into reinforcement-learning cycles, allowing the system to learn not only how successful demonstrations unfold, but also how experienced operators recover when things start to go wrong.
Through this process, the robot gradually acquired the capabilities required for long-horizon task execution—a collection of abilities that AGILINK groups under the term motion intelligence: the ability to generate actions, coordinate bimanual behaviors, and execute extended manipulation sequences under real-world uncertainty.
OmniHand 3 Ultra-M on display at ICRA 2026.AGILINK
Yet motion alone does not explain why balloon twisting remains difficult. The second challenge is contact.
The robot must continuously regulate force, adjust contact locations, and respond to subtle changes in the object’s state. These decisions are difficult to encode through explicit rules. Even skilled human operators often rely on tactile intuition developed through experience rather than consciously articulated strategies.
Analysis of those interventions revealed that many failures did not originate from incorrect action sequences, but from the breakdown of contact itself.
To better capture those interaction dynamics, AGILINK collected contact-centric intervention data and incorporated those interactions into reinforcement-learning training. Rather than learning only which motions to perform, the system also learned how humans maintain stability when contact conditions begin to deteriorate.
AGILINK describes this capability as contact intelligence: the ability to establish, maintain, and adapt physical interaction as force distribution, friction, deformation, and contact geometry continuously evolve.
The distinction between the two capabilities is subtle but important. Motion intelligence determines what the robot intends to do. Contact intelligence determines whether it can continue doing it. For balloon twisting, both are necessary. One provides the sequence of actions. The other keeps those actions physically viable.
YouTuber KhanFlicks follows OmniHand’s motions while learning to fold a balloon dog at the AGILINK booth.AGILINK
Between a balloon slipping away and a balloon bursting lies a narrow region of stability. Successful manipulation depends on finding that region—and remaining within it throughout the task.
Introducing the OmniHand 3 Ultra-M Dexterous Hand
The balloon dog demonstration showcased a manipulation capability. It also revealed a broader question. How much contact intelligence can be achieved through learning alone? A robot can only regulate what it can perceive. It can only respond as quickly as its hardware allows.
As manipulation tasks become increasingly complex, researchers are finding that progress depends not only on better policies, but also on richer sensing and faster physical response.
That realization formed the backdrop for AGILINK’s second major announcement at ICRA 2026. Alongside the balloon dog demonstration, the company introduced the OmniHand 3 Ultra-M.
OmniHand 3 Ultra-M closely matches the size of an adult human hand.AGILINK
The two exhibits represented different stages of the same technological trajectory. If the balloon dog demonstrated what contact intelligence can already accomplish today, Ultra-M was designed to explore what contact intelligence may require next.
Its most distinctive feature is a fully direct-drive architecture. By adopting direct-drive actuation throughout the system, the hand is designed to enable faster and more transparent force regulation and higher force-control bandwidth, enabling faster response as contact conditions change. For contact-rich manipulation, responsiveness can be as important as sensing itself.
By adopting direct-drive actuation throughout the system, the OmniHand 3 Ultra-M is designed to enable faster and more transparent force regulation and higher force-control bandwidth, enabling faster response as contact conditions change.
The platform also incorporates tactile sensing across nearly the entire hand. Each fingertip contains a miniature vision-based tactile sensor, while more than 300 three-dimensional tactile sensing points are distributed throughout the palm. Together, they provide information not only about where contact occurs, but how contact is evolving.
The system is designed to estimate pressure distribution, shear forces, local deformation, slip tendencies, and other interaction dynamics that often remain invisible to conventional position-based control systems.
According to AGILINK’s tests, individual sensors achieve force resolution of approximately 0.005 N—roughly equivalent to detecting the weight of a sheet of paper resting on a fingertip. Spatial resolution reaches approximately 0.04 mm, while sensing density approaches 50,000 sensing points per square centimeter.
OmniHand 3 Ultra-M recognizes feather texture through vision-based tactile sensing.AGILINK
For dexterous robots, contact has traditionally been a largely hidden process. Ultra-M is designed to make that process more observable.
Rather than simply detecting that contact has occurred, the system attempts to resolve where interaction is happening, how forces are distributed, whether instability is beginning to emerge, and how manipulation strategies should adapt in response.
The balloon dog offered a glimpse of what contact intelligence can already accomplish. Ultra-M explores a different question: what capabilities may be required to push contact intelligence further?
The Physical World Remains the Hardest Benchmark
The significance of contact intelligence extends far beyond balloon animals. Many tasks that continue to resist automation involve unstable or deformable interaction: cable insertion, garment handling, flexible packaging, delicate assembly, connector mating, tool use, and household manipulation.
These tasks are difficult not because robots cannot reach the correct location, but because maintaining stable interaction after contact begins remains extraordinarily hard.
For decades, robotics achieved many of its successes by reducing uncertainty. Factories were engineered to make robotic motion predictable, repeatable, and highly structured. The physical world behaves differently.
A growing share of robotics research is shifting toward interaction itself—understanding how robots can establish, maintain, and adapt physical contact within environments that remain fundamentally unpredictable.
Objects shift. Materials deform. Friction changes. Contact evolves. Real environments rarely follow scripts. Seen through that lens, the balloon dog was never really about the balloon dog. What attracted attention at ICRA was not simply a visually impressive demonstration, but what it revealed: intelligence in the physical world is ultimately measured through interaction.
As motion generation continues to mature, a growing share of robotics research is shifting toward interaction itself—understanding how robots can establish, maintain, and adapt physical contact within environments that remain fundamentally unpredictable.
For robots moving beyond structured environments and into less predictable real-world settings, managing contact may become as important as motion itself.
Thirteen years ago last August, I was camped out in NASA’s Jet Propulsion Laboratory press room in Pasadena, Calif., waiting to see whether the Curiosity rover would survive its descent and skycrane-assisted landing on the surface of Mars. It did, and it was awesome.
Since then, Curiosity (also known as Mars Science Laboratory) has traveled nearly 37 kilometers, drilled into and sampled 42 different rocks, and as of publication has snapped nearly 763,000 photos. The fact that this robot is still hard at work, getting real science done at the age of 13, is absolutely incredible—not only is Mars an actively hostile environment for robots, but the only kind of maintenance that JPL engineers can do is to send very, very careful software updates.
Nevertheless, the clever folks at JPL have managed to keep Curiosity safe, warm, mobile, and sciencing, despite well-worn wheels and less and less power every day. One of those folks is Alexandra Holloway, the assistant team chief for engineering operations for Curiosity, who spoke to IEEE Spectrum about keeping Curiosity roving, what its future looks like, and how JPL has used that experience to make rovers like Perseverance even more capable.
How astonished should we be that after 13 years on Mars, Curiosity is not only still doing science, but actually getting more capable?
Alexandra Holloway is the assistant team chief for engineering operations on the Curiosity Mars rover at the Jet Propulsion Laboratory.Alexandra Holloway
Alexandra Holloway: I’m astonished! The longevity comes from a lot of ongoing work. It’s not just that Curiosity was built robustly; it’s also because we’re continuously putting in effort to ensure it can continue to have that lifespan. I think about all the different kinds of embedded systems there are, from cars to refrigerators, and none of them have the kind of longevity that we have with the rover. It’s mind-boggling, and it’s inspiring.
Is the Perseverance rover, which is nine years younger than Curiosity, significantly different in terms of its hardware and software?
Holloway: In terms of hardware, the rovers are actually very similar. Both use a RAD 750 processor and have the same amount of memory. However, Perseverance has an extra processor specifically for visual odometry, which allows it to drive autonomously. This difference reflects their primary mission designs: Perseverance was designed for driving long distances, while Curiosity is a mission focused on sampling as it goes. So Perseverance’s onboard scheduling capabilities are there to optimize its driving. In fact, just last year, Perseverance surpassed Curiosity’s driving distance after only about three years on Mars.
Curiosity Rover Memory and Software Fixes
Do you have some examples of significant tweaks the team has made to keep Curiosity roving?
Holloway: One of my favorite examples comes from a processor anomaly that happened on Sol 2172 [Ed. note: “Sol” is the term for a Martian day—about 24 hours and 40 minutes]. Curiosity has two computers, A and B. We landed on A, swapped to B due to a NAND memory anomaly early on (Sol 200). For years, we were chugging along on B, until one day there was a problem—B booted up, but it couldn’t mount its drive partition. We’d never seen this before. To preserve B’s data, we swapped back to A, which we hadn’t trusted in two thousand Sols. A also had a degraded memory, with only two gigabytes of usable storage space instead of four. We painstakingly transferred data from B over to A and then down to Earth, and eventually we ran out of stuff we wanted to transfer, which was really good, because A then started acting funny in the same way it did on Sol 200. It was acting like its memory was coming unsoldered. That’s bad.
We quickly swapped back to B, formatted it, and got it working again. The problem then became that we couldn’t trust A’s memory at all, but we needed a second computer as a “lifeboat” for diagnostics and transfers if B failed again. We realized we had one other place of memory: where we keep our flight software. We have four copies of the flight software (two current versions and two older versions) in different banks of very small amounts of memory, just 32 megabytes each. What if we just jettisoned the old flight software copies and used that 64-megabyte NOR memory as our file system for computer A?
So that’s what we did. It was so elegant! Computer A is operating with less than 1 percent of its original memory, but we can run a mission on it. A small mission, but we haven’t had to jettison any core capabilities. We can still drive, we can manage data, we can even theoretically do science. Everything works fine, just much slower and much smaller. That flight software release was even called “R-Hope“ because we hoped it would work.
What are the constraints on Curiosity’s lifespan?
Holloway: Our biggest hardware challenge is wheel wear. It looks like we’re driving on this sandy terrain with some rocks in it, and our intuition said that we could just drive over these rocks and they’d get pushed down into the sand and it would be no big deal. But what we ended up seeing was that those little rocks are actually the tips of giant boulders buried in the sand, and they’re razor sharp. Our wheels were getting ripped apart driving over them, especially our front wheels, so we started driving backwards.
We also monitor consumables. We consider the number of times we move our actuators. That’s a consumable. Curiosity hasn’t taken a selfie in a while, and one of the reasons is that it’s really hard on the joint actuators. Our onboard memory is a consumable, but surprisingly we’re not anywhere near our life cycle for memory. Our biggest consumable is power; we have an RTG, a nuclear power source, which decreases its output as it ages.
Newer missions are flying Snapdragon [processors], but Curiosity’s RAD 750 is a power hog. One of the things that we’ve rolled out that’s going really well is a way of reducing the amount of time we spend with the computer powered on, by harvesting time when we finish activities early and going to sleep, which lets us turn off the computers and some of the heating. Another thing we’re looking at is doing stuff in parallel when we’re on, like being able to drive or use the arm while communicating with an orbiter.
So power is decreasing, and that’s causing us to do all this parallelism work and become more efficient and nuanced in the way we operate. But we are not having any degraded science output at this time. Our wheels are still going, our arm is still okay for now, knock on wood. I would say maybe the bottleneck is budget.
Curiosity Rover’s Impact on Future Mars Exploration
What have you learned from Curiosity that will improve future missions?
Holloway: As an embedded flight software person, I think about how we can change, add, or modify software capabilities during the mission. There’s definitely a sweet spot for loading and patching flight software—some of these concepts were pioneered on Spirit and Opportunity and then inherited by Curiosity and Perseverance, making it easier to understand and change the software.
Some of the things that I wish we had now on [the Mars Science Laboratory] include a better understanding of where our power is going. I want to see how much power each component is drawing every minute, so that we could architect a software system that could balance loads better. We have some of this information that was built in by the engineers who designed the rover, but as an operator, I want something slightly different. So if I were building a mission, I would have those discussions earlier and get operators into the room to say, “what do you want your data products to look like?”
The key takeaway for designing future missions is to talk to all your users early in the design process. It needs to happen upfront.
What does Curiosity’s long-term future look like?
Holloway: That’s a conversation that happens, and it’s a really delicate one. We have a lot of science instruments, and a lot of them have to do with contact science and sampling and rely on the arm. If we lose the arm, what science can we still do? Well, we have a lot of remote sensors too, like cameras, environmental sensors, and radiation sensors. All of these things are important for the future of space exploration and humans on Mars.
From a power perspective, our RTG is projected to start degrading science output in the sixth extended mission, but we’re going to be fine through 2035 and potentially even beyond that. So we have a long and exciting future ahead of us. We need to figure out the best way of operating within our constraints, but we’re still kicking.
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
GENE01 designed from scratch and sent to batch production. Two scalable lower bodies. Physical AI deployed for motor control and world-action modeling. All in three months. Generative Bionics is running.
Alex, the newest humanoid robot built entirely by IHMCRobotics, takes its first steps outdoors! This was a significant milestone for our team, especially because Alex is the first humanoid robot developed entirely by IHMC Robotics to venture outside the lab. These outdoor trials were conducted in preparation for a demonstration in Maryland, where Alex later successfully walked completely untethered.
Midweek at ICRA 2026. Cable-climbing robots that work as a squad: CCRobot-S is a team of robots with reconfigurable cable-driven manipulation that collaboratively inspect and maintain long-span bridge stay cables. Parallel operation for speed, morphological reconfiguration for reach.
In 1987, Richard Greenhill, a British photographer who was fascinated by (but had no actual training in) robotics, decided he wanted to build a life-size humanoid that could do useful things, like carrying luggage. He was working at a startup called Intergalactic Robots, but he couldn’t convince anyone there to build such a machine, so he set about building one himself, in his attic.
To help with his project, he organized a weekly get-together of a dozen or so like-minded folks. Every Wednesday night, his wife, Sally, would make a big pot of spaghetti, and the group would tinker with components scavenged from old printers and picked up from junkyards. They called themselves the Shadow Group. They eventually constructed several different robots, but their main project was the two-legged Shadow Walker.
In 1987, photographer Richard Greenhill organized a weekly gathering of DIY enthusiasts to work on projects in his attic, including the Shadow Walker. Richard Greenhill and David Buckley
Greenhill’s friend David Buckley, a robotics and animatronics expert he’d met at Intergalactic, sketched out a rough design based on medical textbooks of human bone structure and muscle movement. The robot’s skeleton, made of maple, was greatly simplified—only one bone in the lower leg and a single wide toe on each foot. The ankle’s double-axis design allowed for two degrees of movement. The knee had no complicating kneecap.
Greenhill didn’t want the robot to use motors, so its movement was controlled using compressed air to extend and contract 28 “air-muscles”—his version of a McKibben muscle, invented in the 1950s to mimic musculature with pneumatics. The muscles were connected to the bones across eight joints (hips, knees, ankles, toes), which provided 12 degrees of freedom.
The robot’s headless torso held the control valves, electronics, and computer interfaces. It stood 168 centimeters tall and 46 cm wide and weighed about 38 kilograms. The group managed to get the robot to stand up reliably and balance itself; it could even regain its center if pushed a little. But walking turned out to be more of a challenge.
Rich Walker joined the group as a teenager and began writing software to get the robot to stand. He was particularly interested in using neural networks to solve balancing problems, although he ran into a number of hardware obstacles, including the unreliability of the sensors and the valves, and the robot’s overall fragility. Over time, Walker and the team developed a standard library of routines to control the robot. Walker wrote a detailed description of the Shadow Walker in 1999, which is available on David Buckley’s website.
The 1st International Robot Olympics
By the time the Shadow Group began developing Shadow Walker, engineers in academia and industry had been working on robotics for several decades. The world’s first industrial robot, the Unimate, debuted in 1961, and in 1967 Donald Michie and others began building a series of Freddy robots to investigate machine intelligence. The IEEE created its first dedicated robotics organization in 1984 when it established the IEEE Robotics and Automation Council, which became the IEEE Robotics and Automation Society in 1987. Also in 1987, the nonprofit International Federation of Robotics was established to promote research, development, use, and cooperation in the field of robotics.
As Shadow Walker pushed the limits for a DIY humanoid robot, industrial humanoids were also gaining ground. In 1986, Honda began working on its experimental (E-series) and later the prototype (P-series) humanoid robots, finally unveiling the P2 in 1996. The P2 stood 183 cm tall and weighed 210 kg. It was the first humanoid capable of stable, autonomous walking. This work eventually led to the development of the groundbreaking ASIMO.
Greenhill’s friend, roboticist David Buckley, consulted medical textbooks to create Shadow Walker’s humanoid design.Richard Greenhill and David Buckley
In the late 1980s, the public was both fascinated and horrified by the potential of robots. Businesses saw robots as a way to increase productivity, while workers worried they would take their jobs. Children viewed them as wondrous toys, while people with disabilities embraced them as tools of liberation. Military experts hoped robots would fight wars without endangering human soldiers, while politicians pondered if robots might eventually get to vote. Philosophers thought robots could challenge our notions of intelligence (and stupidity), while the religious struggled with concerns about the human race in a robot-dominated future.
Shadow Walker’s simplified anatomy included only one bone in the lower leg and a single wide toe on each foot.Science Museum Group
Peter Mowforth, cofounder of the Turing Institute in Glasgow, noted these disparate visions for robots when he announced the 1st International Robot Olympics, to be held in 27 and 28 September 1990 and hosted by the Turing Institute and the University of Strathclyde. The Olympics would round up the world’s best robots and showcase them head-to-head.
Mowforth himself thought all of the competing visions of robots were overblown. Steeped in machine learning research and robotics development, he knew firsthand the limitations of the state of the art: Robots rarely worked as intended, easily broke down, and glitched over seemingly trivial problems. He envisioned the Robot Olympics as a testbed to assess what the latest generation of robots could and could not do.
At the 1990 Robot Olympics, held in Glasgow, Shadow Walker wore pants to conceal its pneumatic “air-muscles” from competitors.Adam Hart-Davis/Science Source
The call for participation was wide open. Instead of having predetermined categories of competition, the organizers opted to see who applied to compete and then group them based on their claimed capabilities. In addition to picking the winners of individual events, the judges would select an overall Olympic champion based on the quality of the hardware, the sophistication of behavior, and novelty. Other prizes were given for young competitors, technologies that showed commercial potential, and design. In the end, more than 50 robots were entered, from a mix of universities, industry, and hobbyist groups from Canada, France, India, Japan, Mexico, the Soviet Union, the United States, the United Kingdom, and Yugoslavia.
There were plenty of disappointments. Trolleyman, a golf-cart-like wheeled robot, suffered a power failure while carrying the opening Olympic torch through the streets of Glasgow. The pile rug in the arena tripped up many robots that had been trained only on flat, smooth floors. David Buckley later concluded that the events were too difficult, and that the Olympics didn’t push development forward.
Of course, there were winners. In a surprise triumph for vintage technology, the fully mechanical 19th-century Japanese Archer from the Museum of Automata in York, England, won gold in javelin, beating out competitors more than 100 years its junior. The overall Olympic Champion was Yamabico, Shoji Suzuki’s entry from the University of Tsukuba, in Japan, which won bronze in obstacle avoidance and gold in wall following, but was disqualified in the talking category for not speaking English.
The Shadow Group had high hopes for Shadow Walker. Unfortunately, though, it failed to take a step, and the biped race was won by the Cardiff University Biped. Shadow Walker now resides in the collections of the Science Museum in London.
The Legacy of Shadow Walker
In 1997, a paying customer in search of a robotic leg compelled the Shadow Group to get serious and become a registered company. Shadow Robot is now Britain’s oldest robotics company. Rich Walker, who had left the Shadow Group to earn a B.A. in mathematics and a diploma in computer science at the University of Cambridge, joined Shadow Robot in 1999 as technical director. Today he’s the director of the company.
Shadow Robot specializes in durable robot hands rather than walking robots. But the focus on hands is also a legacy of the Shadow Group. Walker remembers that the Shadow Group’s first humanoid hand in the late 1990s was impressive simply for being able to pick up a pint of beer (a smooth-sided, thin-walled glass). Today, Shadow Robot’s hands are testbeds for dexterity. Gone are the pneumatic muscles, replaced by actuators that move each finger with precision. The classic model contains 20 motors, allowing for abductive and adductive movement with 24 degrees of freedom.
Shadow Walker’s operator wore a data suit that captured his movements and allowed the robot to copy them.Richard Greenhill
In a recent blog post, Sejal Parsotomo, senior marketing executive at Shadow Robot, wrote that while humanoid robots are great for public relations, specialized dexterity is key for success: A robot that can walk into your factory may be impressive, but a robot that can reliably manipulate objects is transformative.
In its struggles to take more than a few steps, the Shadow Walker showed the inherent difficulty that robots had in mastering even low-level skills. In August 2025, Beijing hosted the World Humanoid Robot Games. Competing in sports such as gymnastics, soccer, and track events, as well as more “useful” tasks like hotel cleaning and sorting medicine, these robots could literally have run circles around the competitors in the first Robot Olympics 35 years earlier. And yet, there is still so much work needed in order for robots to navigate the human-built environment. Despite the astonishing progress, we’re still not all that close to actually useful humanoid robots.
Part of a continuing serieslooking at historical artifacts that embrace the boundless potential of technology.
An abridged version of this article appears in the June 2026 print issue as “Learning to Walk.”
References
Richard Greenhill gives an overview of his life and the founding of the Shadow Group in a post on Shadow Robot’s corporate website.
David Buckley has a compilation of resources on the Shadow Biped Walker, including specifications from the 1999 iteration and a brochure from the 1st International Robot Olympics.
There is coverage of the Robot Olympics worthy of a gossip sheet in La Repubblicaand lovely footage of the competition in this TV-am interview of Peter Mowforth by Lorraine Kelly.
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
Robots are rapidly becoming part of our everyday lives, from drones and industrial machines to home assistants and humanoid robots. As their presence continues to grow, an important question arises: How can we choose the right robot—not only in terms of performance and cost but also in terms of sustainability? This video introduces the Eco‑Score for Robots, a new approach to evaluating the environmental impact of robotic systems. Just as eco-labels help consumers make informed choices in other industries, the Robotics Eco‑Label provides a clear and transparent way to assess how sustainable a robot truly is.
Robotic manipulation has come a long way since the 1990s. We’ve gone from the two-ball paddle juggling robot to AthenaZero, who can juggle barehanded using onboard vision feedback. By moving away from task-specific passive end-effectors such as cups or paddles and using multifingered hands, it can transition between a wide range of patterns including cascade, half-shower, tennis, shower, and box.
We show here the ARISTO (Anthropomorphic, Robotic, Integrated-Sensing, Tendon-Operated) Hand. Developed in collaboration with Sony Group Corporation, this research platform is engineered to address the complex requirements of manipulating small, thin, and fragile objects.
Moby shows what useful mobile manipulation looks like in the real world: picking up, carrying, and placing adaptable payloads. The video shows payload handling across increasing crate loads, including a 50.3-pound load, while maintaining balance, control, and mobility. This is the kind of capability that matters outside the lab—moving real objects, in real spaces, with practical reliability.
What does it take to make a robot look human? Harvard SEAS students Hailey Block, Henry Tavistock, and Evan Crowley created “Hollow Minds,” a pair of animatronic heads capable of speaking, blinking, tracking movement, and displaying lifelike facial expressions.
The longevity here is impressive, but the obvious question here is why the heck you’d ever do this task with a bipedal humanoid robot. It also doesn’t seem to have any error recovery, which is obviously fixable, but highlights the fact that real humans are versatile and humanoid robots are not.
Kacper Nowicki, CEO and cofounder of Nomagic, recently sat down for a deep dive into the “humanoid vs. purpose-built” debate during a panel discussion at the Web Summit in Vancouver 2026.
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
Just months after its debut, Atlas is proving why it is the world’s most capable and dynamic humanoid robot, ready for real work. Lifting a mini-fridge is a feat of strength, but the true breakthrough is in the underlying reinforcement learning and controls systems. The robot is learning to navigate real world adaptability: handling heavy objects by bracing and accounting for the mass and inertia; using whole-body control, not just hands to maneuver; and demonstrating superhuman range of motion and balance. This marks a critical shift in robotics where humanoids move beyond the lab and into dynamic industrial settings.
Watching Atlas move a fridge may be less impressive than whatever the heck it does at 4:10.
SpikerBot is a robot you teach by wiring neurons, not writing code. Drag spiking neurons in the app, connect them to sensors and motors, then press play. It moves, reacts, and changes behavior based on the brain you built.
Already funded on Kickstarter with a robot kit starting at US $219.
Wheeled-legged robots, which have wheels at their feet and achieve high mobility by coordinating wheel drive and leg drive, have been developed. In this paper, we address the problem of how to draw out the potential task-execution capability of the legs by freeing them from the roles of locomotion through external body support.
Using external voice commands, G1 is directly controlled to generate a wide range of actions in real time. This video was recorded in a single take, with on‑site audio recording.
Hummingbirds are impressive flyers, and advancements in high-speed photography, instrumentation, and measurement techniques have revealed much about their aerodynamics, flight behaviors, and wing and body kinematics. However, comparatively less is known about their natural flight dynamics, which is the relationship among a bird’s flight velocities, the control actions of its wings, and the acceleration of the bird in flight. To investigate this, at the Advanced Vertical Flight Laboratory we have designed, built, and flight tested a biomimetic robotic hummingbird on which is implemented the same techniques for flight control as observed in hummingbirds.
The original automated layout robot—the one that showed up when the construction industry was pretty sure robots were lame and then proved otherwise. It has printed millions of square feet of layout across thousands of projects. It built an entire category of construction technology. The category of: Stuff That Actually Does Helpful Work on Real Jobsites. But FieldPrinter 2 is here. It’s faster, tougher, smaller, and smarter. So for FieldPrinter 1, it’s time. Time for a quiet retirement. A mug. Maybe a plaque... But nay, good knight! Thou shalt expire in a blaze of thunderous glory!!
Noble Machines builds general-purpose robots for heavy industry, supporting people with the most hazardous and physically demanding tasks. Attendees at NVIDIA GTC 2026 witnessed the power of autonomous industrial work with Noble Machines Moby.
In March 1982, General Motors announced a rapid and aggressive conversion to robotics. By 1990, GM wanted 14,000 robots in their factories doing everything from painting to welding to assembly. Nowadays, we dream of robots in the factories, doing everything end to end. In the dark. Lights out. Guess what? GM dreamed the same 40 years ago, and they spent an estimated US $60 billion to try to make it reality. In today’s video, we look at General Motors and their dreams of the automated, all-robot factory.
When a group of academics started making open-source robotics hardware, a generation of roboticists got years of their lives back. Now, the bigger challenge is getting robots to think—and that’s starting to be open sourced too.
The shift is still early, but companies including Hugging Face, Nvidia, and Alibaba have all made significant bets on open-source robotics in the last two years, releasing tools and models aimed at the higher-level work of getting robots to reason, decide, and act.
The open source movement that accelerated other AI applications is now being applied to the problem of making robots smarter. If these attempts to bring AI to robotics with open-source platforms succeed, the barrier to building a capable robot could fall as fast as the barrier to building an AI application did.
The world ROS built
Open-source robotics software has been around since the mid-1990s, with early projects like Carnegie Mellon University’s Inter-Process Communication package and the Player Project in the early 2000s laying the groundwork. But these were often tied to specific research groups, and the field remained fragmented.
The Robot Operating System, ROS, changed that when it made its debut in 2007. By bundling tools and attracting more users, it became the de facto standard. The story of open-source robotics, in many ways, starts there.
Despite its name, ROS is not actually an operating system. Rather, it is a software framework that sits on top of Linux and handles robotic fundamentals like moving data between components, talking to hardware, building maps, planning paths, and supporting developer tools, such as data logging and visualization. Before ROS, every robotics team wrote that infrastructure themselves. It often took a year or two before a lab could get to the research it actually cared about.
Brian Gerkey, who helped build ROS in the mid-2000s, says he was drawn to the project because of how much open source had already changed the world, pointing out that nearly the entire internet is built on it.
“I’m a tool builder, and I like to share everything as openly as I possibly can, because I think that’s where we get the most impact out of what we build,” says Gerkey, board chair of Open Robotics and now CTO at Intrinsic, a robotics and AI unit of Google.
As it was developing, the AI community largely took the same approach, sharing research, models, and data openly, and the field accelerated faster than almost anyone predicted. Now some of those same advancements are arriving in robotics.
Open-source AI for robotics
Computer vision, once a hard problem, has advanced dramatically in just a few years, says Spencer Huang, Nvidia’s director of product for robotics. What once required significant expertise can now be done in a few lines of code. Simulation tools have become accurate enough to be useful for training, and access to the tooling that once required a specialized lab is now widely available, much of it open source.
“To get into robotics, you no longer need a Ph.D.,” he says. The result is a much larger pool of people who can contribute, and the field is starting to look less like a specialized discipline and more like a platform that anyone can build on.
Nvidia has built out an open-source robotics stack that covers the full development pipeline. Its Cosmos world models generate synthetic training data and simulate physical environments. Its GR00T models give robots the ability to reason through and execute complex tasks. And its Isaac frameworks handle the orchestration that ties training, simulation, and deployment together. Not everyone needs to train the robots from scratch, Huang says, and most people probably shouldn’t.
“If you gate pre-training, the field just never grows,” he says. “We should be able to provide a high-quality, state-of-the-art pre-trained model that anyone can go and take and fine tune for their own purposes.”
All of Nvidia’s open-source models live on Hugging Face, the open-source AI platform that has become the default place to share models and datasets. Hugging Face launched LeRobot, a community platform for robotics AI, in May 2024. Since its launch, the number of robotics datasets on the platform grew from 1,145 at the end of 2024 to more than 58,000 today, making it the single largest dataset category on the hub.
Hugging Face has also moved into hardware, acquiring robotics company Pollen Robotics. The acquisition came from a realization that software alone was not enough, according to Clement Delangue, Hugging Face’s CEO. The goal, as with the software, was to bring more people in.
The contributors to LeRobot include the biggest names in the industry, academic labs, and hobbyists building robots in their spare time. For instance, earlier this year, Alibaba released RynnBrain, an open-source foundation model for physical AI that the company claims outperforms comparable offerings from Google and Nvidia on benchmarks. That diversity of projects, Delangue says, is important.
“It is not just one model or one dataset or one hardware,” he says. “It is a lot of small contributions that everyone can be part of.”
Commercial incentives muddle the field
The stakes, Delangue says, go beyond convenience. A world where only a few proprietary systems control the robots in people’s homes is a concerning one. “Having robots at home that you don’t really understand, that you don’t really control, that a few people in Silicon Valley control is a scary thought,” he says. “Open source gives an alternative path.”
But getting there is not straightforward. The open sourcing happening now looks different from what produced ROS, which emerged largely from academics pooling their work with no commercial stake in the outcome. The biggest contributors today are companies with clear business reasons to want more people building on their platforms. That’s not necessarily a bad thing, says Bill Smart, a professor at Oregon State University, in Corvallis, who was part of the early open-source robotics community. But the incentives are worth being aware of.
He also worries that the lowered barrier to entry has a downside. Researchers coming from AI without a robotics background are sometimes solving problems the field already solved. A newcomer might spend a week training a neural network to move a robot’s hand from one point to another, unaware that the same task can be accomplished with a few lines of code using decades-old techniques. The incentives are not always pointing in the same direction as the progress.
Smart is not without hope though. Whatever the motives behind the open sourcing, he says, the effect is real. More people are in the field than ever before, the tools are genuinely easier to use, and the community is bigger and more diverse than anything that existed when ROS was getting started.
“Anyone can make a robot move now,” he says. “As an old tech guy, that makes me happy and sad, because I’m no longer special.”
This sponsored article is brought to you by Wetour Robotics.
A field technician on a wind turbine, harness clipped, both hands on a wrench, needs to send a command to the diagnostic device hanging at her belt. A logistics worker on a loading dock, gloves on, eyes on the pallet, needs to redirect a connected lift. A person using an assistive mobility device on a crowded street wants to nudge it forward without taking out a phone or speaking aloud. None of these moments call for a smarter robot. They call for a smarter way to be heard by the machines that already exist.
The industry has been building from one side
The past three years of Physical AI have been a story of remarkable progress on the robot side of the loop. Companies like Boston Dynamics, Figure, and Unitree have advanced actuators, locomotion, and dexterity to a level that would have seemed implausible a decade ago. Google DeepMind’s Gemini Robotics has redefined what vision-language-action models can do in unstructured settings. The trajectory of the hardware and the foundation models is real, and it is accelerating.
But there is another side to this loop, and it has been treated as a solved problem for too long. The interface between humans and machines has defaulted, for 40 years, to three input modalities: screens, buttons, and voice. Each of those assumes the user can stop, look down, and translate intent into structured commands. That assumption breaks the moment the work moves into a real environment. On a turbine. On a dock. On a sidewalk. In any setting where hands are occupied, eyes are committed, or speaking is impractical, the conventional interface stack quietly fails.
Spatial Intent Fusion is the simultaneous processing of three streams of human-centered information, namely spatial position, visual context, and gestural intent: Your body is the interface.
The bottleneck on the human side of the loop is becoming as important as the one on the machine side. And solving it requires a different question. Not how do we make the robot more capable, but how do we let the human participate in the computing system as naturally as the robot already does.
Wetour Robotics’ bet: put the human back into the computing loop
Wetour Robotics is betting that the next architectural leap in Physical AI is not about making the robot more capable. It is about making the human a first-class node in the computing network, with the same kind of low-latency, high-fidelity participation that connected devices already enjoy.
Wetour Robotics’ engineers frame the problem this way: a wristband that recognizes a gesture is not enough. A camera that recognizes a scene is not enough. The information a human carries about what they are about to do is distributed across multiple channels, including where their body is in space, what their eyes are attending to, and what their muscles are preparing to do, and any single channel observed in isolation is ambiguous. Reconstructing intent reliably means fusing those channels at the operating system level, with latency low enough that the loop feels closed rather than mediated.
This approach has a name. Wetour Robotics calls it Spatial Intent Fusion: the simultaneous processing of three streams of human-centered information, namely spatial position, visual context, and gestural intent, fused into a single real-time command for any connected physical device. It is the technical implementation behind a simpler positioning statement the company uses externally: your body is the interface.
Orchestra is a portable intelligent hub running the operating system that handles sensor fusion, intent inference, command translation, and safety arbitration. The reference compute platform is NVIDIA Jetson Orin Nano Super, which provides enough on-device inference capacity to keep the entire control loop at the edge, with no cloud dependency on the critical path. Wetour Robotics
The architecture: three layers, four engines, one loop
Orchestra is not a single device but a layered platform, designed from the start to be sensor-flexible and actuator-agnostic. The architecture decomposes into three perception layers and four coordination engines.
Orchestra itself is the local compute and orchestration core: a portable intelligent hub running the operating system that handles sensor fusion, intent inference, command translation, and safety arbitration. The reference compute platform is NVIDIA Jetson Orin Nano Super, which provides enough on-device inference capacity to keep the entire control loop at the edge, with no cloud dependency on the critical path. Edge inference is non-negotiable for this application. Full-chain latency from biosignal acquisition to actuator command is held under 100 milliseconds, the envelope inside which closed-loop control feels natural rather than laggy.
VisionLink handles visual and spatial perception. Cameras feed into vision models that identify objects, estimate distances, and track environmental context. VisionLink is designed not as a passive recognition layer but as a real-time command generator: its outputs feed directly into Orchestra OS to be fused with biosignal data.
Conductor is the biosignal pipeline. It ingests raw surface electromyographic (sEMG) data from a wrist-worn device, classifies temporal patterns into discrete gestures or continuous control signals, and outputs actuator commands. The technically interesting property of sEMG for this use case is that the signal precedes visible motion. Motor unit action potentials appear at the skin surface roughly 50 to 80 milliseconds before a finger completes the corresponding gesture. Wetour Robotics calls this property pre-motion intent sensing, and it is what allows Orchestra to anticipate user intent rather than react to it.
On top of the three perception layers, Orchestra OS runs four coordination engines. The Perception Engine ingests and normalizes raw sensor streams. The Intent Engine performs Spatial Intent Fusion across modalities, resolving what the user is trying to do given where they are, what they are looking at, and what their hand is signaling. The Orchestration Engine translates intent into device-specific command sequences for any connected actuator. The Safety Engine arbitrates conflicting commands, enforces operational envelopes, and gates execution against runtime safety conditions.
The trade-offs we’re honest about
No system that bridges the human body and the digital world is finished. Three engineering challenges remain open, and the company addresses each with a deliberate trade-off rather than a claim of having fully solved it.
Baseline stability of sEMG under motion. In a stationary user, continuous gesture recognition from sEMG is reliable. Once the user is walking, climbing, or otherwise moving, motion artifacts and electrode drift degrade the signal in ways that are difficult to fully compensate for. Rather than overpromise on continuous control in dynamic settings, Orchestra defaults to a smaller set of robust discrete gestures in complex operating environments, and reserves continuous control modes for contexts where the signal-to-noise ratio supports them.
Miniaturization of edge AI compute. Running the Orchestra control loop entirely at the edge requires real on-device inference, which has historically meant trading off between compute capacity, battery life, and form factor. Wetour Robotics’ approach has been a compact carrier board paired with a thermal design and a battery module sized for all-day wearability. The result is a hub that travels with the user rather than tethering them to a desk, and that performs the full perception-to-actuation loop without offloading to the cloud.
Heterogeneity of third-party device protocols. The actuator side of the loop is a fragmented landscape. Different manufacturers expose different command interfaces, different communication stacks, and different safety conventions, and a Physical AI operating system has to integrate with all of them. Wetour Robotics uses an AI-agent layer to negotiate connection and protocol translation adaptively, so that Orchestra OS can ingest data from a wide range of devices, run them through neural network models that infer human intent, and emit the right command on the right protocol for the device on the other end.
Why this matters, and why it helps the rest of the field
The history of computing is a history of interface revolutions. Command lines gave way to graphical user interfaces, which gave way to touch, which gave way to voice. Each transition expanded who could participate in the system and what they could do with it. The next transition is not about a new screen or a new microphone. It is about treating the human body itself as a participant in the computing network, capable of contributing intent at the same speed and fidelity that any other connected node can.
The history of computing is a history of interface revolutions. The next transition is not about a new screen or a new microphone — it is about treating the human body itself as a participant in the computing network.
This path is not a competitor to the work being done on humanoid robots, foundation models for embodied AI, and dexterous manipulation. It is the missing complement to that work. The hardest open problem for humanoid systems is the data: every natural interaction between a human and the physical world is a potential training signal, and most of those interactions are currently invisible to any computing system. As more humans become first-class nodes in the loop, those interactions become observable, structured, and ultimately useful for training the next generation of embodied AI, including the humanoid robots being developed today.
In other words: putting the human back into the computing loop is not just about better interfaces for individual users. It is about generating the kind of grounded, in-the-wild human-machine interaction data that the broader Physical AI ecosystem will need to keep advancing. The robot side and the human side of the loop are not two competing futures. They are two halves of the same one.
That is what Wetour Robotics means when it says: Your body is the interface.
Over the next few decades, billions of autonomous, AI-powered robots will work alongside people in factories, perform tedious tasks in warehouses, care for the elderly, assist in unsafe disaster areas, deliver packages and food to our doorsteps, and eventually help out in our homes. Some will look like us, and many won’t. What is certain is that regardless of form factor, robots will all rely heavily on AI in order to deliver real-world value.
In 2025, total investments in robotics companies reached a record US $40.7 billion, accounting for 9 percent of all venture funding. The multibillion dollar question therefore is this: What will it take for AI-powered robots to begin to have a serious economic impact? Many of today’s robotics and AI companies are making bold claims, such as that humanoid robots will soon be coming into our homes, but there’s still a big gap between promise and reality.
The promise of robots that live and work alongside us has been the stuff of science fiction for a very long time. And while many programmers have tried to make that promise a reality, the physical world is just too complicated for traditional computer programs to handle the endless complexity it presents. Thanks to AI, robots are no longer being programmed—instead, they learn to operate in the real world. With enough practice, they can learn to perceive and understand the world around them, reason about that world, and use that reason and understanding to perform tasks that are useful, reliable, and safe.
The two of us have worked at the forefront of AI and robotics for the last decade, as a Professor in Robotics at Oregon State University and Co-Founder of Agility Robotics, and as former CEO of the Everyday Robots moonshot at Google X. Our experience deploying AI-powered robots in real-world settings has given us a perspective on where AI can be used to great benefit in complex robotic systems in the near term and where we are still on the frontier of science fiction. We believe AI will enable an inflection point in robotics advances, but that it will be through the well-engineered application of coordinated systems of different AI tools rather than a single ChatGPT-style breakthrough.
As the excitement around AI is matched only by the uncertainty of what will be possible, here are five hard truths that will define AI in robotics.
1. The YouTube-to-Reality Gap Is Real
For years, we have been seeing videos on YouTube with humanoid robots performing amazing moves on everything from a dance floor to an obstacle course. The inside knowledge in robotics is to “never trust a YouTube robot video.” The gap between real robots that can perform real work in unstructured human environments and carefully scripted and edited robot performances remains significant. The latest performance to get a lot of attention was a martial arts show featuring Unitree humanoid robots performing with children at the Chinese 2026 Spring Festival Gala. While impressive, this falls into a long lineage of tightly scripted robotic performances, where everything has been carefully choreographed and planned in advance. The low-level controls, synchronization, and choreography were stunning, yet the Spring Gala robot performance showed a level of autonomy and intelligence much closer to industrial robots building cars in a factory than something that will show up in your living room any time soon.
Seeing these kinds of demos nevertheless raises questions about where robotics really is. If robots can perform kung fu moves and do backflips and dance, why aren’t they also showing up on factory floors yet? And why can’t they do the dishes in my home after dinner? The simple answer is this: Making AI-powered robots capable of performing general tasks in varied human environments is still really hard. While impressive technological feats like those at the Spring Festival may make it look like we could be very close, the use of AI in these demos is only for low-level motor control (to keep the robots from falling over) and therefore is only a small part of the solution for robots to be general purpose in the real, unstructured spaces where we humans live and work.
2. Data Is An Unsolved Challenge
Large Language Models (LLMs) like OpenAI’s ChatGPT and Anthropic’s Claude were initially trained on an internet-scale database of text. The world woke up one day in late 2022 to ChatGPT demonstrating that AI computers could suddenly “speak” to us in prose or verse and about seemingly any topic. LLMs have turned out to generalize well and are now able to take multimodal input (text, images, video) and produce multimodal output. Importantly, the corpus of training data was both enormous and human-generated, which are characteristics that form the gold standard for AI training.
The fastest path to robots as part of everyday life may emerge through a range of robot forms performing increasingly sophisticated applications and employing a range of AI tools.Agility Robotics
Giving AI a body (in the form of a robot), so that it can engage with people in the physical world, continues to be a very difficult and broadly unsolved problem. AI models for general-purpose robotics must simultaneously satisfy multiple, often conflicting, physical, geometric, and temporal limitations while operating in unstructured, dynamic environments. In order to generalize, robot models need to be trained on data gathered in a high-dimensional configuration space, where “dimensions” represent text, lighting conditions, degrees of freedom, joint limits, velocities, force, and safety boundaries, just to mention a few. Importantly, this must be good data—it must contain many examples from what amounts to an infinite number of possible configurations in the physical world.
Since there are very few existing sources of data like this, approaches like teleoperation, video analysis, motion capture of humans, and self-exploration in simulation and in the real world are all seen as important ways to collect data. It’s a herculean task. For example, at Everyday Robots at Google X, we ran 240 million robot instances in our simulator over the course of 2022 to collect training data, mostly to train a trash-sorting model. Similar amounts of data will be needed for every skill to get to a similar level of capability, which is not yet human level.
3. There Will Be No Single Robot AI
We are far away from a moment where a single AI model might allow general-purpose robots to live and work alongside us.
General-purpose robots can have wheels or legs. They can have one, two, three, or more arms. Some have propellers and can fly, while others may be designed to operate under water. Some will drive on busy roads. The physical world is infinitely varied and complex. And then there are all the people and other animals that will be surrounding the robots. How do you train a model to operate a robot safely and reliably in all of these settings? The simple answer is: You don’t. At least not for quite some time.
We believe the winning AI architecture leading to the next big breakthroughs in general-purpose robotics will be “agentic AI” for robots, which are high-level coordinating models that can reason, plan, use tools, and learn from outcomes to execute complex tasks with limited supervision. Agentic, high-level models running on robots will invoke a system of specialized ones for different types of tasks. We will likely soon see multiple robots collaborating and coordinating with each other through their onboard agentic AI models.
AI tools are unlocking new and powerful capabilities in robotics, which in turn will enable new solutions and new markets. It’s encouraging to see these new models being made broadly available, some even as open-source solutions. This availability is akin to what happened with the internet: Real progress occurred when it became ubiquitous. We anticipate an inevitable democratization of complex behaviors in robotics with wide access to these AI tools and technologies.
4. Hardware Is Still Very Hard
Robots are complex systems with many parts that all need to work together with great precision. For a robot to be useful and safe, every part of it must be coordinated, from its perception systems to the computer controlling it, all the way down to its individual actuators.
Actuators—that is, the motors and gears—are a good example of an important part of the robot where what got us here won’t get us there. The actuators used at scale by most industrial robots will not work for robots that will operate in human environments. If these robots accidentally collide with an obstacle, the resulting impacts are harsh, forces are high, and things break. Humans don’t move in this way. We are far more compliant in how we interact with the world, and we’re constantly making contact with our environment and using that contact to help us accomplish things.
Consider the challenge of inserting a key in a lock: Humans typically don’t do this by aligning the key perfectly with the keyhole. Instead, we just feel for the edge of the keyhole and jiggle the key in. Robots need to be able to operate in novel ways to achieve comparable capabilities by using a new class of actuators that are sensitive to force and able to have a compliant interaction with the environment. While these kinds of actuators do exist, they are not yet generally available at scale for robot systems designed to operate around people.
5. Real Value Comes From “Easy” Tasks
There’s a big difference between tasks that look impressive and real-world tasks that provide value. Robotics is a perfect example of Moravec’s paradox, which states that tasks that are hard for humans are easy for computers (like multiplying two big numbers), and tasks easy for humans (like a toddler’s movements) are extremely difficult for computers and robots.
Serving customers is an unforgiving reality check, because customers only care about solving the real problems they have. If we are to deploy AI-based robot solutions, they must outperform the way things are currently done while demonstrating reliable performance metrics and safety. Agility Robotics’ early work to deploy our humanoid robot Digit in customer locations led to the realization that our first obstacle was safety: Robots that balance and manipulate objects in human spaces bring new types of risk to the workplace. In the first humanoid deployments, physical barriers were necessary, and Agility kicked off a multi-year engineering effort to solve the safety challenge, touching nearly every aspect of robot design and relying heavily on new AI-based approaches to human detection and behavior control.
Everyday Robots at Google deployed robots in 2019 that worked autonomously in office buildings doing chores like cleaning cafe tables and sorting trash. We quickly learned how “messy” and difficult the real world is for a robot. This experience informed the architecture and deployment of our AI systems while also gathering real-world data that could be combined with simulation data for training and improving models.
This focus on creating a product to meet specific customer needs and deploying robots in real-world settings is the only way to inform the structure of the AI tools and infrastructure for near-term utility on a path towards long-term broader capability and generality. There will be no “aha” moment, no silver bullet algorithm, and no volume of data sufficient to produce a general-purpose robot without extensive real-world experience.
AI Robots Are Coming, One Step at a Time
As we look to the future, there is no doubt that the world is bringing AI into the physical world through robots. We are at the beginning of a “Cambrian explosion“ of useful, intelligent machines. We believe AI is not one tool, but a huge frontier of technical approaches that is unlocking new capabilities so powerful, they will define our economy moving forward. This will happen not in one single definitive moment, but as an ongoing set of small and large breakthroughs, where AI-driven robots begin to provide real value in a few tasks, and then a few more, with impacts unfolding across numerous $100 billion-plus markets that will dramatically improve the quality of our lives.
Electronic waste is moving up on regulatory agendas in 2026. New European waste-shipment rules, expanded recycling fees on products with non-removable batteries in California, and an e-waste import ban in Malaysia, for example, are all increasing pressure to recover more value before electronics are shredded or exported.
The world is projected to generate 82 million tonnes of e-waste annually by 2030, according to the United Nations’ most recent Global E-Waste Monitor report in 2024. The report estimated that current e-waste management captures less than a third of the recoverable metal value contained in discarded electronics.
For recyclers, much of that lost value is a consequence of what happens before a circuit board ever reaches a smelter or shredder. Boards contain a mixture of components such as memory chips, processors, magnets, and capacitors, as well as valuable raw materials such as copper, aluminum, tantalum, and precious metals. Conventional recycling often mixes everything into bulk streams and destroys components that might otherwise be reused.
Tuurny, a startup based in San Francisco, is developing an automated system to remove and separate reusable chips from circuit boards before the remaining material is shredded. In April, the company announced it had designed a robotic system, called Nantul, to identify and extract RAM integrated circuits, claiming each machine can recover 300 intact RAM ICs per hour.
Sina Ghashghaei, Tuurny’s founder, says the company is preparing its first field deployment with dozens of machines through a six-figure deal with Areera, a television recycler in the United Kingdom, which processes 1,500 tonnes of televisions per month. The deployment is planned for early 2027.
Tuurny’s first target is recovering RAM ICs and other chips used in legacy systems where replacement components can be difficult to source. Ghashghaei says the company is talking with a few legacy chip suppliers and pursuing potential agreements to supply aluminum and copper recovered from circuit boards to smelters and refiners. He declined to identify the companies involved.
Robots for Automated RAM Recovery
Traditional electronics recycling often begins by shredding boards and sorting the mixed output afterward. Tuurny aims to do the opposite: Identify and remove components first, sort them by model or material, then reroute the recovered items to testing labs for potential reuse as new chips or to refiners and smelters for further processing.
Nantul comprises three robotic systems in one. The first is an arm to continuously feed the component-removal robots, paired with two tabletop machines similar to 3D printers or computer numerical control (CNC) machines. A neural network identifies and catalogs components, then searches the internet for manufacturers’ thermal-profile specifications. Nantul uses those specifications to employ a combination of suction, controlled heat, computer vision, and robotic controls to remove chips while minimizing damage. Recovered items are then sorted by model number in material-specific groups.
“We’re creating a new supply chain from old feedstock that didn’t exist before,” Ghashghaei says, adding that manual recovery is expensive and difficult to scale.
Tuurny’s recovery system includes a computer vision system that identifies specific RAM components to assess them for recovery.Tuurny
Minghui Zheng, an associate professor of mechanical engineering at Texas A&M University, in College Station, who studies robotic disassembly and electronics recycling systems, says Tuurny’s approach appears technically feasible, especially when focused on the narrow, valuable target of recovering RAM from more controlled e-waste streams.
“RAM is a good starting point because it has relatively high reuse value and is more standardized than many other electronic parts,” Zheng says. The harder challenge, however, is removing chips “without heat, mechanical, or electrical damage, and making sure it still works reliably afterward.”
Used circuit boards can vary by layout, markings, age, contamination, solder condition, or prior damage. A robot has to identify the correct component, choose a removal strategy, apply heat locally, lift the part cleanly, and preserve enough information about the part for downstream testing and resale.
E-Waste Recycling Strategies
Ghashghaei says Tuurny is building small modular machines using off-the-shelf parts, custom controls, and Nvidia Jetson Nano hardware. The company is trying to keep costs down by reducing hardware complexity to arrive at a price point far below centralized industrial equipment used at large facilities. He says the biggest challenge from an engineering perspective has been developing the autonomous computer vision and robotic control.
Last year, the four-person startup received a NASA-funded grant to support an AI-powered repair assistant for printed circuit boards that used computer vision and a custom large language model (LLM) to guide technicians.
Ghashghaei says Tuurny pivoted from board repair to e-waste processing after concluding that discarded electronics represented a larger market amid growing interest in the U.S. around on-shoring capacity for critical minerals and rare earths. The pivot also positions Tuurny to potentially address supply chain concerns around legacy chips for systems in telecom, aerospace, defense, and other industries where equipment remains in service long after chips leave mainstream production.
In practice, Zheng says the main challenge in making robotic disassembly of electronics commercially viable is ensuring it’s adaptable enough to handle the large variability in e-waste while keeping costs reasonable.
“Every electronic product is different, and used boards may be damaged, dirty, or arranged differently. The robot must be able to find the right parts, remove them carefully, and avoid damaging them in real time, which creates major challenges for robotic perception, decision-making, planning, and manipulation,” Zheng says. “Economically, the recovered parts should be valuable enough to justify the costs of the robot, sensing, testing, maintenance, labor, and scaling up the process.”
For smelters and refiners, the question may be whether Tuurny can supply predictable material streams at commercial volumes. Ghashghaei acknowledged that Tuurny’s scaling efforts could run into its own supply chain constraints in trying to acquire enough components to build more robots.
Zheng called Tuurny’s approach promising but still early. “For now, it is more realistic as a targeted recovery strategy for valuable components like RAM,” Zheng says. “The key question is whether the robotic disassembly technology can work reliably, affordably, and at scale.”
The International Organization for Standardization (ISO) is updating its 12-year-old safety requirements for personal care robots. A lot has happened since the last revision, both on the technology side and with researchers’ understanding of safety for humans collaborating with domestic robots. The proposed ISO update addresses hazard identification, risk assessment, and different use scenarios. It does not, however, set limits, propose testing methods, or have enforcement mechanisms that might address the complexities of human-robot collaboration. And that is a problem, argues technology policy researcher Jae-Seong Lee of the Electronics and Telecommunications Research Institute in Daejeon, South Korea.
Why is the next revision of ISO 13482 a big deal? Jae-Seong Lee: The standard is moving into final approval at a moment when domestic humanoid robot makers are shifting from lab prototypes to products aimed at real homes, real caregivers, and real families. That matters because the standard does more than specify geometry and impact limits. It helps define what counts as acceptable robot behavior in the messy world of everyday life.
What is the core engineering problem? Lee: It is not simply whether a robot can avoid collisions or detect people in its path. The harder problem is that human-robot interaction is bidirectional. The robot changes what the human does, and the human changes what the robot perceives and does next. In other words, safety is not a fixed property of the machine alone; it emerges from the relationship.
Isn’t that already covered by current safety standards? Lee: Only partially. ISO 13482 addresses personal care robots through hazard identification, risk assessment, and intended use scenarios, and related guidance acknowledges noncontact hazards such as unpredictability and incorrect autonomous decisions. But it stops short of binding compliance criteria, test methods, or enforcement mechanisms for the hazards produced by the human-robot relationship.
The technical community understands bidirectional coupling, and the standards framework acknowledges relevant hazards, but no current standard fully converts that knowledge into enforceable rules for domestic autonomy.—Jae-Seong Lee
Why can’t engineers just better define a robot’s operating envelope? Lee: Because the value proposition of a domestic humanoid depends on operating in uncontrolled environments. A robot that is safe only in standardized rooms, with healthy adults, under well-defined conditions is not really a domestic humanoid at all.
In industrial robotics, designers can usually bound the task, the workspace, and the population. In a home, the robot must adapt to elderly residents, children, visitors, pets, clutter, tight spaces, and fluctuating human behavior. Those aren’t edge cases. Those are the baseline. Tightening the domain to be more like that of factory robots would make the home robots less useful.
The proposal mentions training data. Why does that matter? Lee: Because the data already reflect the diversity of domestic life. Companies building humanoid training datasets are reportedly sending paying contract workers around the world to record their chores in ordinary settings. That means the robots will be trained on real-world variability, not sanitized demonstrations. The safety problem is therefore in the composition of the entire human-robot system, not in any one component.
What is the standards gap? Lee: The gap is governance. The technical community understands bidirectional coupling, and the standards framework acknowledges relevant hazards, but no current standard fully converts that knowledge into enforceable rules for domestic autonomy. What is missing is a way to specify safe behavior across the full range of human conditions the robot will actually encounter.
What’s also missing is a decision about who gets to decide whose behavior counts as normal. Whose gait sets the baseline? Whose is an acceptable risk threshold? Whose definition of safe judgment gets written into the requirement language? Those are value judgments, not purely engineering ones. A standards committee cannot avoid choosing a normative reference point; it can only decide whether that choice is explicit and inclusive.
Who could help answer those questions? Lee: The proposal argues that the people most affected by domestic humanoids are not systematically represented in the working groups shaping the standard. It points especially to older adults, who are often the primary intended users of domestic care robots, yet whose movement patterns and cognitive states are not directly embedded in the standards process.
In other words, this revision acknowledges the hardest problems but pushes unresolved issues into advisory language, nonbinding guidance, or future revision scopes. That can be useful, but it also delays the real question: What counts as safe relational behavior in the home?
What are the stakes? Lee: The risk is not only injury, though that is the obvious concern. The deeper risk is that safety assumptions get baked into products and standards before the market, regulators, and users have a chance to question them. Once deployment patterns harden, it becomes much harder to revise the baseline.
What should the engineers on the standards bodies do about it? Lee: The engineers on the standards body should ask not just, “What are the robot’s outputs, and do they stay within safe thresholds?” but “What states does this robot engage with, and does that engagement remain safe across the full range of those states?” That shift sounds subtle, but it changes the design brief. It moves safety from machine-centric measurement toward system-level relational assurance.
Domestic humanoid safety cannot be solved by machine engineering alone. It requires a framework that treats the human not as background noise, but as part of the system, part of the definition of the safety envelope.
For years, the field of robotics has used the terms “dull, dirty, and dangerous” (DDD) to describe the types of tasks or jobs where robots might be useful—by doing work that’s undesirable for people. A classic example of a DDD job is one of “repetitive physical labor on a steaming hot factory floor involving heavy machinery that threatens life and limb.”
But determining which human activities fit into these categories is not as straightforward as it seems. What exactly is a “dull” task, and who makes that assumption? Is “dirty” work just about needing to wash your hands afterwards, or is there also an aspect of social stigma? What data can we rely on to classify jobs as “dangerous?” Our recent work (which was not dull at all) tackles these questions and proposes a framework to help roboticists understand the job context for our technology.
First, we did an empirical analysis of robotics publications between 1980 and 2024 that mention DDD and found that only 2.7 percent define DDD and only 8.7 percent provide examples of tasks or jobs. The definitions vary, and many of the examples aren’t particularly specific (for example, “industrial manufacturing,” “home care”). Next, we reviewed the social science literature in anthropology, economics, political science, psychology, and sociology to develop better definitions for “dull,” “dirty,” and “dangerous” work. Again, while it might seem intuitive which tasks to put into these buckets, it turns out that there are some underlying social, economic, and cultural factors that matter.
Dangerous Work: Occupations or tasks that result in injury or risk of harm
It’s possible to measure the danger of a task or job by using reported information. There are administrative records and surveys that provide numbers on occupational injury rates and hazardous risk factors. While that seems straightforward, it’s important to understand how this data was collected, reported, and verified.
These caveats are an opportunity for robotics to be helpful. If we went out and looked for it, we could probably find some less obviously dangerous work where robotics might be an important intervention, not to mention some groups that are disproportionately affected and would benefit from more workplace safety.
Dirty Work: Occupations or tasks that are physically, socially, or morally tainted
Colloquially, most people might think of dirty work as involving physical dirtiness, such as trash removal, cleaning, or dealing with hazardous substances. But social science literature makes clear that dirty work is also about stigma. Socially tainted jobs are often servile or involve interacting with stigmatized groups (for example, correctional officers), and morally tainted jobs include tasks that people commonly perceive as sinful, deceptive, or otherwise defying norms of civility (like a stripper or a collection agent).
“Dirty work” is a social construct that can vary across time (like tattoo industry stigma in the United States) and culture (such as nursing in the U.S. versus in Bangladesh). One way to measure whether work is “dirty” is by using the closely related concept of occupational prestige, captured through quantitative surveys where people rank jobs. Another way to measure it is through qualitative data, like ethnographies and interviews. Similar to “dangerous,” we see some hidden opportunities for robotics in “dirty” work. But one of our more interesting takeaways from the data is that a lower-ranked job can be something that the workers themselves enjoy or find immense pride and meaning in. If we care about what tasks are truly undesirable, understanding this worker perspective is important.
Dull Work: Occupations or tasks that are repetitive and lacking in autonomy
When it comes to defining dull work, what matters most is workers’ own experiences. Outsiders can make a lot of false assumptions about what tasks have value and meaning. Sometimes things that seem boring or routine create the right conditions for developing skills and competence, such as the concentration needed for woodworking, or for socializing and support, when tasks are done alongside others. Instead of assuming that repetitive work is negative, it’s important to examine qualitative data on how people experience the work and what purpose it serves for them.
DDD: An actionable framework
In our paper, we propose a framework to help the robotics community explore how automation impacts individual jobs. For each term—dull, dirty, and dangerous—the framework gathers key pieces of information to reflect on what physical or social aspects of the task are, in fact, DDD. Worker perspective is an important part of all three considerations. The framework also emphasizes awareness of context—meaning the physical and social environment of an occupation and industry that can influence the DDD nature of a task. Our corresponding worksheet suggests existing data sources to draw on and encourages us to seek out multiple perspectives and consider potential sources of bias in the information.
What makes tasks dull, dirty, or dangerous depends on the perspective of the humans doing those tasks.RAI
The job is also repetitive, but there are aspects that make it not dull. Specifically, workers cite the day-to-day interaction with their coworkers (which includes extensive insider vocabulary, work hacks, and mutual aid groups) and task variety as two of the most enjoyable aspects of the job. Task variety includes inspecting their vehicle and equipment, driving their truck, coordinating with crew members, lifting bins and bags, detecting incorrect sorting of waste, and unloading at the end destination.
This finding matters because some types of robotic solutions will eliminate the parts of the job that workers most appreciate. For instance, the National Institute for Occupational Safety and Health (NIOSH) recommends the adoption of automated side loader trucks and collision avoidance systems. This innovation increases safety, which is great, but it also results in a sole worker operating a joystick in a cab, surrounded by sensor and camera surveillance.
Instead, we should challenge ourselves to think of solutions that make jobs safer without making them terrible in a different way. To do this, we need to understand all aspects of what makes a job dull, dirty, or dangerous (or not). Our framework aims to facilitate this understanding.
Finally, it’s important to note that DDD is only one of many possible approaches to classify what work might be better served by robots. There are lots of ways we could think about which types of tasks or jobs to automate (for example, economic impact or environmental sustainability). Given the popularity of DDD in robotics, we chose this common phrase as a starting point. We would love to see more work in this space, whether it’s data collection on DDD itself or the creation of other frameworks.
At RAI, we believe that the fusion of robotics and social sciences opens a whole new world of information, perspectives, opportunities, and value. It fosters a culture of curiosity and mutual learning, and allows us to create actionable tools for anyone in robotics who cares about societal impact.
This presentation highlights recent efforts at the Johns Hopkins Applied Physics Laboratory to advance agentic AI for collaborative robotic teams. It begins by framing the core challenges of enabling autonomy, coordination, and adaptability across heterogeneous systems, then introduces a scalable architecture designed to support agentic behaviors in multi-robot environments. The talk concludes with key challenges encountered and practical lessons learned from ongoing research and development.
Key learnings
Provides an introduction to LLM-based AI Agents
Describes an approach to applying LLM-based AI Agents to robotic teams
Provides demonstrations of the approach running in hardware with a heterogeneous team of robots
Presents lessons learned and future work in this area
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
Bulk material handling is a critical, labor-intensive operation across various industries, traditionally performed by human operators using heavy hydraulic manipulators equipped with free-swinging, underactuated grippers. This work presents the first complete autonomous material-handling solution deployed on a real-world 40-ton material handler.
I don’t want to minimize this bedroom tidying by Figure (although I suppose I’m going to), but in the context of doing a task like this in place of a human, it really illustrates what these robots are comfortable with, and what they’re not.
Curious about what happens when street dance meets embodied AI? From smooth choreography to dynamic flips, NIX is exploring movement, rhythm, and real-world interaction through embodied AI. We’ll make NIX available—FOR FREE!—to selected partners from global universities, robotics labs, and creative technologists.
We introduce and open-source the Unified Autonomy Stack, a novel solution for resilient autonomy across aerial and ground robot morphologies. The architecture combines multimodal perception, multibehavior planning, and multilayered safe navigation to deliver mission-level autonomy across diverse robot morphologies. It fuses lidar, radar, vision, and inertial sensing to enable robust localization and mapping, vision-language-based scene reasoning, multibehavior planning, and layered safety through map-based avoidance, deep learned policies, and control barrier functions. The system supports Global Navigation Satellite System–denied navigation in perceptually degraded environments, exploration, object discovery, and inspection, and has been validated on multirotor and legged robots in challenging settings, demonstrating resilient performance.
Any robot doing anything consistently over a long period of time is impressive. Having said that, you want to be very careful about claiming that any robot operates at “human performance levels,” especially in a somewhat complex manipulation task, because humans are very, very good at stuff like this.
Robust.AI cofounder and CTO Rodney Brooks, ranked #44 on the Forbes 250 America’s Greatest Innovators list, sits down for a Q&A ahead of his panel discussion at the Forbes America Innovates event in San Francisco. We asked him two questions: What makes innovation in robotics such a challenge? What does the current surge in AI mean for robotics today?
This is one of the best robotic research videos I’ve ever seen—and don’t worry, according to the credits it’s not AI. And make sure to watch after the credits!
EFGCL is a guided-reinforcement learning method that efficiently enables highly dynamic motions through the use of assistive forces. In this work, we successfully achieved several dynamic motions, including jumping, backflips, and lateral flips.
Humanoid robots promise general-purpose assistance, yet real-world humanoid loco-manipulation remains challenging because it requires whole-body stability, dexterous hands, and contact-aware perception under frequent contact changes. In this work, we study dexterous, contact-rich humanoid loco-manipulation.
More than just technology, KATA Friends is a lifelike AI companion designed to see your world, feel your touch, and understand your heart. With expressive movements, evolving emotions, and natural conversations, Noa and Niko both grow alongside you to become a presence uniquely yours. From curious head tilts and playful reactions to ever-changing eye expressions and a soft, innocent voice, every interaction feels warm, personal, and alive.
One of the biggest challenges in robotics right now isn’t the hardware. It’s data. While many data-collection methods are effective, handheld data collection can create a diverse dataset of environments, conditions, and strategies for completing manipulation tasks. The Koala platform codesigned the handheld grippers and robot grippers around the same linkage mechanism, the same degrees of freedom, and the same force transmission. The human feels through the linkages what the robot will feel through its actuators.
Many roboticists (and at least one robotics journalist) have been seduced by the dream of a robot butler. And the rampant popularity of videos showing humanoid robots doing household tasks in improbably clean kitchens and unrealistically tidy bedrooms suggests that we’re not the only ones interested in a robot that can do our chores. But for all kinds of reasons, legged humanoids are not yet ready for industrial or commercial applications at scale, and home applications (if people even want them), I would argue, are even farther away. Even so, ludicrously well-funded humanoid robotics companies are now ramping production while explicitly promising that their robots will be doing ‘housework.’
So what about that robot butler dream, then? It still exists! All you have to do is forget about legs, arms, hands, faces, and focus on what really matters: mobility and manipulation. This is what Hello Robot’sStretch robot is unapologetically all about, and the newest version being announced today, Stretch 4, is closer than ever to a robot that could safely do practical work in the home at an accessible cost.
Hello Robot says Stretch 4 is “built for the real world.”Hello Robot
“With Stretch 4, we wanted to make the transition from a research platform to something that is truly deployable,” explains Aaron Edsinger, Hello Robot co-founder and CEO. This version, while ready for research and enterprise customers now, is designed for pilot deployments to help Hello Robot understand how to scale in the home. “This has been our most difficult design process,” adds co-founder and CTO Charlie Kemp. “We had a lot of fear of ‘second-system syndrome,’ where you add all the features you didn’t get to initially and end up with a monstrosity. But since we founded the company on making simple, minimalist robots, every time we added complexity it was an emotional challenge. Navigating that fear resulted in a nice compromise that sits in a great spot, rather than being a maximalist humanoid.”
Stretch 4 Upgrades
The biggest change from the previous version of Stretch is the addition of an omnidirectional base, meaning that the robot can translate in any direction without having to turn first. This makes it much easier to control (especially for novice users), but omnidirectional bases are significantly more complicated to design and build. What ultimately made it possible for Stretch were new types of omnidirectional wheels developed for powered wheelchairs, along with a solid six months of focused development by Hello Robot.
A redesigned sensorized head gives Stretch more options for teleoperation and autonomy.Hello Robot
Stretch 4 also ditches the cute little pan-tilt head for a more complex sensor suite with a much wider field of view. “We started out wanting to use lots of cheap cameras to keep costs low, like Tesla does,” Edsinger tells us. “But we ended up with an approach closer to Waymo’s: the richer and more reliable your data, the safer and more intelligent the robot can be.” There are a pair of hemispherical lidars, Luxonis cameras for vision and navigation, and a wrist-mounted depth camera for manipulation. The robot’s primary system runs on an Intel NUC 15, plus an Nvidia Jetson Orin NX for researchers to play with for visual processing or AI.
Philosophy on Autonomy
Hello Robot’s general philosophy on autonomy is to have a human in the loop, but that can take many different forms ranging from direct control to purely supervisory control. The robot will ship with a baseline of autonomous capabilities that include mapping, navigation, and self-charging, along with demo-ready features like autonomous grasping. But unlike most other robotics companies, Hello Robot isn’t looking to use their hardware to collect a stupendous amount of data in the concerningly vague hope that commercially viable autonomy will follow.
“Stretch has huge advantages in safety, cost, and capability,” Kemp says. “I’d much rather be the platform that foundation model developers target.” Edsinger agrees: “We do want to partner with foundation model companies to explore things like dexterous in-home manipulation, but we aren’t the ones to build those foundation models.”
In-Home Pilots
While earlier versions of Stretch were primarily for research, Kemp tells us that Stretch 4 has been explicitly designed to be piloted in the homes of people with severe mobility impairments. Hello Robot will be happy to sell you one (or lots, I’m guessing) for commercial or industrial applications, but the broader goal with Stretch 4 is to use remote testing and in-home evaluations to work towards a robot that’s useful and reliable enough that it can provide consistent daily value for disabled users.
A holonomic base and an extendable arm make for a capable robot without the complexity.Hello Robot
Part of why I’m optimistic about Stretch finding near-term success in this role is precisely because it’s not a humanoid. One of the primary arguments for humanoids is that they’re worth pursuing because they can better operate in environments designed for humans, where legs and five-fingered hands are tangible advantages. But those very same environments often exclude an entire subset of humanity—a subset of humanity that we will all likely join at some point, because the best that any of us can ever say is that we are not disabled yet.
Why Not Humanoids?
A key partner for Hello Robot throughout the Stretch development process has been Henry Evans. Evans is paralyzed and cannot speak, although he can use a computer (for controlling robots, among other things) and type at about 15 words per minute. I spoke with Evans about his thoughts on the idea of a humanoid assistive robot, compared to a robot like Stretch. “The question is: What benefit does a bipedal robot offer to a person who can’t walk?” Evans asks. “Their entire environment has been modified to accommodate wheeled conveyances. Automobiles don’t have legs, and neither should home robots. Wheels are cheap, stable, precise, require very few controls, and don’t have to be invented.”
Henry Evans has been testing a Stretch 4 as a home assistive robot.Hello Robot
Evans also points out that humanoids can require the simultaneous control of dozens of degrees of freedom. “A paralyzed person who can’t talk (like yours truly) can control maybe one or two joints at a time with today’s control mechanisms, if they are lucky.” Evans believes that AI, along with Brain Computer Interfaces (BCIs), show promise for dramatically increasing what he can do when it comes to motion. “Remember, though, a paralyzed person has no movements to mimic, so until a perfectly tuned BCI gets here and facilitates a true humanoid body surrogate, I don’t think it will work. And even then, I don’t see the advantage of legs for assistive care robots. I am willing to be proven wrong, though, and will test-drive almost anything once, so bring it on!”
Kemp and Edsinger, who have many decades of humanoid experience between them, feel similarly. “There are applications where the human form is fundamental,” Kemp says. “But for many applications, the value of the human form is unclear or even problematic. Jumping to the conclusion that robots must be humanoid means missing opportunities to take advantage of the structured indoor environments that we’ve already created.”
Georgena Moran and her sisters tested Stretch 4 at the California Academy of Sciences Museum, allowing her to interact with the exhibits from home.Hello Robot
And of course there’s the question of safety, which Evans brings up. “My caregivers and I have been testing robots in my home to assist us for about 15 years, and the very first concerns are: Where is the emergency stop, and how do you activate it? It gets used surprisingly often. The thing is, when a wheeled robot gets emergency stopped, it freezes in place. When a bipedal robot gets run-stopped, it collapses on anything under it, including the patient.” Kemp agrees. “The safety aspect of humanoids in a home freaks me out. I don’t know how someone can confidently think about safety with a humanoid in a home.”
Robots for Sale
However you feel about humanoids, here’s one more reason why Stretch feels like a much more realistic solution for in-home assistive robots right now: You can actually buy one, and at US $29,950, it’s very affordable, as mobile manipulators go. Edsinger and Kemp are planning to leverage in-home Stretch 4 pilot deployments to make the next version of Stretch the one that can be commercially sold for home assistance. At the rate that Hello Robot has been releasing new hardware, that could easily be within the next year or so—and my guess is that Stretch 5 is very likely to be the first practical, affordable assistive robot for home use. It may not look like Rosie, but it promises to be safe, and it works.
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
Introducing GENE-26.5—the first AI brain to give robots human-level physical manipulation capabilities. Cooking a full meal. Cracking an egg one-handed. Conducting lab experiments. Wire harnessing. Even playing the piano. Tasks that were impossible for robots. Until now.
This is Labububot—one of the rarest monsters on Earth. Twelve Labubu heads are reconstituted into a single spherical form: a Frankenstein’s Monster of pop culture iconography. Labububot is a playful critique of social robots, and a question made physical—what do the monsters we make reveal about the monsters we are?
Watch Spot crouch, jump, climb boxes, and leap across gaps, controlled by a neural network trained with reinforcement learning (RL) and multi-expert distillation.
Additive manufacturing has become an enabling technology, but existing techniques are not capable of directly 3D printing high-current electromagnetic actuators due to material and design limitations. In this work, a novel 3D-printable, multilayer, wave-winding topology is created for high-efficiency electric motors.
NASA is pushing the limits of flight on Mars—by spinning helicopter rotor blades so fast, they’re breaking the sound barrier. During recent tests at NASA’s Jet Propulsion Laboratory, engineers accelerated the tips of next-generation rotor blades beyond Mach 1 inside a special chamber that simulates the atmospheric conditions of the Red Planet.
Automated inspection of steel structures using magnetic climbing robots can reduce costs and improve safety, but many such structures feature interior corners that are challenging for wheeled or tracked robots to traverse. We present the first magnetic-wheeled robot to use X-ray fluorescence for steel structure inspection, Sally, capable of overcoming all interior corner transition types, traversing small obstacles, and maneuvering in tight spaces.
Jim Fan, who leads the embodied autonomous research group at Nvidia, returns to AI Ascent to argue that robotics is entering its endgame—and that the playbook is already written.
Login
