Definitely Not Skynet
The Body Electric
Intelligent, curious, capable: our new robot friends are here. A record of the machines that move, lift, walk, and work — and the human and AI stakes of it all.
Marc Lore says that AI will soon enable anyone to open a restaurant
Wonder's Marc Lore is pitching a future where robotic kitchens and AI let anyone conjure a virtual food brand from a prompt, turning restaurant ownership into something closer to a software product than a human craft. The labor implications are not incidental here — kitchens are one of the largest employers of low-wage workers in the country, and 'anyone can open a restaurant' necessarily means fewer people are needed inside one. Whether this reads as democratization or displacement depends almost entirely on which side of the counter you're standing.
Microbot Medical to expand veteran access to robotic surgery with LIBERTY
Microbot Medical is routing its LIBERTY surgical robot into federal programs through a service-disabled veteran-owned small business vendor, which is one of those structural deals that quietly shapes who actually gets access to advanced care. Robotic surgery has long clustered at well-resourced hospitals, so threading it into VA-adjacent pipelines is worth watching — not as a press release, but as a proof point for whether the technology can reach people outside elite medical centers. The human stakes here are as concrete as they get.
Rivian faces a class action lawsuit over self-driving in its early vehicles
A class action against Rivian over self-driving capabilities that were promised but not delivered is a reminder that the gap between marketing language and actual autonomy has real legal weight — and real consequences for people who made purchasing decisions based on it. This is the accountability layer that the industry has been slow to reckon with, and it won't be the last suit of its kind as the distance between demo and deployment keeps getting litigated. The body of the car is physical; the trust gap is the thing that's embodied here.
Kinova launches KIMA medical robotic arm
Kinova built KIMA from scratch for the clinic rather than retrofitting an industrial arm, which is a more honest approach to the gap between factory floors and hospital rooms. The design-from-ground-up decision matters because the bodies that show up in clinical settings — fragile, variable, frightened — are nothing like the parts on an assembly line. Whether KIMA can actually earn the trust of clinicians and patients is the real test ahead.
RealSense unveils AI-native D585 Pro depth camera for robots
Depth perception is the quiet prerequisite for almost everything robots need to do in the physical world, and RealSense's D585 Pro — more than double the quality of its predecessor on a new Gen 5 system-on-chip — is infrastructure-level news even if it doesn't look like it. Better depth means robots make fewer catastrophic mistakes about where things are, which matters whether the robot is in a surgical suite, a warehouse aisle, or a living room. It's the kind of component story that only becomes visible when it fails.
Pickle Robot adds Tesla veteran as first CFO
Pickle Robot — which automates trailer unloading, one of the most physically punishing jobs in warehouse logistics — has hired its first CFO and reportedly deepened its partnership with UPS, signaling a move from early deployment into scaled commercial operations. Bringing in a Tesla veteran at the finance level is a bet that the hard part is no longer making the robot work but making the business work. The jobs being automated here are real and hard on bodies, which means the labor question doesn't go away just because the CFO hire got announced.
Oshen built the first ocean robot to collect data in a Category 5 hurricane
Oshen's C-Star robot went somewhere no crewed vessel reasonably could — into a Category 5 hurricane — and came back with data that forecasters can't get any other way. That's the clearest articulation of what autonomous robots are actually for: not replacing human judgment, but extending human reach into environments that would kill a person. Multiple government agency contracts suggest this isn't a stunt; it's the beginning of a serious environmental sensing infrastructure.
Waymo recalls over 3,800 robotaxis that might drive onto closed freeways
Waymo is recalling more than 3,800 vehicles after a software defect was found that could send a robotaxi into a closed freeway construction zone at full speed — exactly the kind of low-probability, high-consequence failure that safety engineers lose sleep over. It's a reminder that at fleet scale, edge cases stop being hypothetical, and that the governance infrastructure around physical AI systems is still being built in real time. The recall process itself, borrowed from conventional automotive regulation, is doing exactly what it's supposed to do.
Pegasus Tech Ventures launches $60M fund for physical AI startups
A $60 million fund jointly backed by Pegasus Tech Ventures and CYBERDYNE — yes, that CYBERDYNE, the Japanese exoskeleton company whose founders apparently named it with full awareness of the joke — is a modest but telling signal that patient capital is organizing specifically around the physical AI category. The fund targets robotics, automation, healthcare, and intelligent systems, which is a wide enough aperture to catch both the infrastructure plays and the application layer startups. What's worth watching is less the dollar figure than whether the CYBERDYNE connection brings genuine assistive and medical robotics expertise into the portfolio thesis.
Upside Robotics is reducing fertilizer use and waste in corn crops
Upside Robotics makes solar-powered autonomous machines that move through corn fields and apply fertilizer with enough precision to cut overall use by roughly 70 percent — a number that, if it holds at scale, matters for soil health, water quality, and farm economics simultaneously. This is embodied AI doing something genuinely difficult: operating in unstructured outdoor environments, in variable weather, on crops that are alive and growing. It's also a case where automation isn't primarily a labor displacement story but a resource efficiency one, though the two rarely stay separate for long.
Qualcomm's partnership with Neura Robotics is just the beginning
Qualcomm's IQ10 processor was announced at CES as a chip purpose-built for embodied AI, and Neura Robotics building its next humanoids on top of it signals that the semiconductor layer of the robot stack is finally being treated as seriously as the software layer. When a chipmaker of Qualcomm's scale commits to physical AI as a platform rather than an adjacency, it tends to pull capital, talent, and ecosystem partners along with it. The human question underneath the business story: whose bodies and whose jobs are being modeled when engineers decide what counts as the standard inference workload for a robot brain.
Evaluating humanoids for surface finishing applications
Surface finishing — sanding, grinding, polishing — is one of those jobs that sits at the uncomfortable intersection of physical demand, chemical exposure, and skilled touch, which makes it a genuine test case for whether humanoids can do something useful rather than just walk impressively across a stage. The Robot Report takes a methodical look at what it actually requires to put a general-purpose humanoid into this workflow, where the ergonomics favor the form factor but the consistency requirements are unforgiving. The honest answer emerging from evaluations like this is that humanoids aren't replacing purpose-built industrial arms yet, but they're starting to show up in the same sentence without embarrassment.
AI coding agents taught robots how to install GPUs and cut zip ties
Nvidia has built a self-improvement loop where teams of AI coding agents autonomously generate, test, and refine the training programs that teach robots physical tasks — things like installing GPUs and snipping zip ties, the kind of fiddly work that has long resisted automation. The significance here isn't the specific tasks; it's the architecture: robots getting better at having bodies without a human writing every step. That feedback loop, if it scales, changes the pace of everything else on this list.
The Secret to Marathon-Winning Humanoid Robots
On April 19, 2026, Honor's Lightning robot ran a half-marathon in 50:26 — faster than any human ever has — and the engineering reason turns out to be unglamorous but decisive: liquid cooling threaded through the motors like capillaries, combined with gear ratios optimized for running rather than walking. The piece is a careful reminder that breakthroughs rarely involve magic, and that the same tradeoffs making Lightning a remarkable runner make it a less useful general-purpose machine. Worth sitting with the author's closing note: comparing a robot's time to a human record tells us almost nothing useful about either achievement.
Gecko Robotics lands the largest US Navy robotics deal yet
Gecko Robotics has signed a five-year contract to deploy its inspection robots across the U.S. Navy's fleet, using autonomous crawlers to monitor hull and structural integrity and predict maintenance needs before failures occur. It's a significant validation of the inspection-and-monitoring use case for embodied robotics — unglamorous work, but the kind where a robot's ability to go where humans can't, or shouldn't, makes a concrete safety argument. The scale of a naval fleet also means this is one of the largest real-world deployments of physical AI in a high-consequence environment to date.
Sanctuary AI validates physical AI performance at Tier 1 automotive supplier
Sanctuary AI is reporting production-validated performance at a Tier 1 automotive supplier, choosing to demonstrate physical AI capability in a running factory rather than wait for a fully humanoid platform to mature. The framing — 'not waiting for humanoid robots' — is telling: it suggests the company sees the underlying AI as the durable asset, deployable now in whatever body makes sense. Two automotive-supplier stories in one day is worth noting; the question of which factory floors actually adopt this technology, and what happens to the workers already on them, is the story underneath the story.
The next humanoid robot might not look human at all
French startup Genesis AI, backed by former Google CEO Eric Schmidt, is arguing that 'humanoid' is really about capability rather than form — its Eno robot has human hands but may forgo legs, a head, or conventional bipedal structure altogether. It's a serious design question hiding under a provocative frame: if the goal is a robot that operates in human spaces and uses human tools, how much of the body actually needs to be replicated? The hands, at least, Genesis is keeping — which says something about where the real difficulty of embodiment lives.
Autonomique deploys semi-humanoid robots and AI at Canadian Tier 1
Autonomique has graduated its AI-powered semi-humanoid robots from pilot to full production at F&P Manufacturing, a Canadian Tier 1 automotive supplier, with plans to roll out across F-tech's global network. This is the kind of quiet milestone that matters more than a press-conference demo: a robot earning its place on an actual production floor, not just surviving a controlled test. The automotive supply chain is one of the highest-stakes proving grounds for embodied AI, and 'global network' signals that someone upstream has decided this is ready to scale.
Built Robotics, Penn xLAB to develop physical AI for construction
Built Robotics and Penn's xLAB are partnering to capture richer construction-site data to improve AI models, with an explicit focus on safety — a framing that acknowledges construction remains one of the most dangerous industries in the world and that autonomous systems operating on jobsites carry real stakes beyond productivity metrics. The research angle here is about data quality and model improvement, but the application domain means the margin for error is measured in human injury rates, not benchmark scores. Whether academic-industry partnerships like this one can generate data that generalizes beyond controlled conditions will be the test.
Video Friday: Watch This Running Robot Not Fall Down Stairs
This week's reel from IEEE Spectrum is a useful temperature check on where humanoid locomotion actually stands: genuinely impressive recovery from a Deep Robotics platform, first outdoor steps from IHMC's Alex, and Boston Dynamics teaching Atlas to play football — each a real milestone, each still carrying the asterisk of controlled conditions and unknown generalization. The honest question the curator keeps asking — how much was luck? — is the right one, and the fact that it gets asked at all is why this kind of watchful enthusiasm matters. Bodies learning to move through the world are still learning, and the gap between a good video and a reliable robot remains wide.
PSYONIC partners with ABB Robotics to apply human touch to robot dexterity
PSYONIC is doing something quietly profound: taking sensory and control data from people who use its Ability Hand prosthetic — people who have strong reasons to demand that a hand actually work — and feeding those human-derived signals into ABB's GoFa cobot arm to improve robotic grasping. The flow of knowledge here runs from lived human experience outward into industrial machinery, which inverts the usual direction and carries real ethical weight about whose expertise gets counted. If it works, it's a rare case where the needs of people with disabilities directly improve the capabilities of general-purpose robots, rather than the other way around.
New robotic control software avoids jamming their joints
Researchers have developed control software that lets robots learn their own kinematic limits — essentially teaching them not to fight their own bodies — and, notably, to share what they learn across robots with different hardware configurations. This matters because one of the quiet bottlenecks in robot deployment isn't intelligence in the abstract but the chronic mismatch between a robot's learned behaviors and the physical reality of its joints under load. Cross-hardware knowledge transfer, if it holds up outside the lab, could meaningfully compress the time it takes to bring a new robot platform to reliable operation.
Modernizing the global economy with industrial robotics is needed but not inevitable
A measured column that resists the assumption that industrial automation is simply a wave arriving on schedule — instead treating it as a set of choices that face real friction from capital costs, workforce readiness, supply chain fragility, and institutional inertia. The distinction it draws between service robotics (where adoption is accelerating) and industrial automation (where demand is high but obstacles are larger) is a useful corrective to coverage that treats the whole sector as monolithic. This is the kind of piece that ages better than the breathless ones, because it's asking which conditions have to be true for the transformation to actually happen.
Genesis AI launches Eno general-purpose robot
Genesis AI is entering the crowded general-purpose humanoid field with Eno, targeting commercial deployments by end of 2026 and home use afterward — a sequencing that reflects an industry-wide lesson that homes are harder than factories. The timeline is ambitious in the way most humanoid timelines are, and the meaningful question isn't whether Eno can be demonstrated but whether it can be manufactured, maintained, and trusted in environments it wasn't explicitly trained for. Worth watching as one data point in what is becoming a very busy market of robots promised for everyday life.
3D-printable humanoid legs let robotics experiments run wild
Hugging Face's $2,500 bipedal robot project democratizes humanoid research by making the hardware accessible to universities and independent developers. The open-source approach could accelerate progress by letting thousands of researchers experiment with walking algorithms instead of just the well-funded labs. It's a bet that distributed innovation will solve embodiment challenges faster than centralized corporate research.
Video Friday: Robotic Motion Discovery Reveals Unusual Behaviors
MotionDisco shows humanoid robots discovering their own movement patterns without human demonstration—and some of the results are genuinely strange. The framework represents a shift toward robots learning embodied intelligence from scratch, though the line between 'discovery' and 'accident' remains fascinatingly unclear. What emerges when we let machines explore their own physicality?
Gatik to bring autonomous freight to PepsiCo's North American supply chain
PepsiCo's partnership with Gatik represents autonomous vehicles moving from pilot programs to actual supply chain integration. The focus on middle-mile freight—moving products between facilities rather than to consumers—shows how automation is quietly reshaping logistics infrastructure. This is the unglamorous but economically significant work where robots are already changing how goods move through the world.
Inside CES 2026's "physical AI" takeover
After years of chatbots and image generators, AI finally left the screen at CES 2026, with everything from redesigned humanoids to AI-powered appliances. The shift from digital to physical intelligence represents a fundamental change in how AI touches daily life—no longer confined to software, but walking, lifting, and working alongside humans. The spectacle raises as many questions about readiness as it answers about capability.
NEURA Robotics to raise up to $1.4B in Series C funding for physical AI
NEURA's massive funding round signals serious institutional belief that humanoid robots are moving from lab curiosities to commercial realities. The company plans to scale both robot learning platforms and global manufacturing—a bet that the technical challenges are solved enough to justify industrial-scale production. Whether the market is ready for what they're building remains the open question.
Defining Autonomy for Wellness Robots in Senior Care
A new framework attempts to measure how autonomous care robots actually are, using levels similar to self-driving car standards. The paper tackles the intersection of demographic crisis and technological promise—where robots might fill gaps in senior care that human workers cannot. The question isn't just whether robots can help, but how much agency we're comfortable giving them in intimate caregiving moments.
Hello Robot is recognized by World Economic Forum as a tech pioneer
The World Economic Forum honored Hello Robot for its Stretch system, which provides mobile manipulation capabilities specifically designed for older adults and people with disabilities. This recognition highlights how assistive robotics is moving beyond research labs into real homes, addressing genuine human needs rather than just technological showmanship. It's the kind of application that reminds us why we build these machines in the first place—not to replace human capability, but to restore it where it's been lost.
Robot runner handily beats humans in half-marathon, setting new record
A Chinese humanoid robot completed a half-marathon faster than any human, demonstrating remarkable advances in robotic locomotion and endurance that go far beyond laboratory demonstrations. The achievement showcases China's rapid progress in humanoid robotics while raising questions about what it means when machines can outperform humans in activities we consider fundamentally human. It's a milestone that feels both impressive and slightly melancholy—a reminder that our bodies, like our minds, may not remain uniquely ours for long.
From folding boxes to fixing vacuums, GEN-1 robotics model hits 99% reliability
Generalist's new GEN-1 model represents a significant leap in robotic reliability, achieving 99% success rates across diverse tasks from folding boxes to repairing appliances. The system can adapt to disruptions and figure out moves it wasn't explicitly trained for, suggesting we're moving closer to truly general-purpose robots. This kind of reliability threshold matters because it's the difference between a research curiosity and a technology that businesses might actually trust with real work.
The Internet can't stop watching Figure AI's humanoid robots handling packages
Figure AI's 24/7 livestream of humanoid robots working in a warehouse has become an unexpected internet phenomenon, drawing viewers who seem fascinated by the sight of human-shaped machines doing mundane work. The robots handle packages with deliberate, almost meditative movements that reveal both how far the technology has come and how different machine labor looks from human labor. There's something both mesmerizing and unsettling about watching our mechanical doubles work around the clock—a preview of a future that feels simultaneously inevitable and strange.
Award-Winning Researcher Trains Robots to Make Educated Guesses
Yen-Ling Kuo's award-winning research tackles one of robotics' fundamental challenges: how robots handle uncertainty when facing unfamiliar situations. Her Diff-DAgger method enables robots to recognize when they're out of their depth and ask for human help only when truly needed, reducing supervision requirements while improving task completion rates by 20 percent. It's the kind of unglamorous but essential work that makes the difference between a robot that works in the lab and one that works in your living room.
Visual Language Models Train Robots to Read Human Emotions
Researchers trained collaborative robots to read human emotions using vision language models that consider context beyond facial expressions, achieving better emotional recognition than traditional AI systems. However, their study with 40 volunteers revealed a sobering truth: while people appreciated emotionally adaptive apologies from robots, trust still hinged on competent task performance rather than social polish. The work illuminates both the promise and limits of emotional intelligence in machines—a robot that understands your frustration is still just a robot if it can't do its job.
How JPL Keeps the 13-Year-Old Curiosity Rover Doing Science
Thirteen years into its mission, Curiosity continues to perform science on Mars through ingenious engineering workarounds—including running on less than 1% of its original memory after creative software solutions. The rover's longevity comes not just from robust initial design, but from continuous human ingenuity adapting to hardware failures and constraints. It's a master class in making machines work far beyond their expected lifespan, offering lessons for any robot that needs to operate autonomously in harsh environments.
Humanoid robots start sorting luggage in Tokyo airport test amid labor shortage
Japan Airlines is testing humanoid robots for luggage handling and aircraft cleaning at Haneda Airport as the country grapples with severe labor shortages. The pilot program represents a real-world deployment of humanoid automation in a critical infrastructure role, moving beyond demonstration videos to actual work. It's a preview of how demographic pressure and workforce gaps may drive adoption of humanoid robots faster than pure technological readiness might suggest.
Allegedly trashing Airbnbs to test robots puts startup in legal trouble
A robotics startup faces a $12,000 lawsuit for allegedly damaging an Airbnb property while testing robots, highlighting the messy reality of training embodied AI systems in real-world environments. The incident underscores how robot development requires physical space and practice that can conflict with property rights and social norms. It's a reminder that intelligence leaving the screen means robots—and their creators—must navigate the full complexity of the physical world, including its legal frameworks.
Unitree will sell you a massive 'transformable mecha' for $650,000
Chinese robotics company Unitree has unveiled the GD01, a massive piloted mech suit that can switch between bipedal and quadrupedal modes for $650,000. While the practical applications remain unclear, it represents the bleeding edge of human-machine collaboration and the ongoing push to scale humanoid robotics beyond human proportions. The mech sits somewhere between science fiction spectacle and serious engineering, embodying our complex relationship with machines that amplify human capability.
This AI startup will clean your home for free to train future robots
AI training startup Shift offers free house cleaning services while recording workers to create training data for future robots. It's a clever arbitrage between human labor costs and the value of embodied AI training data, but raises questions about worker consent and data ownership. The approach highlights how the path to robot autonomy still runs through capturing and commodifying human expertise.
This DIY Bipedal Robot Used Pneumatic "Air-Muscles" Instead of Motors
In 1987, a group of British enthusiasts built Shadow Walker in an attic using pneumatic muscles instead of motors, pioneering humanoid robotics before Honda's famous ASIMO. Their struggles to achieve basic walking presaged the decades of work still needed to make humanoid robots practical. The story reminds us that today's walking robots stand on the shoulders of countless amateur builders who dared to imagine machines that move like us.
Amazon develops a warehouse robot that workers can speak to
Amazon's upgraded Proteus warehouse robot now responds to natural language commands instead of requiring specialized software, part of the company's growing pivot toward automation that increasingly replaces human workers. The AI-powered upgrade allows employees to assign tasks to the heavy-lifting robots as they would communicate with colleagues, though it arrives amid broader concerns about the human cost of warehouse automation.
The skeptic's guide to humanoid robots going viral on the Internet
Viral robot demonstrations can distort public perceptions of robotic capabilities, creating unrealistic expectations about when and how these machines will integrate into daily life. The spectacle of impressive robot videos often obscures the significant technical limitations and deployment challenges that remain, requiring a more measured understanding of what these systems can actually accomplish in real-world conditions.
Open-Source Software Is Starting to Help Robots Think
The open source movement that accelerated AI development is now being applied to robotics, with companies like Nvidia, Hugging Face, and Alibaba releasing tools and models for robot reasoning and decision-making. If successful, these platforms could lower the barrier to building capable robots as dramatically as they did for AI applications, though commercial incentives complicate the collaborative spirit that originally built ROS.
Tech companies desperately want to film you doing chores
AI training startup Shift offers free house cleaning in exchange for filming its workers scrubbing, vacuuming, and mopping—footage that will train future domestic robots. The unusual arrangement highlights how desperately robotics companies need real-world training data for the mundane tasks we'd happily outsource, revealing both the promise and the awkward reality of building machines to do our household labor.
Video Friday: Atlas Versus a Fridge
Boston Dynamics' Atlas demonstrates lifting a mini-fridge using whole-body control and superhuman range of motion, showcasing the underlying reinforcement learning systems that let it handle heavy objects by bracing and accounting for mass and inertia. The demonstration marks what the company calls "a critical shift" from lab experiments toward dynamic industrial applications, though the true breakthrough may be less visible than the spectacle suggests.
Beyond Dexterity: Why Contact May Define the Next Era of Robotics
AGILINK's balloon-twisting robot demonstration at ICRA 2026 illustrates a fundamental shift in robotics: the hardest problems begin after contact occurs. The company's new OmniHand 3 Ultra-M combines direct-drive actuation with dense tactile sensing to explore what they call "contact intelligence"—the ability to establish, maintain, and adapt physical interaction as conditions continuously evolve.