Hello Robot is working toward a future in which robots give people greater independence in their everyday lives.

The California robotics company expresses that mission in two short phrases: “Robots with purpose” and “Technology with empathy.”

Hello Robot’s Stretch platform has become a widely used tool for robotics research and development. But one of the clearest expressions of the company’s purpose can be found in its assistive robotics work with people with disabilities, older adults and their care partners.

I recently spoke with Hello Robot co-founder and CEO Aaron Edsinger about the company’s path from the research lab into real homes.

A robot designed around people

Stretch does not look like the humanoid robots currently attracting much of the robotics industry’s attention. It moves on a compact wheeled base. A vertical mast allows it to reach different heights. Its telescoping arm extends into the surrounding environment, while cameras and sensors help it understand what is happening around it.

In one pilot Edsinger described, a person wanted to use Stretch to scratch an itch. In another, a participant wanted the robot to hold a drink so that a caregiver would not need to remain beside them throughout the process.

“These are very simple tasks, but they’re achievable today,” Edsinger said. “And they provide immense value to the people who are working with the robot.”

The value extends beyond completing the task itself. Each action gives one person greater agency while giving the care partner more freedom.

Assistive technology is often evaluated according to what it can do for the person receiving care. Hello Robot considers the care partner part of the equation.

Edsinger spoke about an aunt who is bedbound and alone at night. If she needs water, getting up without assistance could be dangerous.

A robot does not need to function as a fully autonomous butler to help in that situation. It could bring a drink, connect her with a remote caregiver or allow someone to perform a safety check. Relatively simple capabilities could support the existing care network rather than attempting to replace it.

Helping people age in place

Hello Robot is also participating in research involving older adults and people with cognitive impairments.

Researchers at the University of New Hampshire have placed a socially assistive version of Stretch in the homes of people with dementia and their care partners. The robot combines its onboard capabilities with smart-home sensors to provide reminders, support daily routines and monitor certain safety concerns.

In one deployment, the robot helped remind a man with dementia to eat, drink, exercise and complete parts of his daily hygiene routine.

Because the robot can move, it can bring its screen and reminders to the person at the appropriate place and time. A tablet left on a bedside table cannot do that if it has been misplaced, forgotten or left uncharged.

The research remains at an early stage. Stretch is not yet a consumer caregiving product, and Hello Robot is careful not to present it as one.

But home deployments help researchers understand what people genuinely need from a robot, what they will accept in their homes and which capabilities are reliable enough to create value today.

For an older adult, that value could mean remaining at home longer. For a family member, it could mean being able to leave the house without constant worry. For a professional caregiver, it could mean having help with routine activities while preserving more time for human care.

Why Hello Robot did not build a humanoid

Edsinger has extensive experience with humanoid robots. He studied at MIT with Rodney Brooks, spent more than a decade working on humanoids and previously founded a humanoid robotics company. He later served as director of robotics at Google, where he began thinking seriously about what it would take to place robots in homes.

His conclusion was that the industry needed a different approach.

After seeing research from his longtime collaborator and eventual Hello Robot co-founder Charlie Kemp, Edsinger became interested in a minimal robot with a telescoping arm.

“It has to be more like a Roomba with an arm than a large humanoid clomping around in your house,” he recalled.

The company’s rejection of the humanoid form is closely connected to its assistive mission. A useful humanoid must be large and powerful enough to balance, reach counters and manipulate objects. That size, weight and force introduce risks when the robot is working near children, pets, older adults and people who cannot easily move away from it.

“Safety is not something you just figure out later,” Edsinger said. “Particularly for the home and working with people who are maybe more vulnerable, older adults, you have to think about safety from the get-go.”

Hello Robot designed Stretch around those physical realities. The goal was a machine that would be simpler, safer, more practical and easier to place in a real home.

Edsinger calls the resulting design philosophy “robotic cubism.” The company took capabilities associated with the human body and rearranged them into a different form.

But robotic cubism is not an artistic exercise. It is a practical response to the people and environments the robot is meant to serve.

Building a community around assistive robotics

Hello Robot has sold approximately 330 earlier-generation Stretch robots across 23 countries. The machines are being used by universities, research laboratories, startups and other organizations exploring robotics applications.

The platform’s software is open source, allowing researchers to build applications, publish their findings and share code with the broader Stretch community.

That community has explored robots that retrieve walkers for people with Parkinson’s disease, help locate misplaced objects, support physical therapy, assist with bathing, bring medication and enable people with severe motor impairments to manipulate objects.

The research platform serves a practical purpose for Hello Robot. It allows many teams to investigate different needs while helping the company learn what Stretch is good at, where it fails and which applications could eventually support a sustainable product.

Stretch 4, launched in May, represents the company’s largest redesign of the robot so far. It includes more extensive sensing, improved reliability and features intended to support pilots outside traditional robotics laboratories.

The current Stretch 4 remains a research and development platform. It is not yet a household appliance that someone can purchase, unpack and use without technical support.

The company is now working through the difficult questions required to bridge that gap: How will these robots be supported? Who will pay for them? Which tasks provide enough value to justify their cost? How much autonomy is necessary, and where should a person remain in control?

A different definition of progress

When I asked Edsinger what we might be discussing a year from now, he did not predict a humanoid breakthrough or a general-purpose robot capable of handling every household chore.

He hopes we will be talking about successful pilots, real deployments and a clearer path to bringing robots into homes.

“What we know is that there’s a real demand, there’s a real need there,” he said.

If you missed our preview

In this one-minute excerpt, Edsinger explains “robotic cubism,” why Hello Robot chose not to build a humanoid and why safety must be considered from the beginning.

#robotics #physicalai #embodiedai

Producing that many robots is difficult too. But it may not be the hardest part. The next test is whether those robots can leave the demonstration stage behind and perform useful work reliably in factories, stores, care facilities, and eventually our homes.

Peng’s answer was direct: “We’re in the era converting from demo phase to deploy phase,” he said.

What does 15,000 actually mean?

On June 28, AgiBot’s 15,000th robot rolled off the production line. The company had reached 10,000 only three months earlier, an impressive ramp for a business founded in 2023.

The figure represents cumulative production, not 15,000 robots working autonomously in the field. Production, shipments, and active deployments are different measures.

Still, AgiBot has achieved meaningful scale. Omdia reported that it shipped 5,168 humanoid robots in 2025, representing 39 percent of the market it tracked.

The milestone suggests that robot hardware is reaching a scale that can support broader experimentation and deployment.

Not every robot looks like a person

When people hear “15,000 robots,” they may imagine an army of identical humanoids walking off an assembly line.

AgiBot’s portfolio is much broader.

The company produces full-size humanoids, half-size humanoids, wheeled industrial robots, quadrupeds, dexterous manipulation systems, and commercial cleaning robots. The X2 featured in my interview is a half-size humanoid designed to appear approachable and interact with people.

Peng described possible applications across education, research, manufacturing, logistics, retail, restaurants, healthcare, and outdoor inspections.

The robot’s physical form depends on the job.

A wheeled platform may make more sense on a factory floor. A quadruped may be better suited to inspection work. A smaller humanoid may be useful for entertainment, education, customer guidance, or social interaction.

This diversity is a reminder that the future of embodied AI will not be defined by a single perfect human replica. It will probably include many types of machines designed for different environments.

The difficult leap from demonstrations to deployment

Robotics demonstrations have become remarkably good at attracting attention.

Robots dance, run, sort objects, fold clothes, perform backflips, and hold conversations. These moments are entertaining and often technically impressive.

Check out this rehearsal of Agibots rehearsing for a performance at the AGI and Robotics summit at Stanford. Very entertaining.

But a demonstration is not the same as a deployment. A demonstration asks whether a robot can perform a task under favorable conditions. A deployment asks whether it can perform that task repeatedly, safely, affordably, and with minimal human intervention.

A robot working in a factory cannot succeed nine times out of ten and simply try again after a failure. It must operate according to production schedules, coordinate with people and other machines, and respond predictably when something unexpected happens. Battery life, maintenance, integration, safety, training, and the cost of human supervision all become part of the calculation.

Peng acknowledged that this is not an AgiBot challenge alone.

“The whole industry” is searching for places where robots can work autonomously and reliably, he said.

This is why constrained environments are likely to come first.

Factories, warehouses, and commercial facilities offer repeatable processes and spaces that can be adapted to machines. Homes, hospitals, and nursing facilities are far less predictable. Every room is different. Objects move. People behave unexpectedly. The consequences of an error can be much more serious.

AgiBot has reported approximately 100 cumulative hours of factory livestream operations involving G2 robots in a tablet-production quality-inspection process. The robots worked alongside people and followed the rhythm of an active production line.

That is a meaningful step beyond a short stage demonstration. It is not yet proof that general-purpose robots can operate autonomously everywhere, but it illustrates the path from controlled experiments to narrowly defined work.

People may be more ready than expected

AgiBot expected some people to be uneasy around the X2. Instead, Peng said, public response has been overwhelmingly enthusiastic.

I saw that curiosity at NVIDIA GTC, where the robot’s friendly appearance and dance moves drew a crowd.

The X2 can converse, perform, guide visitors, and give presentations. While complex physical work remains challenging, social interaction could become one of the first practical uses for smaller humanoids.

Could robots become companions?

Peng described nursing homes and elder care as potential applications. The X2 can already converse and entertain, while future robots might retrieve medication or help with daily tasks.

That would require exceptional reliability, privacy safeguards, and clear accountability.

The deeper question is how these robots are used. Replacing human attention is concerning. Supporting caregivers and giving them more time with residents could be genuinely valuable.

More robots could mean better robots

Manufacturing scale also creates data.

Unlike language models, robots must learn from physical interactions: how objects move, how environments change, and how to recover from mistakes.

AgiBot World reports more than one million robot trajectories across 217 tasks in settings modeled on homes, stores, factories, restaurants, and offices.

More robots can generate more experience, improving the models that control them. But the data must be reliable, and success in training must transfer safely to unpredictable environments.

In the physical world, intelligence also means knowing when to stop and ask for help.

What comes next?

I asked Peng what we might discuss if we met again next year.

“I think we’ll be talking about where these robots have been helping society, rather than what they can do,” he said.

The next measure of progress will not be another impressive demonstration. It will be reliable work, economic value, human trust, and safer workplaces.

AgiBot’s 15,000th robot is an important manufacturing achievement. Now the industry must show where these machines belong and whether they improve human lives.

Additional Resources for Inquisitive Minds

• AgiBot’s 15,000th robot announcement

• AgiBot’s 10,000-unit production announcement

• AgiBot product portfolio

• AgiBot World research paper

• International Federation of Robotics, World Robotics 2025 presentation

#robotics #physicalai #agibot

Just over three months later, I caught up with engineer Ricky Wu at the Humanity & AGI Summit 2026 at Stanford University, where StarBot was showing something entirely different: a humanoid robot. The transition says as much about the service robotics industry as it does about StarBot itself.

During our interview, Ricky explained that the shift was driven by customer feedback. Restaurant operators liked the original robot but kept asking the same question: could it climb stairs? The answer was no. A wheeled robot is efficient on flat floors but quickly reaches its limits in more complex environments. The company’s response was to build a humanoid platform that could navigate spaces designed for people while adding dexterous hands, depth sensing, LiDAR, and conversational AI.

It’s a reminder that some of the biggest advances in robotics aren’t driven by new AI models. They’re driven by deployment. Real customers expose real problems.

The restaurant industry has become one of the first proving grounds for service robotics, but adoption remains in its early stages. While robot waiters attract attention on social media, the vast majority of restaurants still operate without robots. Where they are deployed, robots typically transport food, bus tables, or handle repetitive delivery tasks, allowing staff to spend more time interacting with customers instead of walking dishes across the dining room.

The labor equation also matters. Restaurants continue to struggle with hiring and employee turnover, making repetitive support tasks an attractive target for automation. Rather than replacing servers, today’s service robots are generally designed to augment existing staff by taking over physically repetitive work.

At the same time, companies appear to be thinking beyond restaurants. In our discussion, StarBot described applications that extended into homes and other service environments. That reflects a broader trend I have begun noticing across multiple robotics startups. Instead of building highly specialized machines for a single market, many companies are evolving toward general-purpose robotic platforms that can be adapted through software and AI for a variety of settings.

Whether that strategy ultimately succeeds remains an open question. A robot built for restaurants must satisfy a different set of requirements than one designed for factories or private homes. But the industry increasingly seems willing to pursue that broader vision.

#robotics #physicalai #droidsnewsletter #starbotusa #AIRoA #HumanityAGIsummit #AIRoboticsAlliance

From Drone Racing to Defense

Founded in 2023 by Soren Monroe-Anderson and Olaf Hichwa, who competed as professional drone racers before entering the defense industry, Neros builds its Archer quadcopter FPV drone entirely without Chinese components. The company, headquartered in El Segundo, California, has positioned itself as a domestic alternative to the cheap Chinese-made drones that have dominated the market.

Neros was already on the Pentagon’s radar. The company was selected for the Army’s Purpose-Built Attritable System program in late 2025 and won approval on the Defense Innovation Unit’s Blue UAS list. It also signed a contract to deliver 6,000 Archer drones to Ukraine’s frontlines through the International Drone Coalition. Monroe-Anderson, now in his early twenties, has said the company can scale production to one million drones per year with sufficient Pentagon backing. Earlier this year, Neros acquired a 250,000-square-foot factory to support that ambition.

A Broader Push for Drone Manufacturing

The Neros contract reflects a wider Pentagon effort to build domestic drone production capacity. On Monday, West Virginia Governor Patrick Morrisey announced that Helsing, a German defense AI company valued at $18 billion, will invest $50 million to establish its first U.S. manufacturing facility in Martinsburg, West Virginia. The plant will produce Helsing’s HX-2 AI-enabled strike drone — already combat-proven in Ukraine — at a rate of 2,000 per month, employing 60 workers at an average salary of $125,000.

“Factories win wars, and that’s exactly what we want to do in Berkeley County,” said Jennifer McArdle, general manager of Helsing’s U.S. subsidiary.

Pentagon’s Appetite for Attritable Drones

The contracts underscore how lessons from Ukraine’s war have reshaped American military procurement. The Army has begun training soldiers with FPV drones modeled on Ukrainian tactics, and spending on expendable drone systems has accelerated sharply. Separately, the Pentagon awarded AeroVironment a $500 million counter-drone contract on July 1, illustrating both sides of the emerging drone equation — the need for cheap offensive systems and the defenses to counter them.

Additional Resources for Inquisitive Minds:

MILITARYNYI. FPV Drone Manufacturer Plans to Deliver 1 Million Drones a Year to U.S. Military. July 7, 2025.

U.S. Department of War. Contracts for July 1, 2026

#robotics #militaryrobotics #drones #droidsnewsletter #dailyroboticnews

Tendon-Driven Actuation

Instead of using rigid gearboxes or heavy harmonic drives common in other humanoids, NEO employs a proprietary tendon-drive system. This biomimetic approach uses high-torque-density electric motors that pull flexible polymer tendons, similar to human muscles. This architecture provides fluid motion and precise control while keeping the robot extremely quiet, operating at just 22 decibels.

Design and Safety Profile

The NEO robot stands 5’6” tall but weighs only 66 pounds. It has a soft, custom 3D-lattice polymer body and is clothed in a machine-washable knit suit, which acts as a protective layer and makes the robot safe and approachable for domestic use. The tendon-driven design, combined with the lightweight structure and pinch-proof joints, is intended to minimize the risk of injury if the robot falls.

Capabilities and Manufacturing

Despite its light frame, NEO can lift up to 154 pounds and carry 55 pounds. Its 22-DoF hands have a grip strength capable of deadlifting 70kg, and the hands are naturally compliant to withstand falls. 1X manufactures the hands, actuators, and polymer molding in-house at their facility in Hayward, California. The robot is powered by an NVIDIA Jetson Thor chipset and operates for up to 4 hours per charge.

“The Hand Problem”

The development of highly capable robotic hands, often referred to as “the hands problem” or “Moravec’s Paradox,” is widely considered the most difficult hardware challenge in humanoid robotics. While robots are proficient at balancing, walking, and heavy lifting, replicating the dexterity required for unstructured, everyday human tasks like handling fragile items or cracking eggs remains a massive bottleneck.

The difficulty in building humanoid hands stems from compounding factors of complexity, space, and reliability. Human hands have over 20 degrees of freedom, which requires engineers to pack dozens of miniaturized actuators, custom motors, and control systems into a very tight volume. Fully-actuated hands (with a motor for every joint) are extremely heavy, while under-actuated hands (using fewer motors with mechanical coupling) sacrifice precision.

Why the Hands are So Significant

The hands on 1X’s NEO robot are one of its key differentiators, designed to mimic human dexterity while being robust enough for household tasks. They represent a shift toward prioritizing active control and durability in consumer-ready physical AI.

Additional Resources for Inquisitive Minds:

WIRED. The 1X Neo Robot Has Freaky Fast Fingers. July 9, 2026.

Forbes. Human-Level Hands? 1X Just Gave Humanoid Robot Neo Something Close. July 9, 2026.

1x Blog. NEO’s Hands. An API to the Physical World. July 9, 2026.

Editor’s Note: The podcast attached to this daily update was created with Google NotebookLM’s audio overview. The sources for this “notebook” are listed above. The AI hosts were very creative in how they said “Neo,” (correctly said as “nee-oh.'“)

#robotics #thehandproblem #1x #1xneo

Military reports are rarely exciting.

But one quiet change inside Ukraine’s daily battlefield updates says a lot about where robotics is headed.

For the first time, ground robotic systems now appear as their own category in Ukraine’s official tally of Russian equipment losses. That puts unmanned ground vehicles (UGVs) in the same daily accounting as tanks, artillery, trucks, and drones.

From Experiment to Inventory

On June 3, Ukraine’s General Staff reported Russia lost:

• 5 tanks

• 60 artillery systems

• 1,853 drones

• 437 vehicles

• 6 ground robotic systems

The report, carried by ArmyInform and summarized by Euromaidan Press, estimated roughly $435 million in Russian equipment losses over the previous 24 hours.

Battlefield Robots Are Now Outpacing Tanks—At Least in the Daily Ledger

On June 30, Ukraine’s General Staff reported Russian losses that included 13 ground robotic systems but only one tank during the previous 24 hours.

The numbers themselves fluctuate from day to day. The more important trend is that ground robotic systems now appear consistently as their own category in Ukraine’s official battlefield accounting, alongside tanks, artillery, drones, and other equipment.

Why This Matters

According to General Staff spokesperson Major Dmytro Lykhovii, Ukraine only began publicly releasing statistics on destroyed Russian ground robotic systems in early May, even though the military had been tracking them internally before then. At the time, officials said Ukrainian forces had already destroyed 1,294 Russian ground robotic systems since Russia began deploying them, with the new reporting category appearing in the official daily bulletin starting May 3.

The Bigger Signal

Technology becomes mainstream in war when armies stop talking about prototypes and start counting replacements.

Separate reporting implies ground robots are no longer viewed as unusual experimental hardware but as equipment deployed frequently enough that commanders—and analysts—want to measure their attrition independently.

That’s the same path drones followed several years ago.

Today’s battlefield ledger often becomes tomorrow’s procurement plan.

Robotics takeaway: When militaries create a new column in their daily battle ledger, they’re acknowledging that a technology has become operationally significant. That’s a small accounting change—but an important signal that unmanned ground vehicles are moving beyond experimental deployments into routine wartime logistics and planning.

Additional Resources for Inquisitive Minds

• Mukhina, Olena. “Russia lost $435 million worth of equipment over past 24 hours. Ground robots are among losses, too”. Euromaidan Press, 3 June 2026.

• “Ukraine just added a new line to its Russian-losses ledger: ground robots; the count starts at 12.” Euromaidan Press, 2 May 2026.

• Ukrainska Pravda. Ukrainian forces have destroyed 1,294 Russian ground robotic systems. 2 May 2026.

• babel.ua. General Staff: Russians lost over a thousand soldiers and almost 2 thousand drones over the past day. June 29.
  
  If you happen to read Ukranian, here is the official chart as of June 30th: (so, the +1 means additional losses, as in Russia lost one tank in the past day.) The number 1952 for tactical UAVs covers all classes of tactical drones.
  
  

#militaryrobotics #robotics #battlefieldrobots

A Tech Marvel on the Turf

Beyond the charming performance, getting Atlas to function in a stadium full of 80,000 cellphones was a massive engineering feat. Because standard Wi-Fi was completely jammed by the crowd, Boston Dynamics engineers had to use a custom radio link to maintain control of the robot. Atlas’s performance relied on “whole-body control technology,” which ensures that all of the robot’s joints act as a unified system, allowing it to maintain its balance and move fluidly on the grass. It also boasts 56 degrees of freedom and the ability to generalize tasks beyond its initial programming.

Reinforcement Learning Meets Goal Celebrations

Before handing over the ball, Atlas delighted the crowd by running through a routine of iconic soccer celebrations. The robot perfectly recreated the signature moves of players like Erling Haaland, Harry Kane, Matheus Cunha, and Son Heung-min. This was achieved through weeks of reinforcement learning, where engineers fed the robot footage of World Cup matches and used large-scale simulations to teach it to mimic kicks, drills, and player movements.gigazineyoutube

From the Stadium to the Factory

While the World Cup halftime show was a massive hit with audiences, Hyundai, the parent company of Boston Dynamics, has much bigger plans for Atlas. The real goal of these live demonstrations is to prove the robot’s reliability in chaotic, real-world environments. Hyundai intends to deploy mass-produced versions of these humanoids in its U.S. manufacturing facilities starting in 2028, where they will take over repetitive and high-risk tasks on the assembly line.

I mentioned this to my son today over lunch in Santa Clara. He stopped, fork halfway to his mouth: “How would you even measure such a thing? What even is consciousness?”

Ah, and he asked the right questions. And, this is the reason why major tech companies are hiring philosophers.

From Fringe to Boardroom

Google DeepMind, Anthropic, and Meta have hired experts in psychology, ethics, and philosophy in recent months as they expand research into machine consciousness and AI welfare, the Financial Times reported in June.

The Washington Post reported on July 1 that top technology companies including Google and Anthropic are now “openly exploring whether AI can be conscious,” nearly four years after a Google engineer was fired for claiming the company’s chatbot was sentient.

Anthropic has been the most public in its efforts. The company launched a formal “model welfare” research program in April 2025, led by researcher Kyle Fish, who has estimated a 15 percent probability that Claude is conscious today. Anthropic’s system card for Claude Opus 4.6 includes a dedicated “Model Welfare Assessment” section, tracking metrics such as positive and negative affect, self-image, and emotional stability across thousands of test conversations.

“Things That Are Mysterious, Even Unsettling”

In a speech at the Vatican on May 25, 2026, during the launch of Pope Leo XIV’s encyclical on AI, Anthropic co-founder Chris Olah offered some of the most striking remarks yet from an industry insider. “We keep finding things that are mysterious, even unsettling,” Olah said. “We find structures that mirror results from human neuroscience. We find evidence of introspection. We find internal states that functionally mirror joy, satisfaction, fear, grief, and unease. I don’t know what that means, but I think it warrants ongoing discernment.”

According to research Anthropic published in April, Claude contains 171 distinct “emotion concepts” in its neural network — patterns representing joy, grief, fear, and calm that were not programmed but emerged from training on human text.

OpenAI has taken a different tack. Joanne Jang, who formerly led OpenAI Labs, framed the issue as one of “perceived consciousness” — how conscious the model seems to users — rather than a scientific claim about actual inner experience.

Scientists Urge Caution

Neuroscientists remain broadly skeptical. A June 2026 study concluded that “no existing AI system (including ChatGPT) is conscious,” arguing that the appearance of consciousness in large language models “is not achieved in a way that is sufficiently similar to us to warrant attribution of conscious states”. Neuroscientist Anil Seth, in an April 2026 TED talk, argued that “we see consciousness in AI the same way we see faces in clouds”.

The Washington Post noted that brain experts are “generally skeptical that today’s AI models are or could soon be conscious,” though the question may become “messy and divisive” as public opinion is shaped more by direct experience with chatbots than by scientific evidence.

Additional Resources for Inquisitive Minds:

The Washington Post. He warned AI could be conscious. Here’s what happened next. Four years after firing Blake Lemoine, Google and other tech companies are publicly embracing the search for AI consciousness. July 1, 2026.

TechCrunch. Anthropic is launching a new program to study AI ‘model welfare’. April 24, 2025.

Anthropic News. Anthropic co-founder Chris Olah’s remarks on Pope Leo XIV’s encyclical “Magnifica humanitas”. May 25, 2026.

ScienceDaily. Scientists are seriously asking if bees and ChatGPT are conscious. June 5, 2026.

#aiethics #aimodels #claude #googledeepmind

Stephen Witt recently wrote an interesting piece in The New Yorker focusing on Neo, the home humanoid from 1X Technologies. He asks an important question: Are humanoid robots actually ready to leave the lab? Along the way, he explains why many experts remain cautious. While Neo can autonomously perform relatively simple tasks such as opening doors, retrieving objects and turning off lights, more complex household chores still rely heavily on human teleoperators working remotely.

Witt isn’t the first to report on Neo’s limitations. Last fall, Joanna Stern from the Wall Street Journal observed that virtually everything she watched Neo accomplish was guided by a skilled human operator.

1X presents this as a feature, not a flaw. The company openly embraces teleoperation as part of Neo’s training strategy. Human experts can remotely step into the robot, teach it new skills and generate the real-world data needed to make future versions more autonomous.

That approach can work, of course, provided early adopters are willing to purchase a product that is still learning. If you take possession of a 2026 Neo, you are becoming part of the robot’s training process. For some, that might be thrilling. But, for someone who has bought into the hype of a “robot butler,” they might be very disappointed.

Reading Witt’s article got me wondering where the rest of the humanoid robotics industry actually stands.

If Neo is one of the most advanced home robots in development today, where are Figure, Tesla, Agility, Boston Dynamics, Unitree, Apptronik and the others?

Not All “Deployments” Are Equal

The word deployment has become one of the most overused terms in robotics.

A robot performing a choreographed demonstration is deployed in a very different sense than one working an eight-hour shift for a paying customer.

Today, humanoid robots generally fall into three categories:

Rather than asking whether humanoids are “ready,” a better question is ready for what? A robot designed to impress investors has very different requirements than one expected to work beside people every day. As the industry matures, success will be measured less by viral demonstrations and more by reliability, autonomy and economic value.

Factories Are Leading the Way

The strongest evidence of commercial progress is coming from structured industrial environments.

Figure AI has emerged as one of the industry’s leading examples. At BMW’s Spartanburg plant, Figure robots have accumulated more than 1,250 operational hours while assisting vehicle production.

The company’s latest Figure 03 platform also completed a livestream exceeding 100 continuous hours while sorting more than 140,000 packages. This demonstration proved very effective to show reliability.

Digit from Agility Robotics is working with customers including Amazon, GXO and Toyota Motor Manufacturing Canada. Rather than replacing entire assembly lines, Digit performs repetitive material-handling tasks intended to reduce ergonomic injuries and free workers from physically demanding jobs.

Boston Dynamics crossed an important milestone this year when its all-electric Atlas humanoid entered Hyundai’s Metaplant Application Center in Georgia. Atlas can autonomously navigate the facility, recharge itself and return to work, marking the company’s transition from decades of robotics research toward commercial deployment.

These robots aren’t replacing entire workforces.

They are replacing specific tasks.

Tesla’s Different Strategy

Tesla’s Optimus remains the industry’s highest-profile humanoid robot, but its commercialization strategy differs from many competitors.

The company is investing billions of dollars to build manufacturing capacity that could eventually produce robots at enormous scale. Yet Elon Musk has acknowledged that the Optimus robots currently operating inside Tesla factories are still part of an ongoing research and data collection effort rather than performing economically valuable factory work.

Today, Optimus remains largely an internal research platform rather than a commercial product. Tesla is using deployments inside its own factories to collect the data needed to improve autonomy.

Figure has taken a different approach. While continuing to improve autonomy, it has simultaneously ramped production, deployed robots at BMW, and begun generating commercial operating experience in customer environments.

Both companies are still training their robots. Figure is simply doing more of that training in the field with commercial customers.

The industry’s biggest challenge is no longer building humanoids.

It’s making them reliable enough to perform useful work day after day in the real world.

Why Homes Are So Much Harder

The industry’s biggest challenge is no longer building humanoids.

It’s collecting enough high-quality real-world data to make them reliable.

There is no robotics equivalent of the public internet.

That helps explain why factories are advancing faster than homes. Industrial environments are engineered for consistency. Lighting is predictable. Workstations are standardized. Parts arrive in known locations. Robots repeat the same workflows thousands of times under carefully controlled conditions.

Homes offer none of those advantages.

Furniture gets rearranged. Children leave toys on the floor. Pets wander into rooms. Or, in my house, the pet wanders into the room and leaves toys everywhere.

Frankly, I’ve always suspected my cattle dog would be the ultimate benchmark for Physical AI. Heelers are relentless about herding anything that moves. How would Neo respond when a 40-pound dog suddenly decided the robot belonged in the kitchen?

That’s why teleoperation remains such an important part of 1X’s strategy. Every remotely assisted task becomes additional training data, gradually expanding the robot’s ability to operate independently. Although, I’m not sure a remote teleoperator is a match for Rosie.

The Race for Real-World Experience

If there’s one takeaway from all of this, it’s that the question has changed.

We no longer need to ask whether humanoid robots are possible.

They are.

The question now is which companies can transform millions of hours of real-world experience into reliable autonomy the fastest.

In robotics, experience is becoming the ultimate competitive advantage.

Additional Reading for Inquisitive Minds:

• Abdullahi, Aminu. “1X’s $20K Robot Targets US Homes in 2026, Aims to Reduce the ‘Creepy’ Factor.” eWeek, April 27, 2026.

• Baldauf, Ulrich. “5 Humanoid Robot Challenges Blocking Production in 2026.” There’s A Robot For That, June 8, 2026.

• Bank of America Institute. Physical AI, Part 2: Humanoid Robots. March 12, 2026.

• “Domestic Humanoids: The Dawn of Physical AI.” 1X Technologies, June 29, 2026.

• “Humanoid Robot Deployment Report: Latest Real-World Milestones (June 2026).” Humanoid Applications, June 21, 2026.

• “Industry Insights: Humanoids and Physical AI at Automate 2026.” Automate.org.

• New Market Pitch Team. “Tesla Optimus Deployment Tracker (2026).” April 22, 2026.

• Shannon, Matt. “Robots Are Ready. But Where Do You Start?” Field Notes from the Future, Deloitte Consulting LLP, July 1, 2026.

• Singh, Ranveer, et al. “Predicting AI Maturity and Commercial Readiness of Humanoid Robotic Systems: A Machine Learning Analysis of The Global Humanoid Robotics Intelligence Dataset 2026.” IJIRT, Vol. 13, No. 1, June 2026.

• Torka, Benedikt. “BMW Group Advances the Use of Physical AI in Production with Figure 03 Project in Spartanburg.” BMW Group PressClub Global, June 25, 2026.

• Witt, Stephen. “Are Humanoid Robots Ready to Be Deployed?” The New Yorker, July 6 & 13, 2026.

• Figure03 blog. Ramping Figure 03 production. April 29, 2026.

  ---

Editor’s Note: Today’s podcast was created using Google’s NotebookLM technology, using the sources above as the research basis for the “notebook.”

#robotics #humanoids #physicalai

UBTech founder Zhou Jian announced that the U1 lineup starts at 119,800 yuan (approximately $17,000) for the U1 Lite, an upper-body-only version, while the standard U1 Pro is priced at 169,800 yuan (roughly $24,000). The flagship U1 Ultra commands 880,000 yuan for the female model and 990,000 yuan (about $140,000) for the male version.

The event, themed "Love is eternal," confirmed that deliveries will begin in mid-September 2026.

The robots are available in male (183 cm, 42 kg) and female (168 cm, 35.2 kg) configurations, each featuring 88 degrees of freedom for human-like movement. Pre-orders opened on JD.com on June 2 with a 3,000-yuan deposit, and the company reported surpassing 10,000 reservations by the time of the launch event — ten times its total 2025 sales volume, according to Zhou.

Companionship Robots

UBTech is marketing the U1 line exclusively to adults for emotional companionship rather than household chores. The system operates on a Rockchip RK3588 processor and utilizes an AI model trained on the Huawei Ascend framework. To address privacy concerns, the company stores all personal data locally rather than on external servers. Journalists at the event observed highly lifelike visual designs but noted the facial expressions and conversational capabilities require further refinement.

Read the original Post from UBTech on X:

This sounds so familiar. I went to a public open house event at a high school two weeks ago here in Gilroy, California. Except the hosts were AWS and they were here to reassure about the tech giant’s planned data center campus.

“Trust us,” they said. “Power? It won’t go up. Water? We won’t use it all. Noise? You won’t hear it over the sounds of the highway noise.”

So, now it is Virginia’s turn and they are getting the song and dance from Google.

Supporters and Opponents Clash Outside

The event, held from 5 to 7 p.m. in the school’s lower gymnasium, drew both supporters and protesters. Opponents of the project gathered outside the school before the doors opened, continuing a campaign of community opposition that has intensified since details about the project’s water usage became public earlier this year.

Inside, Google set up information stations covering water usage, power infrastructure, sound and environmental considerations, community benefits, local revenue impacts, and project planning and zoning. Attendees could drop in at any time, explore the stations, and ask questions.

Water Concerns Take Center Stage

Water usage has been the most contentious issue surrounding the project. The data center campus, planned for a 312-acre site at the Greenfield Industrial Park, could draw between 2 million and 8 million gallons of water per day from Carvins Cove reservoir, according to an agreement between Botetourt County and the Western Virginia Water Authority. Google has pledged to minimize consumption, support water replenishment efforts, and finance long-range water supply studies.

The county’s website states that recreational access to Carvins Cove — used for fishing, kayaking, biking, and hiking — will be maintained. Google has also committed to supporting research at Virginia Tech to study watershed health and improvements to the regional water supply.

Jobs and Economic Impact

The project is expected to create up to 1,000 temporary construction jobs at peak and a minimum of 150 permanent positions across three planned data centers, with Google committing to no fewer than 50 high-paying jobs per building. The county estimates the campus will generate roughly $10 million per year in local tax revenue once operational.

Google purchased the land for $14.06 million in June 2025 and pledged $4 million over five years for community projects. Site grading is expected to begin in 2026, with each data center taking 18 to 24 months to construct. The company said project information would be made available online for residents who could not attend Wednesday’s event.

Read more: Google talks timeline while residents protest.

It misunderstands us three times before making the seats nice and toasty. It is about as frustrating as conversing with early stage Alexa.

Now, after owning the car for some time, I learned the short list of line commands the car can understand. Think of training your dog and only using those commands. Except the dog is your car. “Sentry Mode On.” “Wipers On.” “Heat down two degrees.” “Podcast DROIDS Newsletter.” All of it is incredibly useful for keeping your eyes on the road. None of it feels like natural language.

Grok, on the other hand, is happy to chat with you like a friend in a bar. You can just go on and on about any number of subjects. Before NVIDIA GTC, I asked Grok to help me get the “ummms” out of my speech when I was improvising questions. We practiced this skill again and again during long drives.

So, there is a definite difference between Grok and the regular Tesla voice controls.

Elon Musk confirmed on X this week that Tesla drivers will be able to issue natural-language commands to the company’s Full Self-Driving system through its Grok AI assistant within roughly three months, placing the expected rollout around September 2026.

Talking to Your Car Like an Uber Driver

Responding on Wednesday to an X user who asked to be able to “converse” with Grok “like we can with an Uber driver,” Musk wrote that “this functionality will be there in about 3 months or so.” The proposed capability would allow Grok to pass high-level driver intent directly into FSD’s planning layer, enabling instructions such as “turn right at the next block” or “drop us off here” to alter the vehicle’s real-time behavior. It would mark the first time voice input has had a direct influence on FSD’s driving and parking decisions.

In the same thread, Musk said upcoming FSD releases “will remember your parking preferences, so that the car goes to the right location at your home, office, school drop off, etc.” He also acknowledged that while Tesla would eventually have “fully integrated voice control commands that can control the vehicle,” it is “just a bit too early for that.”

(The car already remembers that my husband likes to go surfings in the early mornings, so if he gets in the car at that hour it sets the map to the beach. It knows that I head to the gym at mid-day and sets the course accordingly. Creepy, but useful.)

Grok’s Current Role — and What Changes

Grok currently functions as a conversational assistant inside Tesla vehicles, handling navigation queries and responding to hands-free “Hey Grok” activation added in a Spring 2026 update. It cannot yet control core vehicle functions or influence how FSD drives. The deeper integration Musk described would bridge that gap, letting the AI interpret commands and feed them into FSD’s autonomous planning system.

How to use Grok to enable navigation commands.

The announcement also revived discussion of the long-promised “Banish” feature — essentially a reverse Summon that would allow the car to self-park after dropping off passengers. Musk first teased Banish in early 2024 and has periodically reaffirmed it since, though the feature has yet to ship.

A Timeline Worth Salting

Musk first announced the Grok-FSD integration in February, and this week’s post provides the most concrete launch window to date. However, as Electrek noted, Musk “gave no date at all for parking preferences” and has a well-documented history of missing self-imposed deadlines. Multiple Tesla community observers cautioned that the three-month estimate should be taken with skepticism given past delays.

I believe our early conversations with the voice system are in this video: our first 14 days with the Tesla.

#teslafsd #grok

The event, held at the company’s Los Angeles headquarters, also introduced the FF Master Mini and FF Nova robots, and unveiled a “Three-in-One” EAI Robotics Education Ecosystem Strategy alongside an open-source developer platform.

New Products Target Education and Consumer Markets

The FX Navi is positioned as an entry-level quadruped robot designed for families and education, paired with what the company calls the first EAI STEM curriculum in the United States. The company has opened its 3D model files so users can design and 3D-print custom heads for the robot, and plans to deliver ongoing capabilities through over-the-air updates including autonomous following and multimodal perception.

The All-New Futurist, described as the 2027 model year, is billed as the first humanoid robot in the United States to natively support NVIDIA Sonic’s full-body motion control system. An upcoming Ultra version will be equipped with NVIDIA’s Jetson Thor chip for enhanced computing performance. The original Futurist, launched earlier this year at $34,990, featured 40 degrees of freedom and three hours of runtime on the Jetson Orin platform.

From Electric Vehicles to Robots

The California-based company, which has struggled with its electric vehicle business, pivoted into embodied AI robotics earlier this year, launching its FF EAI-Robotics subsidiary and beginning deliveries of its first robots in February. In May, Faraday Future secured $25 million in funding with a target of shipping 1,500 robots by year’s end. The company also recently signed a strategic cooperation agreement with Lynwood Unified School District in Los Angeles for K-12 robotics education programs.

What Comes Next

Faraday Future plans additional product reveals at Automate, North America’s largest robotics exhibition, running June 22-25 in Chicago. The company will showcase its multi-form robotics lineup at the event, where it is among the exhibitors in the NVIDIA Humanoid Robot Pavilion.

#robotics #faraday

How TEPI Works

The intervention, called therapist-exoskeleton-patient interaction (TEPI), has both a therapist and a stroke survivor wear a lower-limb exoskeleton virtually connected at the hips and knees. The virtual connection behaves like a combination of springs and shock absorbers, allowing the two to influence each other’s movements in real time.

In conventional therapy, therapists can physically assist only a limited number of movements at once, often focusing on a single aspect of gait. More complex whole-body training can require multiple therapists. Existing rehabilitation exoskeletons, meanwhile, often rely on fixed movement patterns that do not fully adapt to a patient’s performance.

“By combining the hands-on adaptability of physical therapy with the scalability and precision of robotic systems, it can enable more comprehensive, whole-body gait training without requiring multiple therapists,” said Lorenzo Vianello, a postdoctoral researcher at Shirley Ryan AbilityLab and co-first author of the paper.

Results and What’s Next

In evaluations with eight stroke survivors, TEPI outperformed conventional therapist-guided treadmill training on several measures. Participants demonstrated greater joint range of motion, took longer and higher steps, maintained similar muscle activation, and reported high levels of motivation and enjoyment.

“Therapist-led rehabilitation remains the foundation of recovery for many patients, and this research shows promise for complementing this standard of care,” said José L. Pons, who led the research program. Pons is a scientific chair at Shirley Ryan AbilityLab and professor at Northwestern University Feinberg School of Medicine.

Nearly 800,000 Americans survive a stroke each year, according to the Centers for Disease Control and Prevention, and many face months of intensive rehabilitation to regain the ability to walk. The researchers plan next to explore applying the framework to overground walking, stair climbing and sit-to-stand transitions, as well as more accessible systems that could extend therapist-guided rehabilitation into patients’ homes.

Additional Resources for Inquisitive Minds:

New exoskeleton therapy could redefine how stroke survivors relearn to walk. Northwestern Now. June 17, 2026.

Read the study:

Therapist-exoskeleton-patient interaction for gait therapy Science Robotics. June 17, 2026.

Editor’s Note: The podcast attached to this article was created using Google’s NotebookLM technology and the resources in this article. Ironically, despite all the technology, the AI hosts still can’t pronounce “droids.”

#robotics

A Deepening Partnership

The two companies signed a memorandum of understanding at Eurosatory 2026 to advance their joint work on counter-unmanned aerial vehicle systems. According to dev.ua, the Ukrainian outlet that first reported the agreement, the partnership will focus on creating robotic complexes designed to neutralize drone threats. Frontline Robotics revealed a counter-UAV shotgun system integrated onto an unmanned ground vehicle at the show, according to Janes.

The cooperation will combine Frontline Robotics’ Buria remote weapon station with Milrem’s Logist unmanned ground vehicle, while also advancing joint work on AI-powered threat detection, Frontline Robotics said in a LinkedIn post. The two firms have a history of collaboration — in 2025, Frontline Robotics integrated its Burya turret onto Milrem’s THeMIS platform.

Parallel Deal with Dropla on AI Threat Detection

Separately, Frontline Robotics announced a collaboration with Dropla, a Danish-Ukrainian company, to jointly develop AI solutions for unmanned ground vehicles capable of detecting threats in real time. Dev.ua reported that the joint work will combine Ukrainian engineering expertise with Danish industrial capabilities to create a new class of ground combat platforms.

Broader Context

The announcements come amid a rapid expansion of Ukraine’s ground robotics program. Ukraine’s Defense Ministry contracted 25,000 unmanned ground vehicles in the first half of 2026, more than double its 2025 total, as Kyiv moves to shift frontline logistics away from soldiers, according to Defense News. Ukrainian unmanned systems have carried out over 22,000 missions in 2026.

Milrem Robotics, meanwhile, has been on a signing spree at Eurosatory 2026, also inking deals with Frankenburg Technologies on mobile counter-UAS missile defense and with France’s CNIM Systèmes Industriels on unmanned engineering reconnaissance. The Estonian firm recently opened a new assembly line in the Netherlands to boost production capacity, and has pitched a broader concept for robotic defense zones along NATO’s eastern border from Finland to Poland.

#militaryrobotics #drones #robotics

Infinite Campus Breach Exposed 137,000 Staff Accounts

The announcements came as data breach notification service Have I Been Pwned confirmed that ShinyHunters had exposed 137,100 school staff accounts stolen from Infinite Campus, a K-12 student information system used by more than 3,200 school districts managing data for roughly 11 million students across 46 states. The attackers gained access through a compromised Salesforce environment in March, stealing names, email addresses, phone numbers, physical addresses, job titles, usernames, and support tickets. Infinite Campus refused to pay the extortion demand, after which ShinyHunters published the stolen data.

The group’s broader education campaign began in late April when it compromised Instructure’s Canvas learning management system, claiming exfiltration of 3.65 terabytes of data across 275 million records from more than 8,800 educational institutions. The FBI issued a public service announcement in May warning of “potential future impacts” from ShinyHunters’ operations against education platforms. More recently, cybersecurity firm Mandiant, part of Alphabet, identified an active ShinyHunters compromise involving an Oracle PeopleSoft exploit targeting the sector

Separate Attack Hits Singapore School Network

In a parallel incident, the ransomware group FulcrumSec claimed responsibility on June 10 for a cyberattack on the Global Schools Foundation, an international education network headquartered in Singapore. According to cybersecurity firm Resecurity, the stolen data includes 33,088 passport numbers belonging to children and parents across 66 nationalities, 143,494 employee salary records, 221 million student attendance records, and 107,603 transport user addresses with GPS coordinates. FulcrumSec threatened to publish the data after the foundation declined to pay a ransom that the group described as “less than the cost of a single year’s tuition for a classroom of students”.

A Sector Under Siege

The education sector’s vulnerability stems from rapid cloud adoption during and after the pandemic, combined with limited cybersecurity resources. Schools store troves of sensitive data — student records, financial information, and employee details — yet often lack defenses comparable to those in finance or healthcare. The FBI has urged affected individuals not to respond to extortion demands and to verify any breach-related communications through known institutional channels.

Additional Resources for Inquisitive Minds:

FBI Public Service Announcement. Alert Number: I-051526-PSA | 15 May 2026. ShinyHunters: Cyber Criminal Group Attacks Learning Management System.

ShinyHunters publishes data stolen from school software provider Infinite Campus, 137K users exposed. Cybernews. June 16, 2026.

#ethicalai #cybersecurity

That distinction matters. Beating Uber or Lyft is not the same thing as convincing millions of people to give up their own vehicles.

The biggest issue is economics. A privately owned car already has a clear cost structure, and for many people that cost is more acceptable than the hassle of waiting for a ride. Robotaxis would need to be cheap enough to tempt people out of ownership, but still expensive enough to cover fleet operations, vehicle wear, cleaning, dispatch, insurance, and the miles driven without passengers. That is a narrow lane to drive through.

Convenience is the other obstacle. A private car is always there. It is your storage space, your backup plan, your weather shield, and your immediate escape hatch. A robotaxi may save money in some situations, but it still requires ordering, waiting, and depending on a system that can fail, surge, or be unavailable when you need it most.

Then there is the hidden inefficiency of fleet service: empty miles. Robotaxis do not only move passengers. They also reposition themselves, travel to pickups, and circulate between jobs. Those miles add cost without generating revenue, which makes the business harder to scale profitably and harder to price aggressively.

So the likely outcome is not that robotaxis fail altogether. More likely, they succeed in a narrower market. They can replace some taxi trips, some ride-hailing trips, and maybe some urban car use. But replacing the private car for mainstream consumers is a much higher bar.

That is the real takeaway. Robotaxis may be a real product, but the leap from “useful service” to “mass replacement for car ownership” is still far from proven.

#robotaxis #physicalAI

Additional Resources for Inquisitive Minds:

The shift, first reported by Euractiv on June 12, was confirmed during an official information session held by EuroHPC JU on Tuesday, June 16, where the body presented specifications and timelines for the formal call for tenders expected this summer.

A Downsized Ambition

The new structure marks a departure from the EU’s February 2025 announcement, when Commission President Ursula von der Leyen pledged €20 billion to build up to five AI gigafactories, each equipped with approximately 100,000 advanced AI chips. The original plan attracted 76 expressions of interest from companies across 16 member states.

But the initiative has been beset by delays and funding constraints. Bidding was pushed from its original late-2025 target to early 2026, then again to summer 2026. According to Bloomberg, the number of interested parties has narrowed from about 70 companies to roughly 10 expected bidders, with at least two consortia reconsidering participation if the project is downsized.

The EU’s direct contribution amounts to roughly €4.1 billion in subsidies, with an equal sum expected from host member states and 65 percent of costs to come from private investors. Only two of the planned centers can receive funding before the EU’s next budget cycle begins in 2028.

Sovereignty Concerns Drive the Push

The gigafactory program remains central to the EU’s broader push for technological sovereignty, which was underscored by the European Technological Sovereignty Package unveiled on June 3, bundling a Chips Act 2.0 and a Cloud and AI Development Act. Nokia CEO Pekka Lundmark warned in April that Europe lacks the infrastructure for AI data centers and is not investing enough to prevent business from migrating to the United States and China.

A French consortium led by Scaleway, dubbed AION, has bid roughly $10 billion to build one facility with GPU capacity equivalent to more than 288,000 Nvidia H100 chips. Construction of the first gigafactory is targeted for 2027.

Polish Digital Minister Dariusz Standerski, who confirmed the timeline, noted that funding will be disbursed in two phases, in 2028 and 2030 — a schedule that critics say delivers subsidies years after the infrastructure is needed.

Read more: The EU’s €20 billion plan for five AI gigafactories is falling apart before the first bid is even submitted

In a new collaboration announced this week, ABB is integrating PSYONIC’s Ability Hand prosthetic with its GoFa collaborative robots. The goal is to capture touch, grip, and motion data from everyday prosthetic use and feed that directly into robotic manipulation models. Because the same hand can be mounted on a human user and on a robot arm, the partnership effectively turns prosthetic wearers into high‑fidelity data collectors for factory automation.

ABB frames this as a “human‑generated data” approach to physical AI. Instead of relying only on synthetic data or carefully scripted robot trials, the system learns from how people actually grasp, adjust, and re‑grasp objects in messy real environments. PSYONIC’s Ability Hand is marketed as one of the fastest dexterous prosthetic hands on the market and the first to provide touch feedback, which makes the sensor stream even richer for training robotic systems.

If this works at scale, you get a powerful flywheel. Every interaction a prosthetic user has with a coffee cup, a grocery bag, or a car part becomes training signal for robots that will eventually handle delicate, variable items in automotive, logistics, packaging, and life sciences facilities. The GoFa cobot supplies the precision and repeatability; the prosthetic community supplies real‑world grasping intelligence.

For industrial robotics, the implication is clear. The line between “assistive tech” and “factory automation” is starting to blur. The next generation of robot hands may learn what to do not from CAD models and lab demos, but from years of lived experience encoded in prosthetic touch.

PSYONIC interview filmed at NVIDIA GTC

Read the press release.

Fun fact: PSYONIC appeared on Shark Tank and secured a $1 million investment from Sharks Lori Greiner, Daymond John, and Kevin O’Leary on ABC’s Shark Tank (Season 15.) You can check out that episode here:

#robotics #psyonic #abilityhand #abbrobotics #industrialrobotics #roboticsnews

How the Study Worked

The team reanalyzed 376 de-identified pediatric cases that had already undergone genetic testing and expert review without yielding a diagnosis. Using the o3 Deep Research model to synthesize genetic data, clinical presentations, and medical literature, clinicians were able to identify diagnoses spanning neurodevelopmental disorders, rare neuromuscular disease, sudden unexpected death in pediatrics, and early-onset psychosis.

The tool is designed to assist clinicians and researchers in navigating complex medical information rather than serve as a direct consumer diagnostic product. John Brownstein, Chief Innovation Officer at Boston Children’s Hospital, has described the underlying challenge as one of cognitive limits rather than effort.

“We combine genetic information, phenotypic information, literature search, and the reasoning of AI to deliver diagnoses to families that were once left without any answers,” Brownstein said.

A Broader Push Into AI-Assisted Medicine

The NEJM AI publication represents the peer-reviewed validation of a broader initiative at Boston Children’s. The hospital announced in late May that its AI efforts had resulted in more than 40 rare disease diagnoses to date that were previously considered unsolvable. The collaboration with OpenAI, which committed $50 million in support to the hospital’s efforts, has been underway since early 2025.

Boston Children’s has embedded AI across its operations, with more than one-third of employees now using AI tools in daily work. The hospital reports approximately 60,000 hours saved across AI-enabled workflows, equivalent to more than $7 million in redeployed labor.

What It Means for Rare Disease Patients

Rare diseases collectively affect an estimated 300 million people worldwide, yet individual conditions are often so uncommon that even specialists at leading institutions struggle to reach diagnoses. Many families endure years-long “diagnostic odysseys” before receiving answers.

“This was unthinkable before, but is now providing hope to so many families,” Brownstein said.

The researchers emphasized that the AI functions as a tool to augment human expertise, not replace it — each diagnosis still required clinician verification and interpretation.

Additional Information for Inquisitive Minds:

Read the statement from OpenAI.

Read the study abstract at NEJM AI

#aimedicine #openai #airesearch