Aadeel Akhtar's PSYONIC turns a bionic hand into robotics data infrastructure

Why it matters

PSYONIC's move shows how physical AI may be built from deployed hardware and real human interaction data, not only larger robot models or simulation.

Dr. Aadeel Akhtar is taking PSYONIC beyond prosthetics by using its Ability Hand as a source of human dexterity data for robots, with NVIDIA (@nvidia) and ABB Robotics now attached to the effort.

The news, reported Monday by Forbes, is not a conventional AI launch. PSYONIC is making a more specific bet: that a prosthetic hand worn by people can capture the kind of force, contact, finger-position and correction data that robot-learning systems still struggle to get from cameras, simulation or teleoperation rigs alone.

That bet fits Akhtar's path. Akhtar, PSYONIC's founder and CEO, studied biology and computer science at Loyola University Chicago before earning an MS in electrical and computer engineering and a PhD in neuroscience from the University of Illinois Urbana-Champaign. Illinois has traced his original motivation to a childhood trip to Pakistan, where he met a child with a limb difference; Akhtar's own site says the company was born after a 2014 trip to Ecuador, where he helped fit Juan Suquillo with a 3D-printed bionic hand prototype and saw him regain a pinch motion after decades without one.

The company Akhtar built out of that work is now trying to make its prosthetics installed base do double duty. The Ability Hand is still sold as a bionic hand for people, with pressure sensors, vibration feedback, five articulating fingers, a rotating thumb, USB-C charging, IP64 splash resistance, a 490-gram weight and 32 grip patterns. Forbes reported that the device is worn by more than 300 patients.

From assistive device to robot teacher

PSYONIC's robotics pitch rests on a hardware loop most robot-hand companies do not have: the same hand can be worn by humans and mounted on robots. In March, PSYONIC announced that the Ability Hand had become a native asset in NVIDIA Isaac Lab, NVIDIA's framework for robot learning, and said it was collaborating with NVIDIA on what it calls "real-to-real" transfer.

The phrase matters because it is a direct answer to a problem in physical AI. Simulation can produce scale, but simulated contact is hard. Video can capture human motion, but it often misses the pressure and small adjustments that determine whether a robot grasps a soft, irregular or fragile object. Teleoperation can generate demonstrations, but the equipment is expensive and the resulting data can be constrained by the control rig.

PSYONIC's claim is that a human using the Ability Hand in ordinary manipulation tasks can generate physically grounded data from a sensorized hand, then let robot developers train and validate policies against the same hand when it is attached to an industrial arm, mobile platform or humanoid. In its GTC announcement, the company framed this as turning prosthetic use into a data collection layer for robot manipulation via real-to-real transfer.

ABB gives the idea an industrial testbed

The newer piece is ABB. On June 16, six days before the Forbes article, ABB Robotics and PSYONIC announced work pairing the Ability Hand with ABB's GoFa cobot. The companies said the collaboration will explore how touch and motion data from human prosthetic use can train robots to perform delicate and variable tasks that have been difficult to automate.

That work is being shown around Automate 2026, which runs June 22-25 in Chicago. The timing is useful for both sides. ABB gets to attach a concrete manipulation-data story to its broader push around autonomous versatile robotics. PSYONIC gets distribution and credibility in industrial automation without pretending that a prosthetic startup has already solved factory deployment by itself.

Marc Segura, president of ABB Robotics, framed the work around a long-standing gap in industrial robots: the instinctive handling that humans apply without thinking. ABB and PSYONIC said the collaboration will explore applications where the economics of automation break down when objects vary in shape, fragility or orientation.

The unresolved question is commercial. Neither the Forbes article nor the company announcements disclose the financial terms of the NVIDIA or ABB collaborations. There is no disclosed valuation, strategic investment or customer volume tied to the announcements. What is public is the direction of travel: PSYONIC wants the Ability Hand to be both a prosthetic and a robotics end effector, and it wants the data exhaust from human use to become part of the training stack.

A small company in a capital-heavy race

PSYONIC is not entering an empty field. Dexterous manipulation has become one of the core bottlenecks in embodied AI, drawing work from humanoid developers, industrial robot makers, simulation platforms and specialized hand companies. Tesla, Figure and other humanoid groups can collect robot-specific data through teleoperation and in-house fleets. NVIDIA can push simulation and tooling through Isaac Lab. Long-running dexterous-hand players and newer tactile-hand startups compete on mechanics, sensors and developer adoption.

PSYONIC's wedge is that it did not start as a robotics data company. It started with patients. That makes its dataset potentially different from lab demonstrations because the hand is being used by humans who are correcting grip, force and position in the messy world the robot eventually has to handle. It also makes the company dependent on careful consent, privacy and product boundaries as prosthetic-use data becomes more valuable to industrial robotics customers.

The company's public financials show both traction and constraint. In a May 2026 SEC filing, PSYONIC reported $5.35 million in 2025 net revenue, up from $3.33 million in 2024, and said the increase was driven by Ability Hand sales across human prosthetic users and robotics customers. The same filing said adoption had grown among more than 90 robotics companies, including Toyota, Mercedes, Hexagon and Amazon, and cited integrations with NVIDIA Isaac Lab and MuJoCo. PSYONIC still posted a $1.20 million net loss in 2025 and had $1.61 million in cash and cash equivalents at year-end.

That profile makes the NVIDIA and ABB work more than a branding exercise. For a company with modest revenue and no disclosed institutional mega-round, robotics data is a way to expand the market for hardware it already knows how to build. It is also a way to avoid being boxed into the reimbursement cycles and clinical adoption pace of prosthetics alone.

The hard part is proving that human prosthetic-use data transfers into reliable robot behavior at industrial scale. A factory does not pay for an elegant dataset. It pays for throughput, uptime, safety and lower integration cost. If PSYONIC can show that Ability Hand data helps robots handle objects that conventional grippers cannot, Akhtar's prosthetics company becomes something larger: a data supplier for the physical AI stack.

That is the real founder bet. Akhtar spent years building a hand that could give touch back to people. Now he is trying to show that the same touch can teach machines.