ABB trains robot hands from prosthetic use

ABB Robotics and PSYONIC announced on June 16 a collaboration to train robotic grasping from data generated by prosthetic hand users. The central fact is specific: PSYONIC’s Ability Hand will be combined with ABB’s GoFa cobot to study how contact, motion and grip-force signals from human use can help robots handle delicate, irregular or variable objects. This is not a new humanoid robot, and it is not a claim of large-scale deployment. It is an attempt to solve a much less theatrical problem: teaching machines to grasp things that conventional industrial grippers do not understand very well.

The important distinction is the origin of the data. Many industrial robots work extremely well when the object, position and path are stable. They struggle more with a soft pouch, a fragile component, a poorly oriented tool or a flow of products that never arrive in exactly the same pose. Humans compensate for those variations through small adjustments of fingers, wrist and pressure. PSYONIC already builds a bionic hand for amputees, with articulated fingers, pressure sensing and tactile feedback. By using the same hand on people and on a robot, the two companies want to capture examples of human manipulation and translate them into repeatable industrial actions.

For ABB, the work fits its push toward more autonomous robots that can perceive, reason, move and manipulate in less rigid environments. GoFa contributes the repeatability and precision of a cobot, meaning a collaborative arm designed to work near operators under defined safety conditions. The Ability Hand adds a richer interface at the end of the arm. Together, they should let the companies test whether subtle changes in force, finger position and contact can be reproduced reliably, rather than merely observed in a controlled demonstration.

The practical scope covers automotive, aerospace, packaging, logistics and life sciences, according to the announcement. These are sectors where automation often breaks down at end-of-line handling, sorting or manipulation of non-standard objects. Caution still matters: the source describes exploration and R&D integration, not a finished product ready to replace a team. But the signal is useful. Physical robotics will not advance only through larger AI models or more spectacular humanoid bodies. It will need real contact data, instrumented hands and patient industrial testing to turn human dexterity into a robot skill that can be measured, repeated and trusted on a production floor.