Psyonic plans to pay amputees in Pakistan and India to record grip data for robots

In this plan, the sensor and the product are the same hand. In recent years robotics developers approached Psyonic about putting its hands on robots ^[3]. Worn by amputees, those same hands would record every grip ^[2][4]. "Robotics has a data problem," Akhtar told Semafor, especially around hands ^[18]. According to Semafor, other companies looking for training data offer free home cleanings in exchange for filming the work, or strap head-mounted cameras on factory workers in India ^[6]. Those methods record what a task looks like. Psyonic's also records how hard each finger pressed ^[4].

Touch is where the field is stuck ^[10]. Rich tactile manipulation remains "generally unsolved," said Kanishka Rao, a research director on Google DeepMind's robotics team ^[10]. A hand has to be skilled at "grabbing a raspberry without crushing it, but also lifting a 100-pound kettlebell and swinging it," Akhtar said ^[19]. The hardware also falls short: at about 22 degrees of freedom against 27, the best robot hands have roughly four-fifths of the independent motions of a human hand ^[9][1]. Barclays researcher Zornitsa Todorova said hands are often the biggest single cost in building a humanoid ^[11]. "It's loaded with motors and rare earths and actuators," she said ^[12].

The deal has three parties. Psyonic supplies the hand and pays the wearer ^[2]. A robotics company pays for a trained model ^[5]. Akhtar hopes the model revenue will then subsidize the hand ^[8]. The report does not say what Psyonic will pay wearers, what a hand costs to build, or what a model would sell for. Because it sells models, Psyonic will compete in the same business as Google DeepMind, whose models are built to run on other companies' robot hardware ^[14]. Akhtar also presents the program as a livelihood for people who are unemployed because they are missing limbs ^[20]. Millions of people worldwide have upper-limb amputations, most of them in developing countries ^[17].

So the supply of wearers is large, and the open question is whether enough robot makers will pay. Three outcomes are plausible. In the first, humanoid makers buy models trained on human grip, and the revenue covers hands and wages across the thousands of units Psyonic wants to place ^[2]. In the second, a July review by German researchers turns out to be right that "in-hand manipulation does not benefit from anthropomorphic hand design, as simpler mechanisms are sufficient" ^[13]. Buyers would then follow the hospital robots from Cobot and Diligent Robotics, which move on wheels with simple grippers ^[15]. In the third, the large labs get dexterity from other data. DeepMind showed a humanoid tying a trash bag when it rolled out Gemini Robotics 2 in July ^[16].

I think the force-and-torque log is the part of the plan that camera-based collectors cannot record ^[4][6], and it gives Psyonic a better chance at the first outcome than a video-only operation would have. The view is wrong if Psyonic's first model customers put the model on grippers instead of five-fingered hands, or if model sales cannot pay for one hand per wearer.