Ex-Tesla Optimus Engineer's Dexterous Hand Startup | Proception

In this talk

1. 0:21

   The hand and its data glove

   Proception’s hand is tendon-driven with 22 degrees of freedom, with motors in the forearm; Jay says it can carry 20 kilograms and withstand unexpected collisions. The matching Proception Glove senses contact across the hand at 500 Hz. People can wear the same glove to record hand and head motion, egocentric vision, and interactions with objects for robot training.

2. 2:51

   Building and testing in Mountain View

   Jay shows the office, where Proception develops and assembles its hands, and says the company is building hundreds of units locally. The lab has a finger-cycling test rig, 3D printers, and an electronics printer for prototyping tactile and stretch sensors on fabric or TPU. Jay says the fabric-based sensing surface is designed to tolerate repeated bending, and worn gloves can be replaced without replacing the hand.

3. 9:14

   Why tendon-driven, human-shaped hands

   Jay says his work at Apple and on Tesla’s Optimus program, where he developed sensors including tactile sensors, led him back to startup work. He argues tendon actuation balances human-sized form, strength, compliance, and robustness, while a familiar five-finger shape can draw on human task and tool-use data. Proception positions itself as a hardware and data platform, leaving applications and model development to customers.

4. 14:42

   Integration, trade-offs, and customers

   The hand, including its wrist, weighs about 2.7 kilograms and is intended for adult-sized robots; its actuators sit in the forearm to support strength and extended operation. Jay acknowledges that tendon-driven hands make sim-to-real harder than direct-drive hands, and says customers can use simulation or real data collected with Proception’s gloves. He says the hand pair costs $50,000 at low volume and includes a glove, with volume discounts.

5. 24:49

   Jay’s outlook on humanoids

   Jay says his experience working on Optimus changed his view of humanoids: he points to progress from basic walking to dancing and early manipulation such as pick-and-place and screw turning. He expects capabilities to keep improving, while describing home use as dependent on scaling production to billions of robots. He also mentions clinical, hazardous, and inaccessible environments as potential applications.