Tesla Sued Him for Secrets. Now He Ships Robot Hands.

Last year, Tesla accused former Optimus chief Jay Li of stealing trade secrets upon his departure. Earlier this month, the two parties reached a settlement, ending the case. Today, Li’s new startup Proception revealed an $11 million seed round led by First Round Capital (with participation from Y Combinator and BoxGroup). And the company is already shipping its first iteration of its ProHand robotic hands to research labs and companies.

The ProHand has 22 degrees of freedom (roughly the same as the human hand), allowing it to perform a much broader suite of tasks compared to typical gripper-based robotic hands. Dexterous manipulation has long been considered the Holy Grail of humanoid robots. Without it, they’re confined to a relatively small set of functions. As First Round partner Bill Trenchard, who led the investment, put it, “dexterous manipulation is a very, very, very important part of the whole humanoid story going forward.”

1The data flywheel: gloves that become hands

But what makes Proception unique isn’t just its hardware. It’s the company’s unique approach to teaching robots. To collect manipulation data, Proception outfits humans with sensor-laden gloves so they can gather fine-grained information about how to manipulate objects — like how to adjust grip pressure if an object starts to slide away. Then, it replicates those same sensors in the ProHand’s skin. This allows the robot to feel exactly the same sensations as the human tester. By decoupling the collection process from the robot itself, Proception can scale data collection without adding costs.

Dexterous robotic hands have long been considered the Holy Grail of robotics. As the Wall Street Journal reported last year, “fully functional robotic hands are at least a decade away,” according to Northwestern University robotics professor Kevin Lynch. But Li argues that the real barrier wasn’t the hardware — it was the lack of quality manipulation data to train the robots. With enough data, he says, you can accelerate progress on the hardware side. Rather than trying to build a better gripper, Proception is betting that a glove-based approach will enable it to solve the data problem more quickly.

The robotics startup space is crowded, but the humanoid market is still wide open. Figure, Physical Intelligence, Apptronik, and others are taking different approaches to enabling humanoid dexterity. But what makes Proception stand out is its particular blend of pedigree (Li designed hands for the most famous humanoid project out there) and a unique take on data generation that bridges the gap between human gloves and robotic skin.