Eka Robotics raises $150M at a $1.45B valuation to teach robots by self-play instead of human video

Eka Robotics announced a $150 million Series A on Oct 20, co-led by Andreessen Horowitz and F-Prime with returning seed leads Obvious Ventures and E14 Fund, and a wider list including MFV Partners, Bison Ventures, Glasswing Ventures, Samsung Next and several Fortune 500 strategics. TechCrunch put the valuation at $1.45 billion.

Eka’s $150 million raise is a bet against the imitation-learning playbook used by its billion-dollar-valued rivals. Physical Intelligence and Skild AI, both valued in the billions, train on large volumes of human teleoperation demonstrations, which makes the scale of data collection the moat. Eka says it skips human demonstrations: a robot picks up an object, drops it and retries millions of times inside a physics simulation, the way AlphaZero learned games without seeing human play, and it adds fingertip force sensing rather than relying on cameras alone. Eka calls the model vision-force-action.

Throughput is the claim to watch

Eka says its robots perform up to 1,500 pick operations per hour on a wide range of unseen objects, against a company-stated human baseline of about 600, roughly 2.5 times faster, and that the same robot can screw in a light bulb and insert wires. Transferring contact-rich skills learned in simulation to real hardware has historically been unreliable, and that has been the usual failure point for this approach. If the throughput figure holds outside the company's own cell, imitation learning loses its structural advantage: data collection stops being the scarce resource.

The result could come from somewhere else

A faster controller, a better gripper or a fixed task setup can each produce a high hourly count, and Eka has not published the per-task breakdown behind the 1,500 figure, so the number is a starting point rather than a benchmark result. The counter-case is that self-play in simulation has produced impressive videos before that did not survive a change of objects and lighting.

Eka was co-founded by MIT professor Pulkit Agrawal, who directs the Improbable AI Lab, and Tuomas Haarnoja, a co-author of the Soft Actor-Critic reinforcement-learning algorithm. The checkpoint is whether the 1,500-per-hour rate survives a task the company did not design.