Eka raises $150M for a force-feeling robot model
The MIT professor's manipulation startup is betting that touch, not language, is the missing input.
Pulkit Agrawal announced a $150M Series A for Eka Robotics, led by a16z and F-Prime Capital, with continued support from seed leads e14fund and Obvious Ventures. Lukas Ziegler puts the valuation at $1.45 billion.
Our robots retry until they succeed
Generality without performance is just a prototype.
The pitch is architectural. Most manipulation labs build VLAs (Vision-Language-Action models, which turn camera images and text into robot actions). Eka is building what it calls a Vision-Force-Action model, where a force-sensitive gripper feeds what the robot feels back into the control loop alongside what it sees.
"Force is the language of manipulation" is the first of three convictions Agrawal lists. The second is that reinforcement learning is how dexterous manipulation scales. The third is that fast, reliable physical execution, what he calls System 1, is the bottleneck to solve. Ziegler describes the training approach as simulation plus reinforcement learning, the same recipe that cracked legged locomotion.
The team
Agrawal is an MIT professor. Ziegler names Tuomas Haarnoja, a former DeepMind researcher, as part of the founding team, and credits him with Soft Actor-Critic, one of the most influential reinforcement learning algorithms of the last decade. That is a strong signal for a company whose whole bet is that RL beats imitation for contact-rich work.
Agrawal frames the company against what he calls the fundamental tension at the frontier of robotics, generality versus performance. His line is that generality without performance is just a prototype, and performance without generality is just a single-purpose tool. Eka says customers should not have to choose.
Shown versus claimed
The announcement comes with a demo video and the line "Our robots retry until they succeed," which is the clearest description of what force feedback is supposed to buy you. A robot that senses when a part has not seated can try again rather than carry on with an empty gripper.
What is not in the announcement is numbers. No success rates, no task list, no deployment counts, no customer names. Ziegler says one roboticist compared the demo to the moment robots first achieved reliable legged locomotion, which is a big comparison resting on one person's reaction to a clip. Agrawal says the money goes to scaling the model and accelerating customer deployments, so the real test is whether those deployments show up with numbers attached.
Eka says it is hiring across the team.
Force is the missing variable.
Sound like you were at the demo
One email a day. Under three minutes. The demos, the deployments and the part the press release left out.