Eka Robotics, a Cambridge, Massachusetts start-up founded by MIT associate professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja, has raised a $150 million Series A at a post-money valuation of about $1.45 billion, TechCrunch reported on October 20, 2026. Andreessen Horowitz and F-Prime led the round; Obvious Ventures and E14 Fund, which led the seed, also invested.
Why it matters: this is a billion-dollar Series A for a manipulation company with no named customers, priced on a technical claim: that force sensing, not more vision, closes the last gap in grasping.
Key Facts
- $150M Series A at about $1.45bn post-money, led by a16z and F-Prime (TechCrunch, Oct 20, 2026)
- Model: “vision-force-action”, trained with reinforcement learning in a physics simulator that models grasp forces
- Gripper: under $1,000 in hardware, rated for 5 million duty cycles; target 20 million, the industrial-arm standard
- Company claims: about 1,500 picks an hour of objects up to 1 kg at above 95% reliability, against about 600 for a person
What is a vision-force-action model?
Most robot foundation models are vision-language-action models: they map camera images and instructions to motor commands. Eka’s model, as TechCrunch describes it, centres on the relationship between language, force inputs and actions. The hand carries more force sensors than a conventional gripper, so it reacts to contact rather than to what a camera sees. “Vision is not the primary way in which you close this last millimeter problem,” Agrawal said.
Training runs in simulation. Eka models the forces involved in grasping inside a physics simulator and uses it as a reinforcement-learning environment at scale; the founders say that produces thousands of hours of robot experience in minutes. Haarnoja contrasted this bottom-up approach with the field’s focus on high-level planning. Eka summed up its pitch in its announcement: “generality without performance isn’t a product. It’s a lab demo.”
What has Eka proven so far?
Speed in demonstrations, not durability in production. Eka runs undisclosed pilots with industrial customers on pick-and-place and wire manipulation. Kris Hauser of the University of Illinois Urbana-Champaign told TechCrunch the robot “was able to adapt to a large number of different objects without retraining,” and named long-term factory durability and tasks beyond pick-and-place as open questions.
The duty-cycle numbers show why. At Eka’s claimed 1,500 picks an hour, a gripper rated for 5 million cycles would reach that rating after about 3,300 hours, or roughly 140 days of round-the-clock work, by our arithmetic. Eka’s 20-million target would stretch that to about 18 months. Cheap hardware helps here: at under $1,000, a worn gripper is a consumable, not a capital item.
Andreessen Horowitz general partner Raghu Raghuram predicted superhuman performance on many dexterous tasks within months. For comparison, investors have also paid up for broad manipulation models: see our report on Generalist’s $3bn valuation.
Frequently Asked
How much did Eka Robotics raise?
$150 million in a Series A led by Andreessen Horowitz and F-Prime, at a post-money valuation of about $1.45 billion, as reported by TechCrunch on October 20, 2026.
Who founded Eka Robotics?
Pulkit Agrawal, an MIT associate professor, and Tuomas Haarnoja, a former Google DeepMind researcher. Both work on reinforcement learning for robot control and met as graduate students at UC Berkeley.
What is different about Eka’s robot model?
Eka calls it a vision-force-action model: it centres on force inputs as well as vision and language, and is paired with a force-sensing gripper that costs under $1,000 and is trained with reinforcement learning in a physics simulator.