Eka Robotics, a startup building robot grippers that sense what they hold, has closed a $150 million Series A led by a16z and F-Prime, TechCrunch reported on 20 October. The $1.45 billion figure attached to the story is the post-money valuation, not the amount raised. The money is the smaller number, and the bet behind it is that robots can finally handle delicate objects at factory speed.
The founders are Pulkit Agrawal, an MIT professor, and Tuomas Haarnoja, formerly of Google DeepMind. Their paths crossed as graduate students at the University of California, Berkeley. Both work on reinforcement learning for robot control, and Eka’s wager is that this training method can run at scale inside simulation, rather than depending on video of people or remote operators doing chores.
The technical claim is specific. Most current robot models are vision-language-action systems, which learn how words, images, and movements relate. Eka says its model swaps sight for touch: a vision-force-action design that centres the forces a hand feels. “Vision tells you roughly where the object is,” Agrawal told TechCrunch, but “the actual grabbing” happens through forces. Training happens in a physics simulator that models contact forces. By the company’s own account, the robot piles up practice worth thousands of hours while only a few minutes go by on the clock. The founders would not explain how the hand itself works.
The performance numbers are the company’s. Eka says its robot succeeds on more than 95 percent of pick attempts. TechCrunch puts its pace at about 1,500 objects of up to 1 kg an hour, against roughly 600 for a person. Agrawal says the gripper costs under $1,000 and carries a 5 million duty cycle rating. The standard for industrial arms is 20 million, and Eka lists that as a target, not a result. No independent benchmark appears anywhere in the reporting.
The praise needs the same labelling. Raghu Raguram is a general partner at a16z, and he says the robots are “already exceeding superhuman performance for certain tasks” and expects superhuman results on many dexterous tasks within months. His firm co-led the round, so he owns a piece of the company now. Treat the claim as a pitch, not a verdict.
The independent voices are more useful, though uneven. Kris Hauser, a professor at the University of Illinois Urbana-Champaign, is on the record. He called the demo a “wow factor” and said the robot adapted to many objects without retraining. A second roboticist, who works at a frontier AI lab and asked TechCrunch for anonymity, compared the experience to the first reliable legged robots of five to ten years ago.
That second person supplied the best detail. They gave the robot a raspberry, apparently the favourite test item at that lab. A heavy arm rushed toward their hand, slowed just in time, and touched it “very gently.” TechCrunch says the hand does not crush the fruit because it reacts to contact. The same observer wondered “how many robots they broke along the way,” since an arm this strong can damage itself at speed.
Hauser also supplies the doubts. The hard test, he said, comes when Eka works “with integrators on real assembly lines” and has to run “without breaking for an entire year.” He has seen only pick and place, and noted that many valuable industrial jobs involve more than that. TechCrunch adds that moving out of the lab and onto a production floor is nearly as hard as getting a simulated skill to work on a real arm.
Deployment is thin. Eka’s pilots, run in secret with several industrial customers, cover wire handling and pick-and-place. It has named none of them, and the reporting gives no revenue or shipping date.
The market angle is the real story. A price this high for a company with no disclosed production customers says investors expect touch to be the missing piece for general-purpose robots. If the pilots turn into year-long contracts, expect rivals built on vision alone to be asked why their robots still squash the fruit; if they stall, a $1.45 billion valuation will have been set by a demo.
Reported by TechCrunch (Tim Fernholz), 20 October 2026.
