Eka Robotics Hits $1.45B Valuation With $150M Series A for Force-Sensing Robot AI

The company says its arm can pick up about 1,500 objects weighing up to 1 kilogram in an hour, compared with roughly 600 for a human, and that it reliably grasps objects more than 95% of the time. The robot moves fast enough, and is strong enough, to hurt itself in the process. A roboticist at a frontier lab, who spoke to TechCrunch on condition of anonymity, said that speed suggested Eka trusts its algorithm not to break the hardware.

"They are already exceeding superhuman performance for certain tasks, and so we think over the next few months there will be superhuman performance in many dexterous manipulation tasks," Raghu Raguram, the a16z general partner who co-led the round, told TechCrunch.

Kris Hauser, a professor at the University of Illinois Urbana-Champaign, said Eka's demonstration carried a "wow factor" compared with other robot demos he had seen recently. The robot, he said, "was able to adapt to a large number of different objects without retraining," and showed a "humanness, creepy humanness that it would try over and over again if it failed the first time."

The frontier lab roboticist described watching the machine as an emotional experience comparable to the arrival of reliable legged locomotion five to ten years ago. Handed a raspberry, a favored test item at the lab, the robot rushed toward the researcher's arm, slowed in time and touched it gently. "The experience of a big, heavy robot rushing towards my arm and then slowing down just in time and very gently touching me is a feeling that I am, as we speak, carrying this feeling in my body," the researcher told TechCrunch.

Those results come from how the underlying model was built. Instead of a vision-language-action model, which learns the relationship between language, visual inputs and actions, Eka uses a vision-force-action model focused on language, force inputs and actions. Critics of the current direction of generative AI robotics, among them Rodney Brooks, have argued that such systems will fail without a way to represent force and touch.

"Vision is not the primary way in which you close this last millimeter problem," Agrawal told TechCrunch. "Vision tells you roughly where the object is, what the pose is, but the actual grabbing, the actual searching is happening through forces."

Eka designed its own hand, more flexible than conventional grippers and fitted with more force sensors, so that it reacts to contact and does not crush a raspberry. The founders decline to detail the method for competitive reasons; they say it rests on a physics simulator that models the multiple forces a grasp will encounter, which in turn feeds a reinforcement learning environment that runs at scale, generating thousands of hours of robot experience in minutes.

"Rather than being limited to the data that you're collecting from, say, human demonstrations, Eka is pushing through reinforcement learning this approach where the robot can surpass any kind of demonstrations that is given to it," Hauser said. "What's really exciting is that they're able to not be limited to the data that is presented to them for training."

Agrawal and Haarnoja first met in graduate school at the University of California, Berkeley, and both are established researchers in reinforcement learning for robot control. Eka's central bet is that the technique can be run at scale through simulation rather than relying on video collected from human demonstrators or operators working through tasks over many hours. Agrawal argues that what Eka is bridging is "the generality-performance gap, which plagues a lot of modern robotics."