Robot that 'feels' its grip makes Eka a $1.45 billion start-up

A robotics start-up that teaches machines to feel what they are holding is now worth $1.45 billion. Eka Robotics has closed a $150 million Series A round, led by the venture firms a16z and F-Prime, for a system that pairs a force-sensitive gripper with a new kind of deep-learning model.

Eka was founded by Pulkit Agrawal, a professor at the Massachusetts Institute of Technology, and Tuomas Haarnoja, a former researcher at Google DeepMind. The two met as graduate students at the University of California, Berkeley.

Touch before sight

Most efforts to build general-purpose robots lean on cameras and on models that plan movements in advance. Eka’s founders argue that the hard part comes at the very end, when fingers meet an object. “Most of the field has been focusing on the planning,” Haarnoja said.

“Vision is not the primary way in which you close this last millimeter problem,” Agrawal said. His company’s model reads force through the gripper and adjusts as it goes. “What EKA is bridging is the generality-performance gap,” he said, meaning the trade-off between robots that can handle many tasks and robots that do one task well.

The robotics pioneer Rodney Brooks has made a similar argument from the outside, warning that generative AI models for robots will fail unless they represent force and touch.

What the robot can do

Eka says its arm can pick up about 1,500 objects an hour, each weighing up to a kilogram, compared with about 600 for a human worker, and that it gets a reliable grip more than 95% of the time. The arm moves fast enough that it can damage itself.

The gripper hardware costs under $1,000 and is rated for five million duty cycles. The company is aiming for 20 million, the usual standard for industrial arms. It trains its model in a physics simulator using reinforcement learning, which Eka says can produce thousands of hours of robot experience in minutes.

Raghu Raguram, the a16z general partner who co-led the round, said: “they are already exceeding superhuman performance for certain tasks.”

Kris Hauser, a roboticist at the University of Illinois Urbana-Champaign who has seen the system, said it “was able to adapt to a large number of different objects without retraining.”

Questions still open

Hauser also raised the questions any buyer will ask: whether the system can run for a year on a real assembly line, and whether it can do more than pick and place.

Eka is running pilots with industrial customers on pick-and-place work and on handling wires, though it has not named them.