Eka Robotics raises $150 million for robot hands that sense force

Eka Robotics has raised $150 million to tackle a stubborn part of automation: getting robots to grasp and handle objects reliably. The Series A, co-led by Andreessen Horowitz and F-Prime, values the company at $1.45 billion after the investment. Its approach gives robots information cameras alone cannot provide: what happens when they make contact.

Obvious Ventures and E14 Fund also participated. Eka says the money will help it take its manipulation model into more tasks and accelerate customer deployments. It is already conducting industrial pilots involving pick-and-place work and wire handling.

Contact becomes an input

Eka calls its technology a vision-force-action model. It combines visual information with force feedback from its robotic hands. Rather than develop the AI model, sensors, controller and hardware separately, Eka designs them together, according to a16z’s investment announcement.

Consider inserting a connector. A camera can help locate the socket, but contact can reveal that an edge has caught or the connector is binding. In a16z’s explanation, that resistance helps the system adjust its motion. Force feedback can also help correct a slipping grasp.

Practice without a human demonstrator

The training approach relies on reinforcement learning: robots learn how to act through repeated attempts and feedback. Eka runs that practice in physics simulation, where virtual robots encounter variations in object position, friction and contact. This lets them rehearse failures and recoveries that humans rarely demonstrate, a16z says.

Eka was founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja. The founders say they can generate thousands of hours of robot experience in minutes. But small errors in simulated friction, material flexibility or motor behavior can make a successful virtual strategy fail on real hardware, a16z notes.

Fast picks are not full production shifts

Eka reports more than 95% picking reliability and the ability to handle objects weighing up to one kilogram. Its investor places the reported throughput of up to 1,500 picks an hour in laboratory induction and bin-picking tasks. Sustaining that performance through full production shifts remains the next test.

Hardware durability is another hurdle. Agrawal says Eka’s gripper costs less than $1,000 and is rated for five million operating cycles. The company wants to reach the 20 million cycles expected of standard industrial arms. The price is for the gripper, not an entire robotic installation.

There is also a distinction between handling varied objects and handling varied jobs. Hauser praised Eka’s ability to adapt to different objects without retraining, but questioned whether its techniques would generalize beyond pick-and-place to other valuable industrial tasks.