Robots are getting a grip. Eka Robotics, with its advanced force-sensitive gripper and unique deep-learning model, has just closed a $150 million funding round, valuing the startup at $1.45 billion. Founded by MIT professor Pulkit Agrawal and Tuomas Haarnoja, a former Google DeepMind researcher, Eka’s technology is already surpassing superhuman performance in certain tasks, according to venture partner Raghu Raguram of a16z.

Their robot’s capabilities are shocking, even to veteran robot builders. Kris Hauser of the University of Illinois Urbana-Champaign described it as having a “wow factor” and noted its ability to adapt to a wide range of objects without retraining. Another anonymous roboticist was moved by the experience, comparing it to the first reliable legged locomotion of robots.

Eka’s key innovation lies in its vision-force-action model, focusing on the relationship between force inputs and actions, rather than vision-action models. This allows the robot to pick up around 1,500 objects an hour, compared to about 600 for a human, and it does so with a near 95% success rate. The robot is robust enough to avoid crushing items like a raspberry, thanks to its unique design and reinforcement learning environment generated through physics simulations.

Co-founder Agrawal explains that vision alone is not enough for the final millimeter problem, and that force plays a crucial role. This work is pushing the boundaries of general-purpose robotics, potentially bridging the gap between generality and performance.

The future of robotics may be shaped by Eka’s approach, where a physically adept bot can perform a multitude of tasks with minimal human intervention, making robots more versatile and efficient than ever before.