Eka raises $150M Series A at a $1.45B valuation, with its physical AI model picking about 2.5x faster than humans

TechCrunch reported on Oct 20 that robotics company Eka closed a $150 million Series A led by a16z and F-Prime at a $1.45 billion post-money valuation. The company was founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja, with a16z general partner Raghu Raghuram involved in the raise.

On technology and product, Eka's physical AI model targets object picking and dexterous manipulation, paired with a force-sensitive gripper and robot arm. The gripper carries more force sensors and adjusts its motion in real time on contact to avoid crushing delicate items, with current hardware cost under $1,000 and a rated life of 5 million duty cycles against a goal of 20 million cycles to match standard industrial arms. Eka says its model adapts to a wide range of objects without retraining, retries after failures, and can generate thousands of hours of robot training experience within minutes via physics simulation. Co-founder Pulkit Agrawal said vision only roughly locates an object and that the last millimeter of whether it is actually grasped is decided by force feedback.

On performance, Eka says it picks roughly 1,500 objects per hour at up to 1kg with over 95% reliability, against about 600 per hour for humans, which works out to around 2.5 times human speed on the reported figures. Throughput and reliability are primarily company-disclosed rather than a complete third-party standardized test.

On commercialization, Eka is running confidential pilots with industrial customers focused on pick-and-place, pick-arrange-place and wire handling. The design splits responsibilities so a general model plans the task while the Eka model executes the physical work, for example the general model knowing that a plate in the dishwasher should go into a cabinet while Eka's system safely picks it up and moves it.

The main unresolved problems the article lists are that industrial arms need to run for a full year without failure, that sim-to-real transfer may not carry directly to production lines, that current demos center on pick-and-place while other industrial tasks involve more complex assembly and contact, and the durability gap between 5 million and 20 million cycles. The article does not provide Eka's pre-round investors.