Eka Robotics Raises $150M at $1.45B Valuation on Force-First Robot AI

Eka Robotics has closed a $150 million Series A round that values the startup at $1.45 billion post-money, with a16z and F-Prime leading the financing. The company, founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja, is betting that force — not vision — is the key to teaching robots how to pick things up.

The round was co-led by a16z general partner Raghu Raguram, who told TechCrunch: "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."

The stakes are considerable. Robots remain poor at manipulation despite billions of dollars of investment and endless training videos fed into deep-learning models. Eka claims its system picks up roughly 1,500 objects per hour — objects weighing up to 1 kg — compared to about 600 for a human, with a success rate above 95%. If those numbers hold outside the lab, the economics of warehouse and factory automation change.

Why are veteran roboticists calling this a 'wow factor' demo?

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

Another roboticist, who requested anonymity because of their work at a frontier lab, compared seeing Eka's robot in action to the moment robots first achieved reliable legged locomotion five to ten years ago.

The arm moves fast enough to damage itself, which the anonymous observer said signals confidence in the control software: "they have a lot of trust in their algorithm not breaking the robot. I wonder how many robots they broke along the way?"

In an informal test, the frontier-lab roboticist handed the robot a raspberry. "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," they told TechCrunch. The robot did not crush the fruit.

What makes the model different?

Eka's architecture departs from the industry's dominant approach. Instead of a vision-language-action model — a system trained to map language, visual inputs, and actions — Eka built a vision-force-action model, which focuses on the relationship between language, force inputs, and actions.

Critics of the current generative-AI direction in robotics, including Rodney Brooks, have argued such models will fail without a representation of force and touch. Eka's results lend weight to that position.

"Vision is not the primary way in which you close this last millimeter problem," co-founder 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."

To make this work, Eka designed a robotic hand more flexible than traditional grippers and equipped with more force sensors. The hand reacts to contact, which is why it does not crush a raspberry. The founders are deliberately vague on specifics for competitive reasons, but the method centers on a physics simulator that models the multiple forces encountered during a grasp, feeding a reinforcement-learning environment that runs at scale. The company says it can generate 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 are 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 leading researchers in reinforcement learning for robotic control. Their bet: the technique scales through simulation, not through videos collected by gig workers or robot operators performing tasks over many hours. Simulation was the path to reliable manipulation — historically harder for tasks requiring interaction with objects rather than avoidance of them.

"What EKA is bridging is the generality-performance gap, which plagues a lot of modern robotics," Agrawal argues.

How does Eka fit alongside general-purpose AI models?

Today's most generalizable physical AI models — the ones that can attempt many tasks — are not particularly good at any of them. Models trained to excel at a single task may not close a business case for customers, since the promise of generative AI in robotics is rapid learning across multiple, even novel, tasks.

Current approaches can produce generally intelligent robots. Astra, ChatGPT's latest LLM, can operate a robot in an unfamiliar setting and perform tasks, but slowly and at high cost. Eka's model aims to handle the physicality while a more general model makes decisions. Astra might know that plates get stacked in a cupboard after the dishwasher; Eka's model picks up the dish without breaking it.

"Most of the field has been focusing on the planning," co-founder Haarnoja said. "So they approach from the top: high-level intelligence, pushing that down towards the physical. Whereas we think about it from the bottom up."

Can Eka survive outside the lab?

The company still needs proof in the field. Eka is running secretive pilots with several industrial customers in pick-and-place and wire manipulation, but the leap from lab to production line is almost as large as the gap between simulation and reality.

"What's going to be a big challenge is when they work with integrators on real assembly lines, and it needs to run without breaking for an entire year," Hauser said. The other open question, he added, is "'how generalizable these techniques can be to different tasks?' What I've seen is pick and place, but many industry tasks that are value added involve more than pick-and-place."

Deployability is Agrawal's focus. Eka's gripper hardware costs less than $1,000 and is rated for 5 million duty cycles; the target is the 20-million-cycle bar set by standard industrial arms.

"What we are building is that engineering of how to design entire workflows, but we are starting off being very pragmatic, and saying here are two verticals in pick, arrange, place, and wire handling, let's make them be home runs," Agrawal said. "Once we have made home runs on them, the sky is the limit in how much we can do with the approach we are developing."