Eka is worth $1.45B thanks to a physical AI model that has experts raving

Eka Robotics Raises $150M for Force-Sensitive AI

Eka Robotics, a startup founded by prominent AI researchers, has secured a $150 million Series A funding round at a $1.45 billion post-money valuation. The round, co-led by a16z and F-Prime, backs the company’s novel approach to robotic manipulation, which has generated significant attention from industry experts for its performance in physically handling objects—a persistent challenge in the field. The company aims to solve the “last millimeter problem” in robotics, where visual systems alone are insufficient for reliable grasping.

A Vision-Force-Action Model

At the core of Eka's technology is a departure from conventional vision-language-action models. Instead, co-founders Pulkit Agrawal of MIT and Tuomas Haarnoja, formerly of Google DeepMind, have developed a vision-force-action architecture. This system pairs a custom, force-sensitive gripper with a deep learning model trained extensively via reinforcement learning in a physics simulator. This allows the robot to learn from thousands of hours of simulated experience in minutes, adapting to objects without retraining and exhibiting a nuanced sense of touch, enabling it to handle delicate items like a raspberry without crushing them.

• Performance: Picks up approximately 1,500 objects per hour, weighing up to 1 kg.
• Reliability: Achieves a success rate of over 95% in picking tasks.
• Training Method: Uses reinforcement learning in simulation, allowing the model to surpass performance seen in human demonstration data.
• Hardware: The custom gripper costs less than $1,000 and is rated for 5 million duty cycles.

This “bottom-up” strategy, which prioritizes mastering physical interaction, stands in contrast to top-down approaches where high-level models like Google's Astra attempt to reason their way through physical tasks. While Eka is currently in pilot programs for pick-and-place and wire manipulation, it faces the challenge of proving its system's durability and generalizability in real-world industrial environments. The company's goal is to bridge the “generality-performance gap,” where current general-purpose robots often lack the efficiency required for commercial applications.

Eka's early success suggests a market validation for architectures that prioritize physical interaction and force feedback through simulation, proposing a more direct path to solving industrial automation challenges than top-down, language-driven reasoning alone.