Eka Robotics has secured a $150 million Series A funding round led by a16z and F-Prime, valuing the startup at $1.45 billion post-money. Founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja, the company developed a manipulation architecture featuring a force-sensitive gripper and a new vision-force-action deep-learning model designed to overcome robotic grasping limitations.
The system combines a unique robotic hand with a physics simulator to run large-scale reinforcement learning environments, generating thousands of hours of experience in minutes. In demonstrations, the robot can handle about 1,500 objects weighing up to 1 kg per hour with a success rate exceeding 95% without retraining, impressing roboticists with its speed, adaptability, and delicate touch handling.
This development matters because it demonstrates a shift from vision-language-action models to a vision-force-action approach, successfully tackling the complex physical interaction and touch challenges that have long constrained robotic dexterity in real-world environments.