Eka Robotics, a startup founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja, has closed a $150 million Series A round that values the company at $1.45 billion post-money. The funding, led by venture funds a16z and F-Prime, supports the deployment of a new physical super intelligence (SI) architecture designed to improve robotic dexterity and manipulation.
What Happened
The company’s core innovation is a manipulation architecture that combines a force-sensitive gripper with a vision-force-action model. Unlike traditional vision-language-action models, which map language and visual inputs to actions, Eka’s system focuses on the relationship between language, force inputs, and actions. This approach allows the robot to adapt to a large number of different objects without retraining, a capability that has impressed veteran robot builders.
According to Eka, the robot can pick up approximately 1,500 objects weighing up to 1 kg in an hour, compared to about 600 for a human, with a reliability rate of more than 95%. The system uses a physics simulator to model multiple forces encountered during grasping, creating a reinforcement learning environment that generates thousands of hours of robot experience in minutes. This method allows the robot to surpass the limitations of data collected from human demonstrations.
Raghu Raghuram, the a16z general partner who co-led the round, told TechCrunch that the system is "already exceeding superhuman performance for certain tasks," with expectations for superhuman performance in many dexterous manipulation tasks in the coming months. Kris Hauser, a professor at the University of Illinois Urbana-Champaign, described the demo as having a "wow factor," noting the robot’s ability to retry failed attempts with a persistence he described as having a "creepy humanness."
Why It Matters
The development addresses a critical bottleneck in the SI industry: the gap between generalizable intelligence and physical performance. While large language models like Astra have shown potential for high-level planning in robotics, they often struggle with the precise, real-time physical adjustments required for delicate tasks. Eka’s bottom-up approach, prioritizing force feedback over pure visual planning, aims to bridge this gap.
Critics of current generative SI robotics models, such as Rodney Brooks, have long argued that systems will fail without representing force and touch. Eka’s results provide evidence supporting this view, suggesting that future SI agents for physical tasks may rely less on video training data and more on simulation-driven reinforcement learning. The company’s gripper hardware, costing less than $1,000, is currently rated for 5 million duty cycles, with a goal to reach the 20 million cycle standard of industrial arms.
The Bottom Line
Eka Robotics has secured significant capital and expert praise for its vision-force-action model, which promises to make robots more dexterous and reliable. However, the company still faces the challenge of proving its technology in real-world industrial settings beyond controlled pilots. The success of its approach in scaling from simulation to production lines will be a key indicator of its viability in the broader SI market.