Cambridge, Massachusetts-based Eka Robotics has raised $150 million in Series A funding to develop general-purpose robots capable of performing complex physical tasks faster than humans.
The round was co-led by Andreessen Horowitz and F-Prime, with participation from Obvious Ventures, E14 Fund, MFV Partners, Bison Ventures, Glasswing Ventures, Samsung Next and strategic investors associated with Fortune 500 companies. According to TechCrunch, the financing values Eka at approximately $1.45 billion post-money.
The company was founded by MIT professor Pulkit Agrawal and former Google DeepMind researcher Tuomas Haarnoja. Agrawal is a robotics researcher at MIT's Computer Science and Artificial Intelligence Laboratory, where his work focuses on robotic manipulation and locomotion. Haarnoja previously worked on reinforcement learning, including research into algorithms that allow machines to learn complex behaviors through trial and error.
Eka is addressing one of the most persistent limitations in robotics: manipulating unfamiliar objects reliably.
Modern robots can perform repetitive tasks with high precision in controlled environments. However, relatively simple human activities, such as picking up fragile fruit, inserting cables or handling objects with unfamiliar shapes, remain difficult to automate.
The problem is not simply recognizing an object. A robot must also understand physical interaction: how much pressure to apply, whether something is slipping, and how an object responds as it moves.
Eka is developing a Vision-Force-Action model, or VFA, that combines visual perception with force feedback.
Unlike systems trained primarily to imitate demonstrations of human actions, Eka uses reinforcement learning in simulated environments to develop manipulation capabilities.
Simulation allows its models to experience thousands of hours of interactions in minutes, with variations in materials, objects and physical conditions. The company then transfers those learned behaviors to physical robots.
Eka says its approach is beginning to overcome the longstanding simulation-to-reality gap, where skills developed in simulated environments fail to transfer reliably to real-world hardware.
Its reported results are notable.
The company says its robots can manipulate previously unseen objects at rates of up to 1,500 operations per hour, compared with a human benchmark of approximately 600 operations per hour for the tasks tested.
Eka has also demonstrated robots screwing light bulbs into sockets and inserting wires, activities that require fine manipulation and contact-sensitive control.
Those demonstrations suggest progress toward combining general-purpose manipulation with production-level throughput, although the results have not yet established performance across complete industrial shifts or a broad range of customer environments.
The commercial opportunity spans manufacturing, logistics, food handling and electronics assembly.
Many industrial processes still depend on human dexterity because traditional robotic systems cannot adapt economically to changing objects and operating conditions.
Eka is betting that a common foundation model for manipulation could reduce the cost of programming robots for individual tasks, allowing the same underlying technology to support multiple applications.
The company plans to use the financing to expand its VFA model, develop additional capabilities and accelerate customer deployments.
However, the transition from laboratory demonstrations to industrial operations presents significant challenges.
Factories require consistent success rates over thousands of repetitions, predictable maintenance, worker safety and clear financial returns.
A robot that briefly exceeds human performance in a controlled test must still prove that it can sustain that advantage across different shifts, products and environments.
Eka also faces competition from robotics foundation model developers and manufacturers developing their own AI systems.
Its potential advantage lies in combining a learning architecture with force-sensitive hardware and extensive simulated experience.
The decisive question is whether that combination can achieve reliable, general-purpose manipulation at a commercially viable cost, rather than producing impressive demonstrations that remain difficult to deploy at scale.