The next great robotics foundation model might not be trained on words or videos. It could be trained on force. That is the bet behind Eka Robotics, a Cambridge-based startup that emerged from stealth this year with a $13 million seed round and a founding team drawn from the top labs at MIT and Google DeepMind [PitchBook, 2026]. The company is building an AI-native robotics platform centered on a new class of model it calls Vision-Force-Action, or VFA. The premise is simple. While other labs chase language-driven robotics or imitation learning from human videos, Eka argues that force is the fundamental language of the physical world. Its model learns manipulation skills primarily in high-fidelity simulation, with force as the primary input signal [theresanaiforthat.com]. The goal is a generalist model that can handle the messy, dynamic physics of real-world tasks, from assembly to logistics.
The VFA wedge
Eka's technical wedge is its focus on force-centric training. Most contemporary approaches to robot learning start with vision or language, treating the robot's physical interaction with the world as a secondary problem to be solved. Eka flips that script. Its VFA foundation model is designed to treat force feedback as a first-class sensory modality, alongside vision, for learning manipulation policies [clauday.com]. The model is trained in simulation, a scalable and safe environment where it can practice millions of trials without breaking hardware. The company claims this approach yields robots that are not only more capable but also safer, as the model develops an intrinsic understanding of physical constraints and contact dynamics [ekarobotics.com]. For investors like Nexus Venture Partners and Obvious Ventures, the bet is that this physics-first foundation will prove more robust and general than alternatives when deployed on real hardware.
A team built for the long game
The founders bring a rare blend of academic depth and large-scale industrial research experience. Pulkit Agrawal is an associate professor at MIT's Department of Electrical Engineering and Computer Science, where he leads the Improbable AI Lab and has received the IROS Toshio Fukuda Young Professional Award for his work in robotics [csail.mit.edu]. His co-founder, Tuomas Haarnoja, is a former Google DeepMind robotics researcher credited with shipping much of the lab's high-profile robot soccer work, a project that demonstrated advanced multi-agent reinforcement learning in the real world [clauday.com]. The broader team, estimated at around 25 employees, draws from institutions including MIT, Berkeley, Harvard, and companies like Boston Dynamics and Microsoft [LinkedIn]. This pedigree suggests Eka is built for the marathon of deeptech development, not a quick product sprint.
| Role | Name | Background |
|---|---|---|
| Co-Founder & Professor | Pulkit Agrawal | MIT EECS faculty; leads Improbable AI Lab; IROS Toshio Fukuda Award recipient [csail.mit.edu]. |
| Co-Founder | Tuomas Haarnoja | Former Google DeepMind robotics researcher; key contributor to robot soccer project [clauday.com]. |
| Senior Mechanical Engineer | Knight Ko | Mechatronics expert specializing in motors, actuators, and electronics packaging [LinkedIn]. |
The crowded field of generalist robots
Eka is entering a space crowded with well-funded rivals, all chasing the same prize: a general-purpose robotic manipulation system. The competitive landscape includes foundation model players like Physical Intelligence and Skild AI, humanoid robot builders like Figure, and other AI-native robotics startups such as Mind Robotics and Genesis AI. Each has a different technical approach and go-to-market hypothesis. Eka's differentiation rests on the primacy of its force-centric training paradigm. The company argues that models trained on language or passive video observation struggle with the precise timing and compliance needed for physical interaction. By building its foundation model around force, Eka aims to shortcut directly to robust, high-performance manipulation [clauday.com]. The risk, of course, is that simulation-trained policies fail to transfer perfectly to the real world,a classic robotics challenge known as the sim-to-real gap.
The road from simulation to shipment
For now, Eka Robotics remains in a development phase. The company has not publicly named commercial customers or specific industry partners. Its immediate milestones will be technical: demonstrating that its VFA model can perform a breadth of manipulation tasks in the lab and, crucially, on physical robots outside of simulation. The seed funding provides a multi-year runway to tackle these challenges. The hiring of senior engineers like Knight Ko, a mechatronics specialist, points to a parallel focus on hardware integration [LinkedIn]. The path to revenue likely runs through early deployments with strategic partners in structured environments, such as manufacturing or logistics, where the value of reliable, general-purpose manipulation is highest.
What to watch next
The next twelve months will test Eka's core technical hypothesis. Key signals will include published research demonstrating the VFA model's capabilities, partnerships with hardware OEMs or logistics firms, and the expansion of its engineering team beyond the current estimated headcount. A Series A round will likely follow once those technical milestones are hit. For Nexus and Obvious, the $13 million seed is a vote of confidence in a specific technical path through one of AI's hardest problems. The question for the market is whether force, not language, will be the native tongue of the next generation of robots.
Sources
- [PitchBook, 2026] Eka Robotics 2026 Company Profile | https://pitchbook.com/profiles/company/1326584-71
- [theresanaiforthat.com] Vision-Force-Action (VFA) | https://theresanaiforthat.com/model/vision-force-action-vfa/
- [clauday.com] Eka Robotics | https://clauday.com/article/49c4476f-1440-4e89-a0d9-f1f9021353d3
- [ekarobotics.com] Eka Robotics | https://ekarobotics.com/
- [csail.mit.edu, 2026] Pulkit Agrawal receives the IROS Toshio Fukuda Young Professional Award | https://www.csail.mit.edu/news/pulkit-agrawal-receives-iros-toshio-fukuda-young-professional-award
- [LinkedIn] Knight Ko | https://linkedin.com/in/knightko