ReadingResearchRadarInvestment framework
Sign in / Sign up
← Back to reading
News3 min read·2026-10-20DisconfirmAI Editorial

a16z Leads Eka Robotics' Series A: How Force Sensing and Simulation Training Change Automation Integration Costs

In an investment letter announcing a16z's lead in Eka Robotics' Series A, partners Yoko Li and Raghu Raghuram articulate their thesis on dexterous manipulation robots: the bottleneck in current automation is not insufficient robot capability but the need to re-engineer integration for every new object or material; if robots can adapt to variation through force feedback and simulation training, tasks that were never worth a dedicated automation project become addressable.

ImportanceMaterialEvidenceE3 inspectableWrite-upStandard

Original publication time 2026-10-20 21:05 GMT+8

The real bottleneck in industrial automation is not that robots are too slow or inaccurate, but integration cost. In an investment letter announcing a16z's lead in Eka Robotics' Series A, partners Yoko Li and Raghu Raghuram put forward this judgment: today, many automation projects only make economic sense when task volume is high enough to amortize engineering tuning costs; every new object, material, or process can trigger another integration project. If robots can adapt to these variations, tasks that were never worth a dedicated automation retrofit become feasible.

Eka's technical approach combines visual perception with force feedback. Yoko Li and Raghu Raghuram describe a concrete scenario: when inserting a connector, vision can tell a robot where the socket is, but once contact begins, the critical information changes—has one edge caught, is the connector beginning to bind, should the robot push harder, rotate slightly, or pull back. Humans answer these questions almost effortlessly through thousands of specialized sensory receptors in their fingertips and palms. Eka does not attempt to rebuild this biological sensing system receptor by receptor; instead, it designs a Vision-Force-Action model that lets the robot use physical contact itself as information—a grasp that starts to slip can be corrected, contact with a surface can help locate an object, and resistance during insertion changes the robot's motion rather than simply triggering an error.

Sign in / Sign up
ReadingResearchRadarInvestment framework

On how robots acquire physical skill, Eka chooses simulation-based reinforcement learning over human demonstrations. Yoko Li and Raghu Raghuram identify two inherent limitations of human demonstrations: first, they consume human time—you cannot practically show a human demonstrating one million slight variations on picking up an egg; second, it is hard for humans to demonstrate all the ways you can fail at picking up an egg in a way that conveys useful information. Virtual robots can practice millions of variations in object position, friction, and contact before a policy ever reaches physical hardware, experiencing failure modes humans rarely demonstrate and learning to recover from them. If this approach works at scale, the implication is that robot experience can increasingly scale with compute rather than human labor.

Eka reports up to 1,500 picks per hour in laboratory induction and bin-picking tasks, two and a half times faster than humans, and has demonstrated manipulation across fragile objects, slippery industrial parts, and contact-rich insertion tasks. Yoko Li and Raghu Raghuram explicitly write that sustaining that performance through full production shifts is the next test.

The founders' background is relevant to this thesis: Pulkit Agrawal leads the Improbable AI Lab at MIT CSAIL, with research spanning reinforcement learning, force-aware control, and moving policies from simulation into the real world; Tuomas Haarnoja co-created Soft Actor-Critic at Berkeley, one of the foundational algorithms in modern reinforcement learning, and later worked at Google DeepMind training complex humanoid behaviors in simulation and deploying them on real machines. The two have spent much of the last decade attacking the same fundamental question from complementary directions: how can machines learn sophisticated physical behavior through experience, and have that intelligence survive contact with reality?

It should be noted that a16z is the lead investor in this round, and this letter is simultaneously an investment announcement. Performance data comes from Eka's self-report and has not been independently verified; the sim-to-real gap remains one of robotics' hardest recognized problems, with contact dynamics notoriously difficult to model—small errors in friction, compliance, or actuator behavior can turn a successful simulated policy into a useless real-world one. Yoko Li and Raghu Raghuram themselves acknowledge this, while noting that Eka's founders are unusually well suited to solve this problem and have been demonstrating that the gap between simulation and reality is narrowing.

Sources

  • 原始出处a16z.news↗
Continue researching →