Eka Robotics raises Series A led by a16z for dexterous AI manipulation

Andreessen Horowitz has led a Series A round in Eka Robotics, a startup building a Vision-Force-Action model for robotic manipulation. The firm announced the funding on October 20, 2026, framing dexterous manipulation as one of the largest remaining opportunities in robotics.

Eka’s pitch centers on the idea that most physical work—electronics assembly, food production, logistics—still depends on human hands because industrial robots struggle with real-world variability. According to the a16z announcement, the startup’s system integrates visual and tactile feedback, allowing robots to interpret physical contact as useful data rather than interference. This means robots could adjust mid-task, like correcting a slipping grip or adapting to resistance during assembly, instead of failing outright.

The company positions its technology as a way to bridge the trade-offs in robotics, aiming for systems that are both flexible and high-performing. On its website, Eka outlines three key principles: broad applicability across tasks and environments, real-time adaptability to unexpected conditions, and safe collaboration with human workers. The a16z post suggests this represents a departure from traditional robotics, which often forces compromises between precision and versatility.

Eka’s training methodology relies on simulated environments, where robots can experiment with countless scenarios before real-world deployment. The investors argue this method scales more efficiently than approaches requiring human demonstrations, though they note the long-standing challenge of ensuring simulated behaviors translate accurately to physical hardware. The post implies the startup’s founders have spent years refining techniques to narrow this gap.

Lab tests reportedly show the system achieving up to 1,500 picks per hour—outpacing human workers—across tasks like induction and bin-picking. The a16z post highlights demonstrations with fragile items, slippery components, and complex insertions, though the next hurdle will be maintaining this performance in full production shifts. If successful, the investors suggest the technology could reduce the need for task-specific retuning, making automation viable for previously impractical applications.

The startup’s leadership includes researchers with backgrounds in reinforcement learning and robotic control. One co-founder, affiliated with MIT CSAIL, has worked extensively on sensorimotor learning and sim-to-real transfer. Another, who developed foundational algorithms at Berkeley, later contributed to humanoid behavior training at Google DeepMind. The broader team draws from top academic institutions and companies like DeepMind and Boston Dynamics, with expertise in areas such as self-supervised learning and robotic simulation.

The funding round comes as robotics sees renewed investor interest, though recent deals have often focused on deployment or data infrastructure. While companies like Maven Robotics have already begun rolling out industrial robots, Eka remains in the lab, refining its core technology. The a16z post positions the startup as tackling a critical gap: if robots can reliably handle unstructured tasks, industries facing labor shortages could see meaningful automation gains.

What remains to be seen is whether Eka’s lab results hold up in real-world settings. The investors’ thesis hinges on the idea that solving dexterous manipulation at scale could transform sectors still reliant on manual labor. If the company can demonstrate consistent performance across multiple tasks without extensive retuning, it may attract manufacturers seeking alternatives to human workers. For now, the Series A signals confidence that the long-standing challenge of sim-to-real transfer is nearing a breakthrough.

Sources: unite.ai