The Robot That Screws In a Light Bulb: Eka's Dexterity Moment

Cambridge-based robotics startup Eka has demonstrated a robot arm capable of picking up objects as unpredictable as a rolling light bulb or a tangle of keys — a level of fluency that no commercial robot arm on the market currently achieves. Founded by MIT professor Pulkit Agrawal and former Google DeepMind robotics researcher Tuomas Haarnoja, Eka may represent the clearest evidence yet that robotic dexterity, long a fundamental barrier to broader automation, is approaching an inflection point.

A Light Bulb No Robot Could Screw In

According to Wired AI, the demonstration begins with what sounds like an accident waiting to happen: a robot claw bearing down on a light bulb at speed. Instead of crushing it, the arm decelerates, paws the table to locate the bulb, chases it across the surface after it rolls away, and eventually screws it into a socket. A Wired journalist who has covered robotics for over a decade describes the movement as the most natural he has seen from any machine — and notes that among the few dozen commercial robot arms available today, none can replicate it.

Founders Betting on Scale, Not Secrets

Agrawal and Haarnoja frame the result not as a one-off trick but as evidence that a long-standing obstacle has been cleared. “A couple of years ago, we realized that dexterity can finally be cracked,” Agrawal told Wired AI. The company says it is halfway to its goal, and that the remaining distance is a matter of scaling up — more training, more iterations — rather than unsolved fundamental science. Haarnoja’s robotics research background at Google DeepMind informs the approach.

Why This Matters

“Trillions of dollars flow through the human hand,” Agrawal told Wired AI, adding that he considers this “the biggest problem in the world to be solved.” Robotic dexterity sufficient for grocery shelves, restaurant prep stations, and household tasks would extend automation far beyond the structured environments of factories and warehouses. Wired’s framing — evoking the first time someone tried ChatGPT — signals that observers see a qualitative threshold being crossed, not merely an incremental hardware improvement. Whether Eka’s scaling hypothesis holds under real-world conditions remains the open question, but a robot that chases a rolling bulb and screws it in unassisted is a credible first answer to a problem that has stumped roboticists for decades.