Field Notes

Robotics data is a moving target

Congratulations to Jerry and the Robotics Center team on a great SF Tech Week event last night, focused on robotics data and commercialization.

The panel explored the “ChatGPT moment” everyone is racing toward in robotics data collection. One takeaway: that moment is only the beginning. Production needs a data pipeline that keeps collecting and adapting as use cases and business needs evolve. Production is dynamic, and the data strategy has to keep pace.

Humanoid robotics needs hardware. Will automotive suppliers step up?

At Frontier Tower’s Physical AI event, the discussion turned to the new FCC restrictions and what they could mean for building humanoid robots in the US. The room saw opportunity. My mind went straight to automotive suppliers. They already know how to build motors, sensors, electronics, and reliable hardware at scale.

But pursuing that opportunity means accepting uncertainty: evolving designs, unproven customers, small initial volumes, and investment ahead of predictable demand. Will automotive suppliers commit? Can their organizations move quickly enough—and tolerate the risk long enough—to help build this market? The capabilities are there. The willingness to act before the business case is comfortable may be the real test.

Humans Set the Policy, Machines Earn the Autonomy. At Factory’s Open Source Evening, Octavian Sima shared a simple idea that stuck with me: humans set the policy rather than just improve the tasks. In Factory’s recursive self-improvement loop, human decisions become signal. Machines improve, evaluate and expand autonomy—but only as their track record earns it. Not human-in-the-loop. Human-above-the-loop.

Physical AI is attracting capital. Production is still the test. Why have so many robotics companies historically failed to reach production? AI may unlock new capabilities, but it does not remove the hard work of deployment. Robots still need real-world data, reliable hardware, safe operation, resilient supply chains, and systems that improve after they leave the lab. The winners will build more than impressive robots. They will build the operational infrastructure required to make them work—reliably, repeatedly, and at scale.

A glimpse inside San Francisco’s hardware ecosystem. Stumbled upon the very cool Studio 45 maker space in Noe Valley during its Summertime Hardware Pitch Night. The standout for me was Lit Motors: a self-balancing, enclosed two-wheel electric vehicle designed to carry two people. Think motorcycle efficiency with more of the protection and stability of a car. They’re now collaborating with Audi on a vehicle prototype.