Lightbringer: physical AI race goes to patent owners

Davies writes from 20 years as a patent attorney who has filed more than 300 patents across industrial automation, and he frames the problem as one of timing rather than volume. The examples he reaches for are old disputes: Boston Dynamics' 2022 suit against Ghost Robotics over "core technology" that dictated how a robot recovered from a fall, and the smartphone patent battles of the 2010s, which RoboSense's IP head warned last year robotics disputes may come to resemble.

The piece also points at an unresolved legal question. In swarm robotics, where the innovation is the interaction between dozens of machines, patent offices including the European Patent Office have indicated that claims may need to define the individual entities, the overall system, or the combination, depending on how the entities interact. Davies notes that Boston Dynamics and Ghost Robotics resolved their dispute in 2025, but argues startups rarely have the upper hand in ambiguous cases because of litigation budgets.

His prescription is selective rather than maximal: patent the one or two mechanisms the robot cannot work without, keep hard-to-reverse-engineer processes such as calibration techniques confidential, and build a wider ring of patents that makes designing around the core expensive. He cites KAIKAKU, a London food-assembly robotics startup, whose value lay in moving a bowl at speed without making a mess. Whether that layered approach holds up likely depends on how cheaply AI-assisted filing can be paired with attorney judgment, and whether courts and patent offices settle how behavior itself can be claimed.