Skill count hides the reliability gap in robotics

In Google DeepMind's preview for Gemini Robotics 2.5, the pitch centers on breadth: a single model learning hundreds of skills that stretch from folding laundry to sorting car parts on a factory floor. Grouping domestic chores and automotive manufacturing under one capability metric hides how opposed those environments are. Folding cloth allows generous physical compliance and loose timing. If an arm misses a corner or pauses to correct an awkward grasp, the task simply takes a second longer.

Factory assembly operates under opposite constraints. A robot sorting metal components on an automotive line must hit fixed cycle times and exact spatial tolerances. A dropped bracket halts the station and backs up the entire line.

Generalist foundation models continue to collect broad catalogs of demonstrated skills. For manufacturing plants, skill count is rarely the gating factor for adoption. Production leads need repeatability on a single station: thousands of consecutive cycles without an unhandled error or a manual reset.