In-Context Learning Brings a 'GPT Moment' to Robotics Skill Training

chris_j_paxton · x · 2026-08-28

For general-purpose robots to be useful and economical, they must learn new skills on the fly. In-context learning is making this possible.

Companies like Generalist, Skild, Rhoda, and RobbyAnt are demonstrating skills using long-horizon video demonstrations. The core idea is prompting with demonstrations that mix video and proprioception data. Language alone is insufficient to specify the problem well, but this multi-sensory context approach is starting to work effectively, signaling a potential 'GPT moment' for robotics.

Related event: In-Context Learning May Bring a GPT Moment to Robotics(2 posts)→

Original post →

More from Embodied

Embodied channel →