Survey Maps Four Families of In-Context Learning for Robots, Toward Physical Recursive Self-Improvement

Knowin · hf · 2026-09-30

This literature review organizes in-context learning (ICL) for robots by the interfaces connecting contextual evidence to execution: context-conditioned policies, geometric demonstration transfer, world-model-based control, and skill/agent-based execution. It compares transfer assumptions and the roles of training, correspondence, and memory across manipulation and navigation, links method design to evaluation practices, and outlines an agenda toward physical recursive self-improvement.

Original post →

More from Embodied

Embodied channel →