In-Context Learning for Robots: Methods and Applications
A survey groups robot in-context learning into four interfaces from context to action.
This literature review organizes in-context learning for robots around interfaces that turn demonstrations and interaction into action while neural parameters stay fixed at deployment. It distinguishes context-conditioned policies, geometric demonstration transfer, world-model-based control, and skill- and agent-based execution. The survey compares transfer assumptions and the roles of training, correspondence, and memory across manipulation and navigation. It links method design to evaluations of teaching responsiveness, physical transfer, and retained experience, toward compositional task acquisition and physical recursive self-improvement.
- Survey covers in-context learning with robot parameters held fixed.
- Four families link demonstrations and context to physical execution.
- Analysis spans manipulation and navigation under changing conditions.
- Agenda connects compositional tasks with physical recursive self-improvement.
From the original article
1 min read at huggingface.co
General-purpose robots must infer what a new task requires and translate that understanding into appropriate physical action. In-context learning (ICL) for robots supports this process by using demonstrations and interaction to direct existing competence with neural parameters held fixed during deployment.
By Haojian Huang, Zexi Li, Junhao Guo · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.