A Survey of In-Context Reinforcement Learning

TL;DR

This survey reviews in-context reinforcement learning, where agents solve new tasks by conditioning on context without updating parameters, highlighting recent advances and research directions.

Contribution

It provides a comprehensive overview of in-context RL, summarizing key methods, challenges, and future research opportunities in this emerging area.

Findings

In-context RL enables task generalization without parameter updates.

Recent methods leverage context to improve adaptability.

The survey identifies open challenges and future research directions.

Abstract

Reinforcement learning (RL) agents typically optimize their policies by performing expensive backward passes to update their network parameters. However, some agents can solve new tasks without updating any parameters by simply conditioning on additional context such as their action-observation histories. This paper surveys work on such behavior, known as in-context reinforcement learning.