Vintix: Action Model via In-Context Reinforcement Learning
This work addresses the problem of scaling ICRL for generalist decision-making systems, representing an incremental step in the field.
The paper tackles the scalability challenge of In-Context Reinforcement Learning (ICRL) beyond toy tasks by introducing a fixed, cross-domain model that learns behaviors through ICRL, showing that Algorithm Distillation offers a competitive alternative to expert distillation for building versatile action models.
In-Context Reinforcement Learning (ICRL) represents a promising paradigm for developing generalist agents that learn at inference time through trial-and-error interactions, analogous to how large language models adapt contextually, but with a focus on reward maximization. However, the scalability of ICRL beyond toy tasks and single-domain settings remains an open challenge. In this work, we present the first steps toward scaling ICRL by introducing a fixed, cross-domain model capable of learning behaviors through in-context reinforcement learning. Our results demonstrate that Algorithm Distillation, a framework designed to facilitate ICRL, offers a compelling and competitive alternative to expert distillation to construct versatile action models. These findings highlight the potential of ICRL as a scalable approach for generalist decision-making systems. Code released at https://github.com/dunnolab/vintix