EnvACE: Internalizing Environment Dynamics via World Rehearsal for Agentic Reinforcement Learning

2026-08-06

AI Summary

This work addresses the limited scalability of traditional large language model (LLM) agent training, which relies on costly interactions with real or synthetic environments. The authors propose EnvACE, a novel approach that internalizes environmental dynamics directly into the agent’s parameters. By introducing a “world rehearsal” mechanism, the agent alternately assumes the roles of executor and environment, enabling end-to-end joint optimization of action generation and response simulation to construct an internal world model that supports decision-making. At test time, a private rehearsal mechanism further enhances performance without requiring additional external interaction. Evaluated on BFCL-v4, tau²-Bench, VitaBench, and FinMCP-Bench, EnvACE significantly outperforms environment-augmentation baselines, and controlled experiments confirm its consistent efficacy across varying model scales.

Abstract

Training large language model agents for long-horizon tool use typically relies on interactions with real or synthesized executable environments, whose construction and verification are costly, or on external simulators that are difficult to ground. We introduce EnvACE, an agentic reinforcement learning method that replaces external environment interaction during training with world rehearsal. The policy alternates between acting and rehearsal: it first generates a tool call, then plays the role of the environment to produce the response induced by that action, and conditions subsequent decisions on the rehearsed response. Both roles are jointly optimized end-to-end using task-success rewards. Through world rehearsal, the policy internalizes the relationship between actions and their environment responses in its parameters, yielding an agent world model that directly supports decision making. Across BFCL-v4, tau^2-Bench, VitaBench, and FinMCP-Bench, EnvACE achieves strong and transferable performance, outperforming environment-scaling baselines in the overall evaluation. Controlled studies further show that world rehearsal consistently improves policy learning across model scales. At test time, the internalized world model enables private rehearsal before committed execution, yielding further gains under a moderate rehearsal budget without additional external interaction. Our findings establish world rehearsal as a new path toward scaling LLM agent training beyond the constraints of external environments. Our code is publicly available at https://github.com/Within-yao/EnvACE.

Problem

• agentic reinforcement learning
• environment dynamics
• world rehearsal
• tool use
• LLM agents

Innovation

• world rehearsal
• agentic reinforcement learning
• internalized world model
• tool-use agents
• environment-free training