Can RL Improve Generalization of LLM Agents? An Empirical Study

• URL: http://arxiv.org/abs/2603.12011v1
• Date: Thu, 12 Mar 2026 14:54:59 GMT
• Title: Can RL Improve Generalization of LLM Agents? An Empirical Study
• Abstract summary: Reinforcement fine-tuning (RFT) has shown promise for training LLM agents to perform multi-turn decision-making based on environment feedback.<n>In real-world deployment, agents may operate in unseen environments with different background knowledge.<n>We show that RFT generalizes well across task difficulty within an environment, but exhibits weaker transfer to unseen environments.
• Score: 68.19349692042341
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Reinforcement fine-tuning (RFT) has shown promise for training LLM agents to perform multi-turn decision-making based on environment feedback. However, most existing evaluations remain largely in-domain: training and testing are conducted in the same environment or even on the same tasks. In real-world deployment, agents may operate in unseen environments with different background knowledge, observation spaces, and action interfaces. To characterize the generalization profile of RFT under such shifts, we conduct a systematic study along three axes: (1) within-environment generalization across task difficulty, (2) cross-environment transfer to unseen environments, and (3) sequential multi-environment training to quantify transfer and forgetting. Our results show that RFT generalizes well across task difficulty within an environment, but exhibits weaker transfer to unseen environments, which correlates with shifts in both semantic priors and observation/action interfaces. In contrast, sequential training yields promising downstream gains with minimal upstream forgetting, and mixture training across environments improves the overall balance. We further provide detailed analyses and deeper insights, and hope our work helps the community develop and deploy generalizable LLM agents.

Related papers

• Learning Simple Test-Time Environments for LLM Web Agents [25.206333684249987]
  We introduce trial steps for agents to decompose a complex environment observation into sub-modules.<n>We implement a label-free learning method, Test-Time Environment Decomposition (TTED), to adapt agent behaviors with experience during inference.
  arXiv  Detail & Related papers  (2026-08-29T14:45:09Z)
• Learning Generalizable Behaviors for Terminal Agents [72.62493560701414]
  We propose a simple training recipe that improves reward quality by filtering low-quality environments and augmenting outcome rewards with process-level behavior regularization.<n>Using this recipe, our RL-trained agent account the best performance among evaluated open-source RL-trained 8B models across four terminal-agent benchmarks.
  arXiv  Detail & Related papers  (2026-08-23T22:34:40Z)
• EnvACE: Internalizing Environment Dynamics via World Rehearsal for Agentic Reinforcement Learning [57.312638079204255]
  Training large language model agents for long-horizon tool use typically relies on interactions with real or synthesized executable environments.<n>We introduce EnvACE, an agentic reinforcement learning method that replaces external environment interaction during training with world rehearsal.
  arXiv  Detail & Related papers  (2026-08-06T15:54:36Z)
• State2State: Environment-Derived Mid-Training for LLM Agents [67.08889234341996]
  We study an environment learning paradigm in which agents acquire interaction and manipulation capabilities solely through environment interaction.<n>We propose State2State, an environment-derived mid-training method that converts explored environment states into training objectives.<n> Experiments on ALFWorld and ScienceWorld show that State2State improves agent performance as a standalone environment-learning stage in most settings.
  arXiv  Detail & Related papers  (2026-08-05T15:02:41Z)
• Autonomous Continual Learning of Computer-Use Agents for Environment Adaptation [57.65688895630163]
  We introduce ACuRL, an Autonomous Curriculum Reinforcement Learning framework that continually adapts agents to specific environments with zero human data.<n>Our method effectively enables both intra-environment and cross-environment continual learning, yielding 4-22% performance gains without forgetting existing environments.
  arXiv  Detail & Related papers  (2026-02-10T23:06:02Z)
• Agent World Model: Infinity Synthetic Environments for Agentic Reinforcement Learning [62.499592503950026]
  Large language model (LLM) have empowered autonomous agents to perform complex tasks that require multi-turn interactions with tools and environments.<n>We propose Agent World Model (AWM), a fully synthetic environment generation pipeline.<n>We scale to 1,000 environments covering everyday scenarios, in which agents can interact with rich toolsets.
  arXiv  Detail & Related papers  (2026-02-10T18:55:41Z)
• Scaling Environments for LLM Agents in the Era of Learning from Interaction: A Survey [30.673419015614233]
  A growing consensus is that agents should interact directly with environments and learn from experience through reinforcement learning.<n>We formalize this iterative process as the Generation-Execution-Feedback (GEF) loop, where environments generate tasks to challenge agents, return observations in response to agents' actions during task execution, and provide evaluative feedback on rollouts for subsequent learning.<n>Under this paradigm, environments function as indispensable producers of experiential data, highlighting the need to scale them toward greater complexity, realism, and interactivity.
  arXiv  Detail & Related papers  (2025-11-12T12:56:25Z)
• Grounded Test-Time Adaptation for LLM Agents [75.62784644919803]
  Large language model (LLM)-based agents struggle to generalize to novel and complex environments.<n>We propose two strategies for adapting LLM agents by leveraging environment-specific information available during deployment.
  arXiv  Detail & Related papers  (2025-11-06T22:24:35Z)
• Discovering Minimal Reinforcement Learning Environments [24.6408931194983]
  Reinforcement learning (RL) agents are commonly trained and evaluated in the same environment. Humans often train in a specialized environment before being evaluated, such as studying a book before taking an exam.
  arXiv  Detail & Related papers  (2024-06-18T13:19:26Z)
• DRED: Zero-Shot Transfer in Reinforcement Learning via Data-Regularised Environment Design [11.922951794283168]
  In this work, we investigate how the sampling of individual environment instances, or levels, affects the zero-shot generalisation (ZSG) ability of RL agents. We discover that for deep actor-critic architectures sharing their base layers, prioritising levels according to their value loss minimises the mutual information between the agent's internal representation and the set of training levels in the generated training data. We find that existing UED methods can significantly shift the training distribution, which translates to low ZSG performance. To prevent both overfitting and distributional shift, we introduce data-regularised environment design (D
  arXiv  Detail & Related papers  (2024-02-05T19:47:45Z)
• AACC: Asymmetric Actor-Critic in Contextual Reinforcement Learning [13.167123175701802]
  This paper formalizes the task of adapting to changing environmental dynamics in Reinforcement Learning (RL) We then propose the Asymmetric Actor-Critic in Contextual RL (AACC) as an end-to-end actor-critic method to deal with such generalization tasks. We demonstrate the essential improvements in the performance of AACC over existing baselines experimentally in a range of simulated environments.
  arXiv  Detail & Related papers  (2022-08-03T22:52:26Z)
• Learning Representations that Support Robust Transfer of Predictors [5.65658124285176]
  We introduce a robust estimation criterion -- transfer risk -- that is specifically geared towards optimizing transfer to new environments. Although inspired by IRM, we show that transfer risk serves as a better out-of-distribution generalization criterion.
  arXiv  Detail & Related papers  (2021-10-19T13:00:37Z)
• When Is Generalizable Reinforcement Learning Tractable? [74.87383727210705]
  We study the query complexity required to train RL agents that can generalize to multiple environments. We introduce Strong Proximity, a structural condition which precisely characterizes the relative closeness of different environments. We show that under a natural weakening of this condition, RL can require query complexity that is exponential in the horizon to generalize.
  arXiv  Detail & Related papers  (2021-01-01T19:08:24Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.