Can Agents Generalize to the Open World? Unveiling the Fragility of Static Training in Tool Use
For researchers building LLM agents, this work highlights the fragility of static training and provides a diagnostic framework and mitigation strategy for open-world generalization.
The paper formalizes the OpenAgent problem setting to study how LLM agents fail under distributional shifts in real-world tool use, finding that both SFT and RL agents suffer significant performance degradation. They propose Perturbation-Augmented Fine-Tuning to improve robustness.
While Large Language Model (LLM) agents demonstrate proficiency in static benchmarks, their deployment in real-world scenarios is hindered by the dynamic nature of user queries, tool sets, and interaction dynamics. To address this generalization gap, we formalize OpenAgent (Tool-Use Agent in Open-World), a problem setting characterized by distributional shifts across query, action, observation, and domain dimensions. To systematically diagnose its impact, we construct a controlled sandbox environment where we define fine-grained environmental shifts across a four-tier hierarchy, Perception, Interaction, Reasoning, and Internalization, and conduct a comprehensive series of experiments. Our analysis yields a series of key insights, demonstrating that agents trained via both Supervised Fine-Tuning(SFT) and Reinforcement Learning suffer from varying degrees of performance degradation when confronting open environmental shifts. Building on these insights, we propose Perturbation-Augmented Fine-Tuning, a disturbance-based intervention strategy for SFT that lays the foundation for enhancing agent robustness and utility in realistic environments. Our code will be released at: https://github. com/LAMDA-NeSy/OpenAgent.