Can Agents Generalize to the Open World? Unveiling the Fragility of Static Training in Tool Use
2026-07-01 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors study how Large Language Model agents struggle to handle changing situations in the real world, like different questions or tools than they saw during training. They created a special test environment to mimic these changes in stages related to perception, interaction, reasoning, and internal thinking. Their experiments show that common training methods lead to performance drops in new settings. To improve this, they suggest a new training method called Perturbation-Augmented Fine-Tuning that introduces controlled disturbances to make agents more robust. This work helps understand and reduce the gap between lab tests and real-world use for these AI agents.
Large Language ModelsOpen-World LearningDistributional ShiftSupervised Fine-TuningReinforcement LearningAgent RobustnessPerturbation-Augmented Fine-TuningEnvironmental ShiftsAI Interaction
Authors
Song-Lin Lv, Weiming Wu, Rui Zhu, Zi-Jian Cheng, Lan-Zhe Guo
Abstract
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.