EvoTool: Self-Evolving Tool-Use Policy Optimization in LLM Agents via Blame-Aware Mutation and Diversity-Aware Selection
#EvoTool #tool-use policy #LLM agents #self-evolving #blame-aware mutation #diversity-aware selection #optimization
📌 Key Takeaways
- EvoTool introduces a self-evolving framework for optimizing tool-use policies in LLM agents.
- It employs blame-aware mutation to identify and correct ineffective tool-use strategies.
- The method uses diversity-aware selection to maintain a variety of effective policies and prevent stagnation.
- This approach aims to enhance the adaptability and efficiency of LLM agents in complex tasks.
📖 Full Retelling
arXiv:2603.04900v1 Announce Type: new
Abstract: LLM-based agents depend on effective tool-use policies to solve complex tasks, yet optimizing these policies remains challenging due to delayed supervision and the difficulty of credit assignment in long-horizon trajectories. Existing optimization approaches tend to be either monolithic, which are prone to entangling behaviors, or single-aspect, which ignore cross-module error propagation. To address these limitations, we propose EvoTool, a self-e
🏷️ Themes
AI Optimization, LLM Agents
Entity Intersection Graph
No entity connections available yet for this article.
Original Source
--> Computer Science > Artificial Intelligence arXiv:2603.04900 [Submitted on 5 Mar 2026] Title: EvoTool: Self-Evolving Tool-Use Policy Optimization in LLM Agents via Blame-Aware Mutation and Diversity-Aware Selection Authors: Shuo Yang , Soyeon Caren Han , Xueqi Ma , Yan Li , Mohammad Reza Ghasemi Madani , Eduard Hovy View a PDF of the paper titled EvoTool: Self-Evolving Tool-Use Policy Optimization in LLM Agents via Blame-Aware Mutation and Diversity-Aware Selection, by Shuo Yang and 5 other authors View PDF HTML Abstract: LLM-based agents depend on effective tool-use policies to solve complex tasks, yet optimizing these policies remains challenging due to delayed supervision and the difficulty of credit assignment in long-horizon trajectories. Existing optimization approaches tend to be either monolithic, which are prone to entangling behaviors, or single-aspect, which ignore cross-module error propagation. To address these limitations, we propose EvoTool, a self-evolving framework that optimizes a modular tool-use policy via a gradient-free evolutionary paradigm. EvoTool decomposes agent's tool-use policy into four modules, including Planner, Selector, Caller, and Synthesizer, and iteratively improves them in a self-improving loop through three novel mechanisms. Trajectory-Grounded Blame Attribution uses diagnostic traces to localize failures to a specific module. Feedback-Guided Targeted Mutation then edits only that module via natural-language critique. Diversity-Aware Population Selection preserves complementary candidates to ensure solution diversity. Across four benchmarks, EvoTool outperforms strong baselines by over 5 points on both GPT-4.1 and Qwen3-8B, while achieving superior efficiency and transferability. The code will be released once paper is accepted. Comments: Work under review, 9 pages, 5 figures Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2603.04900 [cs.AI] (or arXiv:2603.04900v1 [cs.AI] for this version) https://doi.org/10.48550/a...
Read full article at source