arXiv cs.AI· Tong Nie, Yuewen Mei, Yihong Tang, Junlin He, Jie Deng, Jian Sun, Wei Ma·· 6 天前AI 评分24
EvoDrive：基于自改进 LLM 智能体的帕累托进化安全关键自动驾驶场景生成
EvoDrive: Pareto Evolution for Safety-Critical Autonomous Driving via Self-Improving LLM Agents
AI 导读
EvoDrive 是首个基于 LLM 智能体的多目标场景生成自动化进化框架，采用模拟器接地的 actor-critic 架构，由记忆驱动 actor 迭代改进生成器、critic 过滤不合理候选，并通过自进化世界评估器分配仿真预算。
正文 · 原文
Abstract:Generating safety-critical scenarios is essential for validating and improving autonomous driving systems, yet it inherently requires maximizing adversariality to expose failures while preserving realism. Existing methods usually manage this trade-off with handcrafted heuristics, confining generation to known priors and overlooking underexplored patterns. While recent open-ended agentic evolution can push this limit, unconstrained general agents lack strict simulator grounding and tend to collapse the multi-objective tension into single-scalar maximization. Here we present EvoDrive, the first automated, LLM-based agentic evolution framework for multi-objective scenario generation. EvoDrive employs a simulator-grounded actor-critic architecture where a memory-driven actor iteratively proposes improvements to the generators and critics filter out implausible candidates, and a self-evolving world evaluator routes promising proposals to optimize simulation budgets. EvoDrive further maintains a Pareto archive of evaluated candidates to preserve diverse attack-realism trade-offs and guide future evolution via simulation feedback. Benchmark results on MetaDrive and CARLA show that EvoDrive not only significantly expands the Pareto frontier across various generators, but also produces valuable scenarios for policy training.
| Comments: | Accepted at the 10th Conference on Robot Learning (CoRL 2026) |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2606.03678 [cs.AI] |
| (or arXiv:2606.03678v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2606.03678 arXiv-issued DOI via DataCite |
Submission history
From: Tong Nie [view email]
[v1]
Tue, 2 Jun 2026 14:01:23 UTC (5,199 KB)
[v2]
Fri, 9 Oct 2026 13:56:29 UTC (5,208 KB)
来源：arXiv cs.AI · arxiv.org