# 用于鲁棒多智能体驾驶策略的不确定性驱动自博弈

> 原文链接：https://robotics.hellobike.com/papers/2026-10-14-2610-19714
> 频道：papers | 更新日期：2026-10-14 | 标签：arxiv, cs.RO
> 核心摘要：提出以认知不确定性引导的自博弈，让探索智能体挖掘难例，并在nuplan Val14、Argoverse 2和Waymo验证。

# 用于鲁棒多智能体驾驶策略的不确定性驱动自博弈

::: v-pre
> 原文链接：[Uncertainty-Driven Self-Play for Robust Multi-Agent Driving Policies](https://arxiv.org/abs/2610.19714)（arXiv 2610.19714 · 发布日期：2026-10-14）

**Uncertainty-Driven Self-Play for Robust Multi-Agent Driving Policies**

提出以认知不确定性引导的自博弈，让探索智能体挖掘难例，并在nuplan Val14、Argoverse 2和Waymo验证。

## 核心要点

- 鲁棒驾驶策略难以泛化到训练域外，困难场景构建也常需人工设计。
- 方法让半数智能体提高邻居认知不确定性，另半数学习安全驾驶。
- 无需微调时，在nuplan Val14和Argoverse 2零样本迁移优于标准自博弈，并提升Waymo表现。
- LSTM隐藏状态覆盖率较基线高85%，高不确定智能体合并失败率为低不确定者14.8倍。

**验证范围与局限**：验证范围为多智能体驾驶策略在nuplan Val14、Argoverse 2和Waymo上的评估。

## 英文摘要 (Abstract)

Learning robust driving policies that generalize beyond their training domain remains a central challenge in motion planning. We propose an uncertainty-driven self-play scheme, in which half of the agents are rewarded for increasing their neighbors' epistemic uncertainty, while the other half learn to drive safely as the exploring agents continuously seek out their weaknesses. Epistemic uncertainty indicates where the policy is least confident and serves as an automatic proxy for scenario difficulty. Exploring agents use this signal to seek out hard cases, generating a challenging curriculum without manual scenario design. The resulting policy outperforms standard self-play, achieving best-in-class zero-shot transfer to nuplan Val14 (99.12% vs. 97.97%) and Argoverse 2 (82.39% vs. 78.03%) without fine-tuning, while also improving performance over standard self-play on Waymo. Our diversity-driven approach achieves 85% higher LSTM hidden-state coverage than the baseline, indicating a much richer set of behaviors. Further analysis shows that epistemic uncertainty serves as a calibrated difficulty detector, with highest-uncertain agents exhibiting 14.8 times the combined failure rate across collision, off-road, and DNF, of lowest-uncertain agents, enabling automatic prioritization of challenging scenarios without ground-truth labels.

作者：Sajad Marvi, Julian Schmidt, Alessandro Canevaro, Julian Jordan, Abhinav Valada
:::

## 原始参考来源 (Sources)

- [arXiv 2610.19714：Uncertainty-Driven Self-Play for Robust Multi-Agent Driving Policies](https://arxiv.org/abs/2610.19714)