发表机构
Valeo Brain; Sorbonne Université, CNRS, ISIR（法雷奥智能大脑公司; 索邦大学、法国国家科学研究中心、智能系统与机器人研究所）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究针对驾驶模拟中自博弈存在的问题，提出建立视角自博弈训练机制，介绍GPU加速模拟器Pictura，通过普通PPO自博弈训练出Alberti策略，无需特权观测，性能接近且能零样本迁移优于特权矢量化策略。
AI 中文摘要
在模拟中进行自博弈可大规模生成稳健的驾驶策略。以往此类行为演示使用了如精确姿态和速度等有特权的矢量化观测，即便针对被遮挡的智能体。这假定感知问题已解决，且与从以自我为中心的相机视角进行部署的智能体的部分观测存在表征差距。常见解决方法是将有特权的策略提炼为相机输入的学生策略，但会使学生模仿自身视角无法证明合理的决策。我们建立了视角自博弈作为一种实用训练机制。引入Pictura，这是一个GPU加速的多智能体驾驶模拟器，在每一步渲染每个智能体的以自我为中心的视角，从源头减轻表征差距。Pictura在单个H100上每秒可维持高达500K智能体步（2M图像/秒）。使用Pictura，我们通过普通PPO自博弈训练了Alberti。它是首个直接从视角图像训练的大规模驾驶自博弈策略，无需特权观测。训练跨越50B智能体步，行驶约35M公里。其性能接近有特权的矢量化对应策略，并能零样本迁移到在Pictura中重新渲染的Waymo Open Motion数据集布局，且在此处优于有特权的矢量化智能体。
英文摘要
Self-play in simulation produces robust driving policies at scale. Demonstrations of such behavior have been made using privileged vectorized observations such as exact poses and velocities, even for occluded agents. This assumes that perception is solved and introduces a representation gap with the partial observation of a deployed agent driving from the perspective view of egocentric cameras. A common fix, distilling the privileged policy into a camera-input student, leaves the student imitating decisions its own view cannot justify. Instead, we establish perspective-view self-play as a practical training regime. We introduce Pictura, a GPU-accelerated multi-agent driving simulator that renders each agent's egocentric view at every step, mitigating the representation gap at its source. Pictura sustains up to 500K agent-steps/s (2M images/s) on a single H100. Using Pictura, we train Alberti by self-play with plain PPO. It is the first large-scale driving self-play policy trained directly from perspective images, without privileged observations. Training spans 50B agent steps for ~35M km of driving. It approaches the driving performance of its privileged vectorized counterpart, and transfers zero-shot to Waymo Open Motion Dataset layouts re-rendered in Pictura, where it outperforms privileged vectorized agents. Project page: https://valeoai.github.io/Pictura/
