---
title: "Pictura：大规模透视视角自博弈驾驶策略"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.26005v1"
permalink: "https://feed.astralor.com/items/itm_cbca33f2865600f74eb98f74"
published: "2026-07-28T17:20:39.000Z"
score: 81
channel: "ai"
tags: ["Alberti","Pictura","PPO","自动驾驶","自博弈","透视图像"]
---

# Pictura：大规模透视视角自博弈驾驶策略

## AI 点评

> 这篇论文真正做到了‘所见即所得’——直接让智能体从自己的摄像头视角学习，而不是依赖特权信息蒸馏。对于从事自动驾驶仿真与强化学习的同行来说，Pictura模拟器的高吞吐量和Alberti的零样本迁移能力都极具参考价值。尤其是，它证明了纯视觉自博弈在合理计算成本下就能达到甚至超越特权方法的性能，这可能会改变后续驾驶策略训练的设计思路。

## 速览

Pictura是一个GPU加速的多智能体驾驶模拟器，在单个H100上每秒支持50万智能体步（200万图像）。利用该模拟器，研究者通过普通PPO算法从透视图像直接进行自博弈训练，得到名为Alberti的驾驶策略。训练规模达500亿智能体步（约3500万公里驾驶里程），无需任何特权观测（如精确位姿、速度）。Alberti的性能接近使用特权向量观测的同类策略，并在Waymo Open Motion Dataset的重新渲染布局上实现零样本迁移，表现甚至优于特权向量智能体。该工作弥合了仿真训练与真实部署之间的表征鸿沟。

## 关键点

- Pictura模拟器在H100上实现50万智能体步/秒的吞吐量，每步为每个智能体渲染透视视图。
- Alberti策略仅使用透视图像和PPO算法进行自博弈训练，训练规模达500亿步，约3500万公里。
- Alberti零样本迁移至Waymo Open Motion Dataset布局，性能超越使用特权信息的向量智能体。
- 该方法直接从传感器输入学习，消除了特权蒸馏带来的表征差距。

## 全文

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/

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_cbca33f2865600f74eb98f74