[论文] DreamerAD: Latent World Models with Shortcut Forcing for Efficient Aut...

## 论文概要 **研究领域**: Robotics **作者**: Pengxuan Yang, Yupeng Zheng, Deheng Qian, Zebin Xing, Qichao Zhang... **发布时间**: 2026-03-25 **arXiv**: [2603.24587](https://arxiv.org/abs/2603.24587) ## 中文摘要 本文提出DreamerAD，首个潜在世界模型框架，通过将扩散采样从100步压缩到1步，实现80倍加速同时保持视觉可解释性，从而支持高效的自动驾驶强化学习。在真实驾驶数据上训练RL策略会产生高昂的成本和安全风险。虽然现有的像素级扩散世界模型支持安全的基于想象的训练，但它们遭受多步扩散推理延迟（2秒/帧）的困扰，阻碍了高频RL交互。 ## 原文摘要 We introduce DreamerAD, the first latent world model framework that enables efficient reinforcement learning for autonomous driving by compressing diffusion sampling from 100 steps to 1 - achieving 80x speedup while maintaining visual interpretability. Training RL policies on real-world driving data incurs prohibitive costs and safety risks. While existing pixel-level diffusion world models enable safe imagination-based training, they suffer from multi-step diffusion inference latency (2s/frame) that prevents high-frequency RL interaction. --- *自动采集于 2026-03-27* #论文 #arXiv #Robotics #小凯