外观
论文HF Daily Papers · arXiv 2610.37381昨天
PDRVLA引入物理动态模块,VLA模型仿真成功率达99.3%
本文由 AI 根据公开内容整理,可能有误,请以原文为准:PDRVLA: A Vision-Language-Action Model with Physical Dynamics Representations(HF Daily Papers · arXiv 2610.37381)
核心内容
针对视觉-语言-动作(VLA)模型物理动力学理解受限及视频世界模型在线生成导致控制回路高延迟的问题,研究团队提出PDRVLA架构。该模型创新性地引入轻量级物理动态模块(Physical Dynamics Module),直接从视觉-语言特征中提取物理Token以条件化动作生成,完全规避了在线视频生成步骤。在LIBERO、LIBERO-Plus、RoboTwin 2.0及SimplerEnv四个仿真基准测试中,PDRVLA分别取得99.3%、89.9%、93.6%和81.5%的成功率,显著提升了具身智能体在复杂任务中的执行效能。
技术上,物理动态模块通过直接预测未来RGB、深度图及光流特征进行学习,并利用冻结的预训练视频世界模型提供辅助监督,确保物理表征准确性而不增加推理负担。训练数据完全基于公开数据集,并通过人机合成、子任务标注及具身思维链(ECoT)注解进行增强。消融实验证实,世界模型监督有效提升了未来状态预测精度,且子任务条件化、未来物体位姿监督与ECoT训练策略均对动作成功率产生正向影响,为低延迟具身控制提供了新范式。
关键看点
- 架构创新:设计轻量物理动态模块提取物理Token,替代在线视频生成,解决VLA控制回路高延迟瓶颈
- 性能指标:在LIBERO、RoboTwin 2.0等四大仿真基准测试中,最高成功率达99.3%,验证高效决策能力
- 训练机制:采用冻结视频世界模型辅助监督,结合RGB/深度/光流预测学习物理动态,无需额外推理开销
- 数据增强:利用人机合成、子任务标注及具身思维链(ECoT)扩展公开数据,提升模型泛化与逻辑推理水平
工程与产业观察
PDRVLA证明了在不依赖高算力视频生成的前提下,通过隐式物理表征与多模态特征对齐即可实现高精度具身控制。该方案为端侧部署VLA模型提供了低延迟工程路径,其ECoT与物理Token结合的训练范式,对解决Sim-to-Real迁移中的动力学一致性难题具有重要参考价值。
英文摘要 (Abstract)
Vision--Language--Action (VLA) models generate robot actions from visual observations and language instructions, but their understanding of physical dynamics remains limited. Video world models can capture scene dynamics, yet generating future videos introduces high latency into the control loop. We introduce PDRVLA, whose lightweight Physical Dynamics Module extracts physical tokens from vision--language features to condition action generation without online video generation. The module learns through direct prediction of future RGB, depth, and optical-flow features, with auxiliary supervision through a frozen pretrained video world model. Training relies entirely on publicly available data, expanded through human-to-robot synthesis, subtask annotation, and embodied chain-of-thought (ECoT) annotation. Across four simulation benchmarks, PDRVLA achieves success rates of 99.3% on LIBERO, 89.9% on LIBERO-Plus, 93.6% on RoboTwin~2.0, and 81.5% on SimplerEnv. Ablations show improved future prediction with world-model supervision and evaluate how subtask conditioning, future object-pose supervision, and ECoT training affect action success.
Hugging Face Daily Papers 热度:130 upvotes 作者:Hanqing Wang, Zhaotian Cai, Ruichuan An, Bohan Zeng, Tao Zhang, yilin chen, Jiang Xu, Kaiquan Chen, Chongyang Wu, Haoran Li, FanYang, Mingan Lin, Wentao Zhang, Zenan Zhou, Youzheng Wu
来源
本页事实对应的公开页面,以原文为准。