用密集具身思维链训练跨机器人视觉语言动作模型，实现零损失推理。
Training Vision-Language-Action Models with Dense Embodied Chain-of-Thought Supervision
- 通过双流架构让视觉语言模型生成密集具身思维链，对齐不同机器人的认知过程。
- 在6000万帧数据上训练，96.8%的帧有思维链标注，支持端到端学习。
- 可在仿真与真实机器人上零损失推理，适合多平台部署的智能控制研究者。
视觉语言动作（VLA）模型的跨机器人迁移仍具挑战，因低层状态与动作空间在不同机器人平台上存在本质差异。我们观察到，操控任务背后的高层认知过程——包括场景感知、物体识别、任务规划与子任务分解——在不同机器人间具有高度共性。基于此，我们提出ZR-0，一个26亿参数的端到端VLA模型，采用密集具身思维链（ECoT）监督来对齐视觉语言模型中的跨机器人表示。ZR-0采用双流架构：预训练的视觉语言模型（系统2）在训练中生成结构化ECoT推理，而基于扩散变换器的动作专家（系统1）通过流匹配生成连续动作块。两者通过交叉注意力耦合，且注意力掩码仅允许动作专家接收提示特征，使推理时可完全跳过ECoT生成而无性能损失。ZR-0在ProcCorpus-60M上预训练，该数据集包含约6000万帧（约1000小时），来自超过40万条轨迹，其中96.8%的帧带有密集ECoT标注。我们在三个仿真基准上评估：单臂（LIBERO）、双臂（RoboTwin 2.0）和人形机器人（RoboCasa GR-1 Tabletop），以及xArm平台的真实世界实验，均表现优异。代码与模型权重已公开于https://github.com/RUCKBReasoning/ZR-0。
原文摘要 · Abstract (English)
Cross-embodiment transfer in vision-language-action (VLA) models remains challenging because low-level state and action spaces differ fundamentally across robot platforms. We observe that the high-level cognitive process underlying manipulation, including scene perception, object identification, task planning, and sub-task decomposition, is largely shared across embodiments. Based on this observation, we present ZR-0, a 2.6 billion parameter end-to-end VLA model that uses dense Embodied Chain-of-Thought (ECoT) supervision to align cross-embodiment representations within the vision-language model (VLM). ZR-0 adopts a dual-stream architecture: a pre-trained VLM (System 2) generates structured ECoT reasoning during training, while a Diffusion Transformer-based action expert (System 1) produces continuous action chunks via flow matching. The two components are coupled through cross-attention, with an attention mask that restricts the action expert to input prompt features only, enabling ECoT generation to be entirely skipped at inference without any performance loss. ZR-0 is pre-trained on ProcCorpus-60M, a large-scale dataset comprising approximately 60 million frames (approximately 1,000 hours) from over 400K trajectories, with dense ECoT annotations covering 96.8% of all frames. We evaluate ZR-0 on three simulation benchmarks spanning single-arm (LIBERO), bimanual (RoboTwin 2.0), and humanoid (RoboCasa GR-1 Tabletop) embodiments, as well as real-world experiments on the xArm platform, demonstrating strong performance across all settings. Code and model checkpoints are available at https://github.com/RUCKBReasoning/ZR-0.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。