MobileVLA-CoT
收藏资源简介:
MobileVLA-CoT 是一个大规模、多粒度的链式思维(CoT)数据集,专门用于具身轨迹推理。该数据集来源于论文《MobileVLA-R1: Reinforcing Vision-Language-Action for Mobile Robots》,旨在为四足机器人提供结构化的推理监督,将自然语言指令对齐到连续控制中。数据集包含丰富的多粒度推理步骤,支持机器人从高层任务规划到低层动作执行的端到端学习。
MobileVLA-CoT is a large-scale, multi-granularity Chain-of-Thought (CoT) dataset specifically designed for embodied trajectory reasoning. Derived from the paper MobileVLA-R1: Reinforcing Vision-Language-Action for Mobile Robots, it aims to provide structured reasoning supervision for quadruped robots, aligning natural language instructions with continuous control. The dataset contains rich multi-granularity reasoning steps, supporting end-to-end learning from high-level task planning to low-level action execution.
MobileVLA-CoT 数据集总结
MobileVLA-CoT 是一个由北京大学计算机学院唐浩团队与北京智源人工智能研究院(BAAI)联合发布的大规模机器人数据集,专注于为具身轨迹提供多粒度思维链(Chain-of-Thought, CoT)标注。该数据集源自论文《MobileVLA-R1: Reinforcing Vision-Language-Action for Mobile Robots》,旨在为四足机器人提供从自然语言指令到连续控制的结构化推理监督。
核心特性
- 任务类型:机器人(Robotics)
- 数据规模:大规模(Large-scale)
- 数据内容:多粒度思维链(CoT)标注的具身轨迹(Embodied Trajectories)
- 核心目的:为四足机器人(Quadruped Robots)的自然语言指令到连续控制(Continuous Control)的映射提供结构化推理监督
相关链接
- 论文:MobileVLA-R1: Reinforcing Vision-Language-Action for Mobile Robots
- 项目主页:https://aigeeksgroup.github.io/MobileVLA-R1
- 代码仓库:https://github.com/AIGeeksGroup/MobileVLA-R1




