Berkeley Fanuc Manipulation
收藏资源简介:
该数据集可用于基于视觉的机器人模仿学习策略、微调视觉表示模型、训练根据任务指令生成机器人动作的生成模型等。它是用 FANUC Mate 200iD 机器人收集的,两个 RGB 视频(第三-提供了人视图和以自我为中心的视图)、机器人轨迹(关节和动作值)和语言指令。右 数据集格式。格式如下。 状态 主摄像头(固定第三人称视角,224 x 224 x 3) 腕式相机(以自我为中心的视图,224 x 224 x 3) 本体感受信息(关节姿势、抓手状态、关节速度) 语言指导(针对任务) 行动 笛卡尔空间动作
This dataset can be used for vision-based robotic imitation learning strategies, fine-tuning visual representation models, training generative models that generate robotic actions based on task instructions, and other relevant applications. It was collected using a FANUC Mate 200iD robot, and includes two RGB videos (a third-person view and an egocentric view), robot trajectories (joint positions and action values), as well as task-specific language instructions. The dataset format is as follows: Status Main camera (fixed third-person perspective, 224 × 224 × 3) Wrist camera (egocentric view, 224 × 224 × 3) Proprioceptive information (joint poses, gripper state, joint velocities) Task-specific language instructions Actions Cartesian-space actions




