Robotics datasets and Physical AI models

Search robotics datasets and Physical AI models, including LeRobot and vision-language-action (VLA). Compare robots, tasks, sensors, scale, and licenses.

Find data for robot manipulation, embodied AI, and vision-language-action research. Compare robot types, camera streams, episode counts, and license metadata in one catalog.

Popular robotics datasets

  • IPEC-COMMUNITY/language_table_lerobot

    This dataset contains 442,226 episodes of xArm robot manipulation tasks on a language table, with RGB camera observations, state/action data, and natural language task instructions, formatted in the LeRobot v2.0 codebase.

  • IPEC-COMMUNITY/droid_lerobot

    This dataset contains 92,233 episodes of Franka robot manipulation tasks with 31,308 task instructions, recorded at 15 fps with three RGB cameras (two exterior, one wrist), including state and action data, suitable for imitation learning an

  • IPEC-COMMUNITY/kuka_lerobot

    This dataset contains 209,880 episodes of KUKA IIWA robot teleoperation data for a single task 'pick anything', with RGB camera observations, state, and action data at 10 fps, formatted in the LeRobot v2.0 codebase.

  • cadene/droid_1.0.1

    The DROID 1.0.1 dataset contains 95,600 episodes of Franka robot manipulation tasks in diverse real-world environments, with 3 RGB cameras, 8-DoF state/action data, and language instructions.

  • IPEC-COMMUNITY/bridge_orig_lerobot

    This dataset contains 53,192 episodes of real-world robot manipulation data collected from a WidowX robot, featuring 4 RGB cameras, 8-dimensional state, and 7-dimensional actions, with natural language task instructions covering diverse hou

  • cadene/droid

    The DROID dataset is a large-scale, in-the-wild robot manipulation dataset collected with Franka robots, containing over 27 million frames across 92,223 episodes with natural language task instructions, and is provided in the LeRobot format