Berkeley AUTOLab Datasets

## Berkeley AUTOLab Datasets Berkeley AUTOLab (Automation Laboratory) at UC Berkeley is a leading robotics research lab directed by Professor Ken Goldberg. The lab has produced multiple widely-used robot manipulation datasets collected on various robot platforms. ### Key Datasets - **berkeley_autolab_ur5**: A dataset of 1,000 episodes (~98K frames) collected on a UR5 robot arm, covering 5 different tasks. Part of the Open X-Embodiment dataset collection. - **berkeley_cable_routing**: Cable routing manipulation dataset on a UR5 robot arm. - **berkeley_fanuc_manipulation**: Dataset on a Fanuc robot arm, also part of Open X-Embodiment. ### Characteristics - **Modality**: RGB images, robot proprioception (joint positions, velocities), end-effector actions - **Robots**: UR5, Fanuc, and other arms - **Collection Method**: Human teleoperation - **Licenses**: CC-BY-4.0 or Apache-2.0 depending on the specific dataset variant The Berkeley AUTOLab datasets are widely used in the robotics community for benchmarking imitation learning algorithms and cross-embodiment transfer learning.