ABC-130k

UC Berkeley / MIT / Amazon FAR / CMU / XDOF · 2026 · datasets.bot · datasets.bot page

One-liner. Largest open bimanual teleoperation dataset: ~131K episodes, ~3,600 hours across ~195 manipulation tasks.

Setup

Datasets / benchmarks: ABC-130k is an open bimanual robot manipulation dataset released with the paper 'Scalable Behavior Cloning with Open Data, Training, and Evaluation'. It contains roughly 130,919 episodes (43,090 annotated) totaling about 3,598 hours collected on bimanual YAM stations with two 6-DoF arms and parallel-jaw grippers, spanning ~195 tasks such as pick-and-place, folding, sorting, handover, insertion, tool use, and assembly. Each episode includes three camera views (one top plus two wrist), joint states, end-effector poses, gripper aperture, and task/subtask annotations, plus 400 hours of accompanying sim-teleop data. License: Apache-2.0. Download: https://huggingface.co/datasets/XDOF/ABC-130k.
Hardware / simulator: Embodiment: parallel_jaw, aloha. Environment: lab, simulation, tabletop. Realness: both.

episodes -> steps -> {states_actions (14 state + 14 action), combined_camera-images-rgb.mp4, metadata}