MolmoAct2 Bimanual YAM Dataset
Allen Institute for AI (Ai2) · 2026 · datasets.bot · datasets.bot page
One-liner. Ai2 bimanual YAM-arm tabletop manipulation: 32k episodes, 720+ hours, 34 tasks, 3 camera views.
Setup
- Datasets / benchmarks: A large-scale bimanual manipulation dataset collected by Ai2 for MolmoAct2 using paired YAM follower arms. It contains 32,246 episodes (76M frames) spanning 34 tabletop tasks with 720+ hours of teleoperated demonstrations. Each episode provides three RGB camera views (left, right, top) plus 14-dim joint/gripper states, in LeRobot format with per-episode annotated language instructions. License: Apache-2.0. Download: https://huggingface.co/datasets/allenai/MolmoAct2-BimanualYAM-Dataset.
- Hardware / simulator: Embodiment: aloha. Environment: tabletop. Realness: physical.
Schema
LeRobot: 3 RGB videos (left/right/top, 360x640, 30fps) + 14-dim joint/gripper state and action; per-episode language instructions.
Links