Mobile ALOHA
Stanford University · 2024 · datasets.bot · datasets.bot page
One-liner. Stanford mobile bimanual manipulation demos (kitchen/home tasks) on the Mobile ALOHA robot; RLDS in Open X / TFDS.
Setup
- Datasets / benchmarks: Mobile ALOHA is a teleoperated mobile bimanual manipulation dataset collected by Fu, Zhao, and Finn (Stanford) using the low-cost whole-body Mobile ALOHA system, with ~50 human demonstrations per task across whole-body tasks such as sauteing shrimp, opening a two-door cabinet, calling/entering an elevator, and rinsing a pan. The TFDS/Open X release contains 276 episodes with 3 RGB cameras (overhead + two wrist cameras at 480x640), a 14-dim state, a 16-dim action, and per-step language instructions. Raw HDF5 demonstrations are also distributed via the project's Google Drive. License: CC-BY-4.0. Download: https://www.tensorflow.org/datasets/catalog/aloha_mobile.
- Hardware / simulator: Embodiment: aloha. Environment: home, lab, kitchen. Realness: physical.
Schema
episodes -> steps -> {observation: {cam_high, cam_left_wrist, cam_right_wrist (480x640x3), state(14)}, action(16), language_instruction, reward, discount, is_first/is_last/is_terminal}
Links