ALOHA Bimanual Teleoperation Dataset
Download ALOHA bimanual teleoperation datasets for robot learning. Sim and real-world manipulation demos in LeRobot format. Apache 2.0 licensed. Free to use.
Open-source bimanual manipulation demonstrations from the low-cost ALOHA platform. Sim and real-world tasks in LeRobot format, Apache 2.0 licensed.
Apache 2.0 -- Open LeRobot / Parquet + MP4 Bimanual 2x 6-DOF
Key Stats
| Metric | Value |
|---|---|
| Robot | ALOHA bimanual (2x 6-DOF + grippers), stationary and mobile variants |
| Tasks | Cube transfer, peg insertion, cup opening, cabinet interaction, coffee making |
| Total size | ~4 GB across all variants |
| Format | LeRobot (Parquet + MP4 video) |
| Modalities | RGB (4 cameras), joint positions, gripper state, actions; mobile base state for mobile variant |
| License | Apache 2.0 |
| Downloads | 27K+ (sim transfer cube alone) |
What is the ALOHA dataset?
ALOHA (A Low-cost Open-source Hardware Assembly) is a bimanual teleoperation platform developed at Stanford by Tony Zhao et al. The project released both the hardware design and a suite of demonstration datasets that have become standard benchmarks for bimanual imitation learning.
The dataset collection spans multiple task variants:
- Sim Transfer Cube (Human): The most downloaded variant (27K downloads). Human-teleoperated cube transfer in simulation with 4 camera views.
- Sim Insertion (Human): Bimanual peg insertion demonstrations in simulation (17.4K downloads).
- Static Cups Open: Real-world demonstrations of cup opening with stationary ALOHA.
- Mobile Cabinet: Mobile ALOHA performing cabinet manipulation, the primary example dataset in the LeRobot README.
All datasets use the standardized LeRobot format (Parquet tables + MP4 video), making them directly compatible with the LeRobot training pipeline, ACT (Action Chunking with Transformers), and diffusion policy implementations.
How to download
# Install dependencies
pip install lerobot datasets
# Load the most popular variant
from datasets import load_dataset
ds = load_dataset("lerobot/aloha_sim_transfer_cube_human")
# Or use the LeRobot API
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
dataset = LeRobotDataset("lerobot/aloha_sim_transfer_cube_human")
# Available variants:
# lerobot/aloha_sim_transfer_cube_human
# lerobot/aloha_sim_transfer_cube_scripted
# lerobot/aloha_sim_insertion_human
# lerobot/aloha_sim_insertion_scripted
# lerobot/aloha_static_cups_open
# lerobot/aloha_mobile_cabinet
RCSV quality assessment
- Strengths: Clean bimanual demonstrations, well-structured LeRobot format, 4-camera coverage, open hardware design enables reproduction.
- Limitations: Simulation variants are low-fidelity compared to real-world. Real-world datasets are smaller. Limited task diversity within each variant.
- Best for: ACT and diffusion policy benchmarking, bimanual manipulation research, LeRobot pipeline validation.