ALOHA

A four-arm bimanual teleoperation and demonstration-recording system with a separate learning stack. [S001]

This ALOHA repository provides teleoperation and data collection for four arms and four cameras. The hardware assembly tutorial is a separate document, and training imitation-learning models requires the separate ACT project. Copying this repository alone does not preserve a complete robot build or learning environment. [S001]

The upstream guide calls for at least six USB 3 ports: four direct connections for the robots and two for cameras. It reports unstable multi-arm connections on some hosts, rejects hubs and extensions for the robot links, and limits cameras to two per hub. These are practical host-selection constraints even before training compute is considered. [S001]

Follow the device-identity section when planning the setup: fixed aliases distinguish left/right leader and follower arms and the four cameras. The pinned guide reports Ubuntu 18.04 and 20.04 with ROS 1 Noetic and specifies a Python 3.8.10 Conda environment; these historical combinations have not been reproduced here. ROS 2 compatibility is described as ongoing work. Interbotix software is another required external dependency, including a documented change to its arm.py file to avoid forward-kinematics computation delaying teleoperation; see the upstream software-installation section before freezing that dependency. It also warns that torque-off arms can collapse because the joints have no automatically engaged brakes. [S001]

The software is MIT-licensed; that does not establish rights to manufacture every purchased arm component. Current full-setup cost, dependency compatibility and our support effort remain unknown. We have not run teleoperation, recorded an episode or tested this hardware. [S003]

Official website · Source code