ALOHA 2

ALOHA 2 is an enhanced version of the low-cost ALOHA hardware platform for bimanual teleoperation and manipulation. It was developed by the Google DeepMind team together with researchers from Stanford (IRIS Lab) and builds on the original ALOHA design. The platform aims to accelerate robot-learning research by enabling large-scale collection of manipulation demonstrations.

Compared with the original, ALOHA 2 introduces concrete improvements: new low-friction rail grippers for both leader and follower (better teleoperation ergonomics, lower latency and higher gripper force), a passive gravity-compensation mechanism built from off-the-shelf parts (more durable than the original rubber-band system), a simplified frame made from 20x20 mm aluminium extrusions, and Intel RealSense D405 cameras (smaller, with a wider field of view, depth and a global shutter).

The system works in a leader–follower setup: an operator moves the leader arms and the follower arms mirror the motion. It uses 6-DOF Trossen Interbotix arms (WidowX 250S as leader, ViperX 300S as follower) and Dynamixel servos. All hardware designs are open-sourced with a detailed tutorial, along with an accurate MuJoCo model (MuJoCo Menagerie) with system identification. ALOHA 2 kits are commercially available from Trossen Robotics.