Overview

ALOHA 2 is the second-generation, open-source reference platform for A Low-cost Open-source Hardware System for Bimanual Teleoperation, originating from Stanford and Google DeepMind and commercialized as buildable kits by Trossen Robotics. The station is a stationary tabletop setup with two 6-DOF leader arms (WidowX-class) and two 6-DOF follower arms (ViperX-class); an operator back-drives the leaders to puppeteer the followers, capturing demonstrations for imitation learning and vision-language-action policies. Version 2 improvements over the original ALOHA include a redesigned low-friction rail gripper for both leader and follower, passive gravity compensation, more repeatable joints, and a switch to the compact Intel RealSense D405 depth cameras. A high-fidelity MuJoCo simulation model and open hardware make it a common benchmark platform for scaled robot-learning data collection.

Highlights

• Enhanced open-source reference design for bimanual teleoperation; sold as buildable kits by Trossen Robotics
• 4 arms total: 2 leader + 2 follower, 6 DOF each, leader-follower puppeteering
• Low-friction rail grippers (redesigned in v2), passive gravity compensation, more repeatable joints
• Intel RealSense D405 depth cameras (overhead + worms-eye + per-wrist mounts)
• Ships with high-fidelity MuJoCo model; widely used for scaled imitation/VLA data collection
• Basis for Google DeepMind Gemini Robotics demos and large imitation-learning datasets
• Approx. kit cost USD 25,000-40,000 depending on configuration

Specifications