Mobile ALOHA

Imitation learning has advanced robotic performance, but most success has been limited to static, table-top tasks. Mobile ALOHA addresses this gap by enabling bimanual, whole-body teleoperation for mobile manipulation. The system integrates a mobile base and an enhanced teleoperation interface, allowing for more complex and dynamic tasks. Using data from Mobile ALOHA, co-training with existing datasets significantly improves task success rates by up to 90%. This allows robots to autonomously perform tasks like sautéing, opening cabinets, and calling an elevator. Mobile ALOHA's success shows promise for robots completing real-world tasks requiring mobility and dexterity.