ALOHA 2: the open bimanual platform for advanced imitation learning
In one sentence Stanford and Berkeley release ALOHA 2, the commercial version of the teleoperated bimanual system used to collect ACT and Diffusion Policy datasets for tasks like cooking and surgery.
ALOHA (A Low-cost Open-source Hardware System for Bimanual Teleoperation) had already become the reference system in robotics labs for collecting human demonstrations on complex bimanual tasks. ALOHA 2 is the commercially available version, with improved construction and complete documentation to allow more labs to replicate it.
The system lets a human operator teleoperate two robot arms simultaneously with haptic feedback, collecting high-quality demonstrations for tasks like cooking, assembling delicate components, or performing simplified surgical procedures.
This data is then used to train policies via ACT (Action Chunking with Transformers) or Diffusion Policy, yielding robots that autonomously execute the learned tasks.
Companies
Stanford University, UC Berkeley
Tools
ALOHA 2, ACT, Diffusion Policy, Trossen Robotics
Tags
Sources