A new robot-learning framework can perform autonomous manipulation tasks at nearly the speed of a human after observing just 50 demonstrations. Developed by researchers including Chelsea Finn and Zipeng Fu, the system called Flash ALOHA has been open sourced, including its research paper, hardware specifications and code.
The system uses direct human guidance to collect training data, avoiding the speed limitations of traditional teleoperation and the tracking degradation associated with UMI tracking at high speeds. This approach allows the robot to acquire the fast and accurate data required to learn precise, dynamic movements in fewer iterations.
