Event

Seoul National University's RISE framework enables humanoid robots to perform multi-stage tasks continuously

1 reports · 1 sources

A research team led by Songhwai Oh, a professor at Seoul National University's College of Engineering, proposed the vision-language-action framework RISE, which separately collects demonstration data for locomotion and object manipulation and then learns jointly, enabling the robot to perform multi-stage tasks continuously without full multi-stage demonstrations. The team says the method does not require additionally collecting multi-stage demonstrations of transitions between the two behaviors. The research team validated the framework in simulation and on a real humanoid robot, the Unitree G1. In addition to basic whole-body joints, the robot used in the experiments was equipped with two Dex3-1 robotic hands, for a total of 43 controllable degrees of freedom; in real-robot experiments, the system completed two-stage, four-stage, and up to five-stage locomotion and manipulation tasks. The team says that as environmental adaptability and safety are further improved, the framework is expected to be used in scenarios such as delivering supplies in hospitals, caregiving assistance, and tidying up items at home.