The article presents a study on Robust Autonomous Modulation (ROAM), a mechanism that helps robots adapt to new situations during deployment. ROAM allows a robot to select and adapt from a range of pre-trained behaviors based on the perceived value of these behaviors. This selection and adaptation process occurs within a single test episode without human supervision. The study shows that ROAM allows a robot to adapt more than twice as efficiently as existing methods when facing a variety of out-of-distribution situations during deployment. The mechanism has been tested both in simulation and on a real quadruped robot.
Publication date: 2 Nov 2023
Project Page: https://anniesch.github.io/adapt-on-the-go/
Paper: https://arxiv.org/pdf/2311.01059
