Fan Shi , Yuta Kojio, Tasuku Makabe, Tomoki Anzai, Kunio Kojima, Kei Okada, Masayuki Inaba
JSK Lab, the University of Tokyo

To our best knowledge, it is the first work to analyze the assistant format and demonstrate several challenging bipedal skills under reference-free settings. We successfully verify the learning process on 5 robots with different kinematics, including the asymmetrical configuration which has large dynamic mismatches with the usual human data. We hope the proposed bio-inspired method could open up new possibilities for general bipedal learning and benefit the multi-skills framework in future work.