eprintid: 10159089 rev_number: 8 eprint_status: archive userid: 699 dir: disk0/10/15/90/89 datestamp: 2022-11-14 11:09:50 lastmod: 2022-11-14 11:09:50 status_changed: 2022-11-14 11:09:50 type: article metadata_visibility: show sword_depositor: 699 creators_name: Yang, Chuanyu creators_name: Yuan, Kai creators_name: Zhu, Qiuguo creators_name: Yu, Wanming creators_name: Li, Zhibin title: Multi-expert learning of adaptive legged locomotion ispublished: pub divisions: UCL divisions: B04 divisions: C05 divisions: F48 keywords: cs.RO, cs.RO, cs.AI, cs.LG note: This version is the author accepted manuscript. For information on re-use, please refer to the publisher’s terms and conditions. abstract: Achieving versatile robot locomotion requires motor skills that can adapt to previously unseen situations. We propose a multi-expert learning architecture (MELA) that learns to generate adaptive skills from a group of representative expert skills. During training, MELA is first initialized by a distinct set of pretrained experts, each in a separate deep neural network (DNN). Then, by learning the combination of these DNNs using a gating neural network (GNN), MELA can acquire more specialized experts and transitional skills across various locomotion modes. During runtime, MELA constantly blends multiple DNNs and dynamically synthesizes a new DNN to produce adaptive behaviors in response to changing situations. This approach leverages the advantages of trained expert skills and the fast online synthesis of adaptive policies to generate responsive motor skills during the changing tasks. Using one unified MELA framework, we demonstrated successful multiskill locomotion on a real quadruped robot that performed coherent trotting, steering, and fall recovery autonomously and showed the merit of multi-expert learning generating behaviors that can adapt to unseen scenarios. date: 2020-12-16 date_type: published publisher: American Association for the Advancement of Science (AAAS) official_url: http://dx.doi.org/10.1126/scirobotics.abb2174 oa_status: green full_text_type: other language: eng primo: open primo_central: open_green verified: verified_manual elements_id: 1951073 doi: 10.1126/scirobotics.abb2174 medium: Print pii: 5/49/eabb2174 lyricists_name: Li, Zhibin lyricists_id: ZLISX72 actors_name: Li, Zhibin actors_id: ZLISX72 actors_role: owner funding_acknowledgements: ICT1900349 [Zhejiang University]; EP/L016834/1 [Engineering and Physical Sciences Research Council]; ICT20005 [Zhejiang University] full_text_status: public publication: Science Robotics volume: 5 number: 49 article_number: eabb2174 event_location: United States citation: Yang, Chuanyu; Yuan, Kai; Zhu, Qiuguo; Yu, Wanming; Li, Zhibin; (2020) Multi-expert learning of adaptive legged locomotion. Science Robotics , 5 (49) , Article eabb2174. 10.1126/scirobotics.abb2174 <https://doi.org/10.1126/scirobotics.abb2174>. Green open access document_url: https://discovery.ucl.ac.uk/id/eprint/10159089/1/combined%20PDF.pdf