Identifying Important Sensory Feedback for Learning Locomotion Skills
Published in Nature Machine Intelligence, 2023
Recommended citation: Wanming Yu, Chuanyu Yang, Christopher McGreavy, Eleftherios Triantafyllidis, Guillaume Bellegarda, Milad Shafiee, Auke Jan Ijspeert and Zhibin Li (2023). "Identifying Important Sensory Feedback for Learning Locomotion Skills." in Nature Machine Intelligence (NMI) 2023. https://www.nature.com/articles/s42256-023-00701-w
This paper presents a saliency analysis to determine the most crucial feedback states for motor skills in deep reinforcement learning, showing that using only key states, a simulated robot can perform various locomotion tasks robustly. The suggested approach can be applied to differentiable state-action mappings, like those in neural network control strategies, allowing for the acquisition of various motor skills with fewer sensing requirements.