Motor Synergy Development in Symmetric Gait ofWhole-body Locomotion Learning (ICRA2023 video)

Motor Synergy Development in Symmetric Gait of Whole-body Locomotion Learning (ICRA2023 stand-alone video)

Enabling humanoid robots to have a natural and energy-efficient gait like humans in an automatic manner is a long-term goal. As a model-free algorithm, deep reinforcement learning relieves the reliance on the model-based controller with complex dynamics and is expected to achieve better performance and adaptivity than the traditional control methods. However, learning in a highly redundant system tends to result in unnatural movement patterns without synergistic joint coordination.
In this video, we analyzed the training process of walking and running tasks by a humanoid using deep reinforcement learning and symmetry promotion. We observed that the motor synergy hypothesized human motion occurred in humanoid locomotion tasks trained by deep reinforcement learning with symmetry promotion. The results illustrate the correlation between motor synergy and the performance of the humanoid robot, reflecting the consistency between humans and deep learning algorithm-trained humanoid robots in learning locomotion tasks under a highly redundant joint system.

Видео Motor Synergy Development in Symmetric Gait ofWhole-body Locomotion Learning (ICRA2023 video) канала Neuro-Robotics Lab