Bionic Gait Generation for Bipedal Robots Based on Human Gait Kinematic Synergies

Abstract

Effective gait generation is crucial for the development of bipedal robots. However, achieving human-like locomotion remains a major challenge in robotics, primarily due to the system’s high degrees of freedom, inherent nonlinearity, and the complex spatiotemporal characteristics of human walking. This study proposed a three-dimensional gait generation framework based on low-dimensional kinematic synergies of gaits. By applying principal component analysis to motion capture data, dominant kinematic synergies and their temporal key points were extracted. Subsequently, a neural network model was developed to map individual characteristics (walking speed, height, and body mass) to gait parameters (key-point parameters and gait cycle duration). The generated gait patterns were validated as reference trajectories for a skeletal model trained via reinforcement learning, successfully achieving non-falling locomotion. This study demonstrates that complex human gaits can be effectively parameterized and reconstructed using kinematic synergies, offering application potential for bipedal robots and walking assistance exoskeletons.