Autonomous Locomotion Control Strategy for Quadruped Robot Based on Visual Lidar Fusion Terrain Recognition

In comparison to other robot types, quadruped robots have greater flexibility and terrain adaptability for the special leg structure. To enhance their autonomy and intelligence in complex environments, a terrain perception integrated hardware system mixed lidar, camera and IMU, is developed for quadruped robots. The system utilizes FAST-LIO for localization, Globally Consistent Dense Elevation Mapping (GEM) for building local maps and maintaining global map consistency, and Efficient Residual Factorized Network (ERFNet) for terrain recognition. This allows for the classification of various terrains such as land, grass, curbs, and stairs. Adaptation strategies for step height and step frequency have been designed for various terrains. In the experimental section, the recognition effectiveness of different terrains was verified, and the feasibility of the perception system was demonstrated by combining control strategies.