Design of Real-Time SLAM Algorithm for Quadruped Robots in Complex Terrains

To address the poor adaptability of traditional SLAM algorithms to quadruped robots in complex terrains, this study optimizes the collaboration between real-time SLAM and quadruped robots. ORB-SLAM2 is improved by adding a dense mapping thread: combined with a Kinect camera and filtering, a dense point cloud is built, and a 3D semantic map is generated by fusing YOLO v5 and super-voxel clustering. Hector-SLAM is optimized using bicubic interpolation to enhance grid map accuracy, and EKF to fuse IMU and odometer data for motion distortion correction and map ghosting elimination, with effectiveness verified via AGV navigation experiments. An exploratory gait for the quadruped robot is designed for obstacle avoidance through tactile sensing; PSO optimizes step length, height, and cycle, with joint kinematic limits as constraints, balancing speed, stability, and other indicators. Results show the improved SLAM provides high-precision perception, and the optimized gait enhances complex terrain adaptability, supporting quadruped robot applications.