There are still many challenges in achieving stable movement and adaptability to different terrains for quadruped robots in unstructured environments. This paper, based on the Gazebo simulation platform, utilizes LiDAR as the primary means of environmental perception, supplemented by a depth camera, to achieve efficient perception and recognition of complex environments. To address the task requirements of quadruped robots, a comprehensive environmental perception and navigation system was developed, including global map construction, local navigation map generation, precise balance control, and global positioning. Based on the generated global and local maps, path planning for the quadruped robot was achieved using a sliding window A* path planning algorithm. Finally, by perceiving obstacles and adjusting speed in real-time, speed commands are sent to the robot controller, enabling real-time gait switching and obstacle avoidance, thereby enhancing the quadruped robot’s ability to cope with complex environments. A complete framework was developed for quadruped robots based on recognizing and perceiving complex environments, conducting path planning, and switching gaits in real-time.