Autonomous operation of a robot dog for point-cloud data acquisition

This paper presents the development of an autonomous operation for a robot dog, the Unitree Go1, to acquire a detailed point cloud of a building’s interior, from simulation to real-world experiments. The system integrates a 3D LiDAR, an RGB-D camera, and a mini-PC, enabling the Go1 to autonomously navigate, avoid obstacles, and collect data. The process includes developing the solution inside a simulation environment within the Robot Operating System (ROS) to simulate the Go1 movement in a building model. Real-Time Appearance-Based Mapping (RTAB-Map) has been adapted for the robot's 3D localization. In addition, real-world experiments were conducted to validate the pipeline. Our proposed framework provides valuable insights for the construction community, outlining current limitations and suggesting directions for future work. To promote transparency and enable reproducibility, we have made the code used and developed in our experiments publicly available at: https://github.com/MooKol/unitree_go1.