Building dense reconstructions with SLAM and Spot

Abstract: Having access to dense reconstruction of ongoing building constructions provides insight into the building process and could serve as both a tool for error detection and documentation of the actual outcome. In this thesis, Spot, the quadruped robot designed by Boston Dynamics is evaluated as a platform for site inspection. The result of the thesis was a prototype system built on top of the ROS framework which integrates the sensors of the robot platform. Combined with the visual SLAM library RTABMap for robust localization with loop closing capabilities and dense camera and LiDAR reconstructions, the system provides means for online map building and data acquisition. The localization performance achieved positional RMSE in the decimeter range for paths spanning further than 100 meters. The resulting LiDAR reconstructions provided comparable results to be evaluated against the planned building model and acheived centimeter accuracy with the current sensor configuration.