Autonomous Navigation: LIDAR-based SLAM, Terrain of Technology Explored

Abstract: In this thesis, an autonomous system capable of navigating and mapping unknown environments is developed. The solution uses a family of algorithms called SLAM or simultaneous localization and mapping, capable of mapping the environment and retaining accurate position data without external sensors such as GPS. Firstly four different SLAM algorithms are implemented and then four different pathing algorithms are tested with a generated map. Everything runs on a hoverboard-based robot using an RPI as the processing unit and LIDAR as the only sensor. The performance is evaluated by analyzing the processor utilization, the positional accuracy and the accuracy of the generated map. It is concluded that the RPI has good enough performance to run the program while leaving processing power for other tasks. The achieved positional accuracy is usually better than 10 cm which is a good result given the circumstances. The generated map has a map-resolution dependent accuracy causing an error of less than $10'\sqrt{2}$ cm between points on the map, it reproduces long distances of >20 m with no further error.