The Effects of Sensor Fusion on Localisation in a Sparse, Outdoor Environment

Abstract: This thesis compares the results of a localisation algorithm for a mobile robot in both a sparse and a densely featured environment whilst varying key parameters. The software and hardware required to enable the mobile robot to localise itself in the sparsely featured, GPS-denied, outdoor environment is described. The project included a rebuild of a robot built in a previous project, however some hardware was retained. The localisation algorithm was an Extended Kalman Filter fused LeGO-LOAM Simultaneous Localisation and Mapping (SLAM) algorithm with wheel odometry and IMU data. The sensors used for localisation and physical robot parameters (speed and robot weight) were varied to test the localisation performance. Contrary to the projects hypothesis, the smallest error in the sparse environment was from the wheel odometry alone and the second smallest error in the dense environment was the LeGO-LOAM algorithm output. The smallest error in the dense environment behaved as expected at low speeds, with high payload and all sensors, however this test had the largest variance between test cases, therefore may be an outlier. The results show that in both the sparse and the dense environment the larger the velocity the larger the error. Recommendations for further development on this thesis topic are included.