Computation Complexity Evaluation of FastSLAM Algorithm for Unmanned Ground Vehicles

FastSLAM algorithm is commonly used in Unmanned Ground Vehicles (UGVs) recently. One of the main problems under research is the computation cost of this probabilistic algorithm. Since the speed of the UGV is limited by the latency of the algorithm, the computation complexity and its effect on the step time of the FastSLAM needs to be investigated. The present work addresses the effects of the number of particles and number of map features on the computation complexity of the FastSLAM algorithm. The study included the prediction, the observation, data association and resampling phase's complexities. Also, the correlation between the uncertainty of the UGV location and the number of particles was addressed. The simulation study was validated experimentally using hardware in the loop (HIL) setup. The analysis showed that when there is a prior knowledge of the average number of map features, an optimum number of particle filters could be set for that UGV in the given environment while maintaining an improved performance of the algorithm.