Experimental Evaluation of Computation Cost of FastSLAM Algorithm for Unmanned Ground Vehicles

Two decades ago, FastSLAM algorithm for mobile robots was introduced. Since then, dozens of research work focused on FastSLAM algorithm performance enhancement while keeping reduced computation cost. Since experimental evaluation of computation cost is dependent on the hardware capabilities of the platform, the present work introduces a quantitative theoretical method for predicting the computation cost of the FastSLAM algorithm. The method relies on the big (O) computation complexity which represents the worst case. The method was evaluated experimentally with different number of particles and different number of map features. The computation cost evaluation analysis was broken down into prediction, observation, data association and resampling computation cost evaluation. The proposed method was proven to be helpful in customization of FastSLAM parameters like number of particles and data association optimization for FastSLAM algorithm developers.