PARTICLE FILTER APPLICATION TO RADIO-BASED RANGE-ONLY SELF-LOCALIZATION

Abstract

This work is aimed at searching for a relatively accurate and computationally efficient algorithm for range-only self-localization. Key requirement to the data sources is cost effectiveness. The radio signal strength indicator (RSSI)-based distance estimation model was chosen, due to wide availability of hardware and cost effectiveness. Key requirements to the algorithm are robustness towards noise introduced by RSSI-based distance estimation errors and flexibility to introduce additional data sources. Particle filter algorithm was chosen, because it satisfies both requirements, although it has bigger computational costs comparing to Kalman filter. A simulation environment was created to conduct experiments with assumed transmitted signal strength of 1W and square root signal decay law. A Particle filter-based actor was implemented. Actor starts by translating geo-spatial coordinates of beacons into relative planar cartesian coordinates, generates particles around beacons. Particles are placed with uniform distribution within a torus with co-planar circular axis radius equal to distance estimation. In each experiment, simulation software walks actor through a pre-defined path, compares obtained position estimates with expected state and calculates MAE and RMSE. Obtained results were satisfactory, in some cases showing MAE=2.5981 and RMSE=2.8474 after short stabilization period of few iterations. This period exists due to suboptimal initial particles placement. A Kalman filter-based actor was implemented as a comparison. A reference Kalman filter-based implementation showed slightly worse overall results with MAE=12.5199 and RMSE=13.2238. This can be explained by a lack of input from IMU. Further research will be directed towards UAV swarm self-localization and collision avoidance.