Sensor fusion framework for robust occupancy grid mapping

Sensor based perception of the environment is an emerging area of research where sensors play a pivotal role in mobile robots to map the environment. For autonomous mobile robot mapping, information from different range sensors like vision sensor, laser range finder, ultrasonic and infrared sensors, etc. are fused to obtained better perception. Despite significant progress in this area, it still poses great challenges to attain robustness and reliability of the maps. In this paper, a new architecture of sensor fusion framework is proposed to make the map robust and reliable. The proposed architecture consists of the three main segments: a) Pre-processing of sensory information b) Fusion of information from heterogeneous sensors and c) Post-processing of the map. As reported in literature, specular reflection of sonar sensor is considered as the fundamental cause of an error in map making. To overcome such problem, pre-processing of information for sonar sensor is proposed in which fuzzy logic algorithm is used to discard the specular information. The proposed fuzzy technique shows that the average performance of the resultant grid is increased by 6.6%. The last part of the paper deals with the post-processing of grid with newly proposed dedicated filter (DF). The updated results using proposed framework show an average improvement of 8.4% in the occupancy grid. The qualitative comparisons show the improvement in the results where the overall occupied and empty area of the resultant map is extremely near to the reference map.