Comparative evaluation of time-of-flight depth-imaging sensors for mapping and SLAM applications

Autonomous robotic systems rely on simultaneous localisation and mapping (SLAM) algorithms that use ranging or other sensory data as input. Numerous algorithms have been developed and demonstrated, many of which utilise data from high-precision ranging instruments. Small unmanned aircraft systems (UAS) have significant restrictions on the weight of sensors they can carry, and light-weight ranging sensors tend to be subject to more error than their larger counterparts. The effect of these errors on SLAM effectiveness will depend on the algorithm in use. Our current work is focussed on evaluating different combinations of sensor and algorithm. This paper presents an evaluation of the performance of three SLAM algorithms that are freely available in the Robot Operating System (ROS), in conjunction with ranging data from two different time-of-flight imaging cameras: a commercially available Mesa Imaging sensor and a prototype sensor based on single-photon avalanche diode (SPAD) technology. Based on the results of this data collection, a difference with respect to the ability of the SLAM algorithms to handle noisy odometry data can be seen. GMapping is able to generate maps consistently when compared with KartoSLAM and Hector Mapping algorithms. However, KartoSLAM was able to create maps that represented the ground truth more accurately.