Abstract
This paper deals with the development of a localization methodology for autonomous vehicles using only a 3D LIDAR sensor. In the context of this paper, localizing a vehicle in a known 3D global map of the environment is equivalent to finding the vehicle's global 3D pose (position and orientation), in addition to other vehicle states, within this map. Once localized, the problem of tracking uses the sequential LIDAR scans to continuously estimate the states of the vehicle. While the proposed scan matching-based particle filters can be used for both localization and tracking, in this paper, we emphasize only the localization problem. Particle filters are a well-known solution for robot/vehicle localization, but they become computationally prohibitive as the states and the number of particles increases. Further, computing the likelihood of a LIDAR scan for each particle is in itself a computationally expensive task, thus limiting the number of particles that can be used for real-time performance. To this end, a hybrid approach is proposed that combines the advantages of a particle filter with a global-local scan matching method to better inform the resampling stage of the particle filter. We also use a pre-computed likelihood grid to speed up the computation of LIDAR scan likelihoods. Using simulation data of real-world LIDAR scans from the KITTI datasets, we show the efficacy of the proposed approach.