Research On Autonomous Localization And Navigation Of Unmanned Ground Vehicles

Autonomous driving is one of the current research hotspots.As an interdisciplinary field,making research on autonomous driving is impossible without professional knowledge on artificial intelligence,computer vison and sensing technologies.The target of autonomous driving,by its very nature,is still to sense the external environment under the control of computers and manipulate its mechanical structure reliably.Therefore,autonomous driving can be classified into robotics in some ways.Aiming at problems related to sensing technologies and environment modeling in autonomous driving,the thesis refits a 1/8 remote-controlled model of an off-road vehicle as the experiment platform of localization and autonomous navigation with energy-saving high-performance NVIDIA Jetson TX2 with ARM architecture and builds up a complete software system of localization and navigation based on ROS.Within the system,perception of environment utilizes basic concepts of probabilistic robotics.By non-parametric particle filtering,experiments of map building and localization in the in-door environment are conducted,successfully reproducing the whole process of in-door navigation.To improve real-time issues of autonomous localization and navigation on Jetson TX2,GPUs on TX2 are introduced for GPU-based large-scale parallel computation.The thesis proposes a CUDA-based optimization solution to the package costmap₂d in ROS after analyzing its workflow and data schemes.After optimization,real-time performance can be satisfied at a power consumption of fewer than 15 watts.Based on the consideration above,this thesis designs a field trial to validate the effect of the solution.Performance tests conducted on three different computers prove that the solution can dramatically improve the performance of costmap₂d in various cases,decreasing the execution time by up to 88%.Furthermore,it is also proved by another set of experiments that optimized version of costmap₂d is less sensitive to data size and that autonomous localization and navigation can be implemented on maps with the larger size.