The Research And Implementation Of 4WS-4WD Unmanned Ground Vehicle Autonomous Navigation System

The autonomous navigation system of unmanned ground vehicles is an important part of intelligent transportation,which is of great significance to the development and progress of human society.In recent years,the software and hardware technologies related to autonomous navigation systems for unmanned ground vehicles have developed rapidly.The unmanned ground vehicle’s in-vehicle computing power is increasing geometrically.The in-vehicle sensors are beginning to be industrialized on a large scale.Artificial intelligence technology makes it possible to decisions making in complex scenarios.Compared with the hardware part of the autonomous navigation system,the software part of the autonomous navigation system is more likely to be the current technical bottleneck for autonomous driving that cannot be implemented on a large scale.The research content of this paper mainly focuses on the trajectory planning,motion control and end-to-end autonomous navigation strategy.This article mainly carried out the following work:(1)Based on self-developed four wheel independent steering and four wheel independent drive wire-controlled vehicles,a layered autonomous navigation system for unmanned ground vehicles is built.The autonomous navigation system includes:software layer,Hardware layer and execution layer.In order to facilitate the cross-platform porting of the autonomous navigation system,the software layer is divided into upper and lower parts,and the motion decomposition module at the lower layer handles the differences brought by unmanned ground vehicles with different chassis structures.(2)Based on the Hybrid A * method,the trajectory planning of unmanned ground vehicles is realized.Aiming at the problem that the target trajectory output by the A *method does not satisfy the minimum steering radius constraint of the vehicle,the kinematic constraints of the vehicle are considered when the path node is expanded to ensure the output target trajectory is executable.In addition,a dubins path check module is also introduced to accelerate the target trajectory search process.(3)An adaptive pure pursuit method under longitudinal distance constraints is proposed to realize the motion control of 4WS-4WD vehicles.Aiming at the problem of "cutting edge" and the problem of vehicle sway appearing in the traditional pure pursuit method,the traditional pure pursuit method is improved.The fuzzy controller of the look-head distance is built,so that the look-head distance of the pure pursuit method can automatically adapt to the change of the curvature of the target trajectory,which solves the problem of "cutting edge".In addition,the adaptive longitudinal distance constraint is introduced into the traditional pure pursuit method to ensure that the vehicle will not swing back and forth near the target trajectory during the trajectory tracking process.(4)An end-to-end autonomous navigation strategy for unmanned ground vehicles was developed based on a deep deterministic strategy gradient method.The autonomous navigation strategy can be applied to autonomous navigation of unmanned ground vehicles in an unknown environment.Aiming at the long training time of autonomous navigation strategy,the method of network weight pre-training is used to accelerate the training process.Experimental results show that: network weight pre-training can speed up the training process to a certain extent.