Trajectory Planning And Tracking Control Of Wheel Loader Based On Driving Data

The working environment of wheel loaders is extremely poor.After two hours of continuous operation,the driver’s physical strength and endurance significantly reduces;vision and hearing deteriorates;and concentration of the energy declines.These factors pose a great threat to driving safety.The annual salary of domestic loader drivers is approximately 120,000 yuan,and the annual labor cost of a wheel loader is approximately500,000 yuan.The wide application of self-driving loaders holds the key to the development of unmanned earthwork operations technology and the realization of safe,efficient,and green operations.Compared with the traffic environment of ordinary passenger cars,which is typified by dense traffic and complex road networks,the operational conditions of wheel loaders are marked by simplicity and a high degree of repetition,making it easier to realize automated driving.However,owing to the lack of an evaluation method for the cycle operation of wheel loaders,formulating a reasonable level of specification for cycle operation,and obtaining specification for the optimal Vshaped cycle operation are complex tasks.In addition,the trajectory plan does not match the actual operational driving characteristics of the wheel loader,and the trajectory tracking control system does not have the ability to adapt to a new environment and task so that it cannot meet the intelligent control requirements of an automated-driving wheel loader requiring adaptive ability for different driving environments.In order to meet the requirements of intelligent control with an adaptive ability for different driving environments,a set of wheel loader trajectory planning and trajectory tracking control methods is proposed,based on actual driving data of a wheel loader.First,the driving data of the wheel loader in the V-type operation mode are collected,and the evaluation method for the cycle operation of the loader is studied.Second,the mechanical model of the wheel loader is constructed,and the constraint conditions,such as speed,acceleration,and minimum turning radius of the wheel loader are comprehensively considered.Based on RRT *,CC-steer,and path speed decomposition algorithms,a collision-free trajectory planning method is studied to accommodate the forward/reverse driving state of the loader in the working area.Then,an adaptive model predictive control(AMPC)path tracking method with the ability to adapt to the path curvature is studied.This method considers variations in the path curvature,based on model predictive control(MPC)theory.Finally,based on the deep deterministic policy gradient(DDPG)learning algorithm,a deep reinforcement learning-based trajectory tracking control method that is adaptive to new environments and tasks under a data-driven framework is studied.The implementation of this study provides theoretical support for the adaptive ability of an automated-driving wheel loader under different driving environments.The primary research content of this study is as follows:(1)The driving data of the driver in the V-type operation mode of the loader are collected,the operational characteristics of the wheel loader driver are analyzed,and the method for evaluating the advantages and disadvantages of the loader cycle operation is studied.The driving data acquisition system of the loader is constructed,and the driving data of skilled and unskilled drivers in the V-type mode are collected.The related parameters of cycle operation,such as cycle operation time,driving distance,and braking sliding work,are extracted.On this basis,the above parameters are analyzed under different operation cycles and operation modes.The proposed evaluation method provides a theoretical basis for the selection of the operation time and driving distance of the automated wheel loader in the V-type operation mode.(2)A vehicle dynamics model that can reflect the handling and stability of the wheel loader is established,and a collision-free driving trajectory planning method that satisfies the forward/reverse driving state of the loader in the working area is studied.Based on the future path curvature information and current vehicle status information,a dynamic deviation model reflecting the deviation in the lateral position and the heading angle of the wheel loader during the trajectory tracking process is established.A vehicle dynamics model that can reflect the steering stability of the wheel loader is constructed.Using the driving data of the wheel loader,such as vehicle speed,acceleration,minimum stable turning radius of the entire vehicle,a set of trajectory points,which consider both space and time factors,is generated,based on RRT * and CC-steer algorithms.This set of trajectory points also includes an obstacle avoidance function and conforms to the actual operational characteristics of the loader.(3)An adaptive model predictive trajectory tracking control method with the ability to adapt to the curvature of the driving path is studied.Based on the mechanical model and dynamic deviation model of the wheel loader,an AMPC trajectory tracking system is established,which takes the path curvature in the time-varying state as a disturbance input,acceleration and articulation angle as control inputs,and vehicle speed,lateral position deviation,and heading angle deviation as outputs.The trajectory tracking simulation of the self-driving wheel loader is verified,aiming at the planned driving trajectory and using the built AMPC trajectory tracking system,and the AMPC trajectory tracking control with the ability to adapt to the curvature of the path is realized.(4)A trajectory tracking control method for a wheel loader based on deep reinforcement learning is studied.Based on the DDPG deep reinforcement learning algorithm,four environment models and four DDPG agents corresponding to the environment models are constructed.Through interactive learning based on trial-anderror mechanisms,DDPG agents that approach the optimal control strategy under different environmental models are obtained.By considering the planned V-shaped operation trajectory as the target,and using the built-in deep reinforcement learning trajectory tracking system,the trajectory tracking simulation of the self-driving wheel loader in different environments is verified.This method can eliminate the dependence of the trajectory tracking control method on the vehicle model while realizing trajectory tracking control with the adaptive ability of deep reinforcement learning for new environments and tasks.