Research On Intelligent Model And Method Of Motion Control For Autonomous Vehicle

Autonomous vehicle is a popular field of research in the world.It’s important for transportation and military.Motion control of autonomous vehicle is one of the core issues of the research.Autonomous vehicles gradually move from closed-road scenarios to open-road scenarios,the scenarios are becoming more and more.The traditional hierarchical motion control of autonomous vehicle uses a combination of motion planning and feedback control,which lacks intelligence and is difficult to deal with complex scenes.Complex scenes pose a challenge to motion control of autonomous vehicles.The intelligent motion control of autonomous vehicle is an effective way to solve safe driving in complex environment.In recent years,the more popular end-toend model adopts the learning method,which has intelligence,but it lacks interpretability and can not support for multiple driving behaviors.In this regard,this paper studies the intelligent motion control model and method of autonomous vehicles.The specific research contents are as follows:First,research on hierarchical model of autonomous vehicle intelligent motion control.Firstly,the characteristics of the current autonomous vehicle motion control model are compared and analyzed,and the "meta-action and decision-vehicle control" hierarchical model framework supporting multi-driving behavior in complex scenarios is proposed.Secondly,the driving behavior is decomposed and the hierarchical granularity is determined.The controlled model is proposed which abstracts each layer of the motion control model,then verifies the feasibility of the layered model with controlled model.Finally,the overall motion control model structure is described in combination with the controlled model.Second,meta-action decision model modeling based on deep Q network algorighm.Firstly,the meta-action decision model is modeled by the deep Q network algorithm.Secondly,the approximate optimization method of deep reinforcement learning under semi-Markov is studied.Finally,the meta-action decision model is implemented in combination with the lane change behavior to verify its decision-making ability and the results were analyzed.Third,vehicle control model modeling based on deep deterministic policy gradients algorithm.Firstly,the vehicle control model is modeled by the deep deterministic policy gradient algorithm.Secondly,the vehicle control model and the meta-action command are combined.Finally,the vehicle control model is implemented by combining the lane change behavior,verifying its control ability and analysis the result.Finally,combining the meta-action decision and the vehicle control into motion control of autonomous vehicle and implement the model.Firstly,modify the TORCS simulator for autonomous vehicle and develop an interactive environment to provide an environment for the complete implementation of the motion control model.Secondly,the two models are merged and modeled;finally,the motion control of the lane changing behavior under a complex scenario is verified and the result is analysed.