Research On Autonomous Navigation Strategy And Control Algorithm Of Unmanned Vehicle

Unmanned technology has broad application prospects and has great potential to reconstruct the entire transportation system in the future.As technology continues to advance,people are looking forward to the future of unmanned driving."Unmanned driving" is to achieve autonomous navigation without human participation in control.The unmanned autonomous navigation system has become a research hotspot because of the complexity of the environment it faces.Among them,there are two major problems.First,the data provided by the information source is not accurate enough,not stable enough and not comprehensive enough to provide effective data support for decision-making.Second,it is difficult to construct a model that can adapt to various environmental decisions.To this end,The thesis is based on the unmanned complex environment,based on the relevant theory of complex adaptive systems,built an unmanned autonomous navigation system to achieve autonomous navigation.Firstly,aiming at the autonomous navigation problem of unmanned vehicles in complex and ever-changing environment,based on Agent’s way of modeling complex systems,an agent-based autonomous navigation system is constructed.The autonomous navigation system mainly includes a scene conversion module,an information set module,and a policy set module.The information set module mainly solves the problem of supporting decision data through information fusion.The policy set module provides policy support for autonomous navigation in complex environments.Then,the self-localization information in the information set module is studied by fusion.The original sensor information has the disadvantages of uncertainty and limited sensibility.The redundant sensor can enhance the accuracy and robustness of the information,but the correlation of the sensor information is usually unknown.In this paper,a decentralized fusion architecture based on GPS,SINS and Stereo vision odometer is proposed,which effectively improves the accuracy and robustness of self-positioning information.Next,the rule-based autonomous navigation strategy in the policy set module is studied.A rule-based autonomous navigation strategy on structured roads is designed.Under the rule constraint,the trajectory is generated based on the time window,and the complex decision problem is transformed into the target state determination and trajectory evaluation problem of the trajectory,and the adaptive speed and safety distance are based on the following model.The trajectory is optimized by quantifying the real-time risk and comfort of the driving.Finally,the decision to change lanes,deceleration and acceleration on complex structured roads was realized,and autonomous navigation of complex structured roads was realized.Finally,the learning-based autonomous navigation strategy in the policy set module is studied.The decision model in complex environment is modeled as neural network.According to the evolution law of complex adaptive system,unstructured road training scene is constructed based on Unity3 D,and genetic algorithm is used to evolve decision neural network.By gradually increasing the complexity of the scene,the driverless car learns to turn,learn to avoid obstacles,and learn to drive.Finally,by adding a number of learning-capable driverless cars to the scene to co-evolve to simulate highly complex unstructured roads.Finally,autonomous navigation of complex unstructured roads is achieved.