Research On Automatic Climbing Method Based On Excellent Driving Behavior

Because of the limitations of the human body,there are visual blind areas,drunk driving,fatigue driving and other factors in the driving process,which will lead to the frequent occurrence of traffic accidents.How to create a safe and comfortable driving environment has become an urgent problem in the field of transportation.However,autonomous driving technology can avoid driving problems caused by human’s innate limitations.And the aim of autonomous driving technology is the realization of anthropomorphic driving.Anthropomorphic driving requires the guidance and data support of human’s actual excellent driving behavior and the establishment of learning model.Firstly,five driving conditions(left turning,right turning,vehicle starting,vehicle braking and climbing process)that are common in urban scenes are analyzed.Based on the driving simulator platform,the test route and test flow are designed.52 drivers with more than two years’ driving experience and different driving characteristics were selected to conduct multiple simulated driving experiments,and a large amount of information of driver operation behavior and vehicle motion state were obtained.Secondly,the significance of selecting excellent driving behavior was expounded.After analyzing the advantages and disadvantages of various clustering algorithms,a gaussian mixture model based on EM algorithm was selected to evaluate driving behaviors.First,all behaviors were analyzed by gaussian mixture model to select the optimal behavior and marking the 23 drivers to which it belongs.Because the environmental familiarity will have some impact on driving performance.The optimal behavior is clustered again,and the proportions of excellent behaviors of drivers in familiar and unfamiliar scenarios are compared.The results show that excellent drivers have better driving behaviors in familiar scenarios,which provides a new collection method for driving behavior database.Finally,in the five driving conditions,the comprehensive driving ability of the driver can be well reflected in the climbing process.On the other hand,because of the particularity of terrain,the traffic accidents are more likely to happen,which requires higher driving skills of drivers.Therefore,this paper establishes an excellent climbing driving behavior learning model.The mathematical principles of the random forest and neural network models are introduced,and the main parameters of the models are obtained through many experiments.Finally,it is concluded that the random forest model has better prediction accuracy than neural network model.The effectiveness of the direct mapping of input to action in the end-to-end pattern is verified.