Research On Indirect Shared Control Strategy Of Intelligent Vehicle For Human-machine Cooperative Driving

Due to the current technical bottlenecks and social difficulties,highly or fully automated intelligent vehicles are difficult to achieve large-scale applications in a short time.Drivers will continue to participate in driving tasks for a long time in the future.Therefore,the new driving mode that the driver "human driving" and the intelligent driving system "machine driving" cooperate to complete the driving task,namely,human-machine driving,will exist for a long time.To study the human-machine cooperative driving strategy of intelligent vehicle,this paper establishes a hierarchical cooperative driving strategy including tactical decision-making level and operation execution level.At the tactical decision-making level,the human-machine cooperative driving system not only needs to plan the reference trajectory,but also needs to dynamically allocate the man-machine control authority,and the accurate evaluation of driving risk is the common basis of both.Aiming at the problem that the existing risk assessment methods seldom consider the motion trend of dynamic obstacles,a driving risk assessment model based on trajectory prediction is established,and a humanmachine driving decision-making method is then proposed.In the operation execution layer,aiming at the problem that the existing man-machine control command fusion methods seldom consider the smoothness of human-machine command fusion,a human-machine command fusion method based on multi-objective MPC is proposed.The main research contents of this paper include:Firstly,for the risk assessment of the driving environment,a travel trajectory prediction model considering multi-vehicle space-time interaction is established.Based on the spatial grid expression method of the driving environment,the LSTM encoderdecoder architecture is constructed to extract the time-series features in the trajectory information,motion state information,and road boundary information of the surrounding vehicles.To consider the influence of different historical moments on prediction time,a temporal attention mechanism is designed to further explore the temporal interaction between historical times and prediction times.And the time-series features of surrounding vehicles are aggregated into a tensor according to the corresponding grid to represent the spatial interaction of surrounding vehicles.The features of different grids are weighted and fused through the designed spatial attention,to mine the spatial interaction between surrounding vehicles and self vehicles in different locations.The proposed trajectory prediction algorithm is trained and tested based on the NGSIM dataset.The root means square error between the predicted trajectory and the actual trajectory is used as the performance evaluation index and compared with the original LSTM model.The results show that the trajectory prediction algorithm proposed in this paper can improve the trajectory prediction accuracy.Secondly,a human-machine driving decision-making method based on driving risk assessment is constructed.Based on the risk field theory,the static obstacles and road structure in the environment are modeled,and the static risk field model is established.Considering the influence of the motion trend of dynamic obstacles on driving risk,a motion risk field model is established based on the predicted trajectory.To consider the impact of different prediction times on the current driving risk,a risk attenuation factor is proposed to further optimize the risk distribution of the motion risk field.Combining the dynamic and static driving risk field,the comprehensive driving risk field model is obtained,and a human-machine cooperative driving decisionmaking method including local path planning and dynamic allocation of humanmachine control authority is proposed.On the one hand,based on the NMPC method,the error between driving risk field strength and tracking global trajectory is minimized,and the local reference trajectory is obtained in real-time.On the other hand,based on the results of the driving risk assessment,the mapping relationship between weight factor and driving risk is established,and the dynamic allocation of human-machine control authority is realized.Then,an indirect fusion strategy of human-machine control commands based on multi-objective MPC is established.Considering the real-time ability and calculation accuracy,a linear time-varying prediction model based on the monorail vehicle model is established.Considering the ability of MPC to solve with constraints,the constraints of tire sideslip angle and sideslip angle are proposed as the constraints of the controller based on the features of vehicle dynamics.Based on the above foundation,a multiobjective optimization function including matching the driver’s command,tracking the dynamic trajectory,and smoothing the control command is designed.The humanmachine control command fusion problem is transformed into the controller multiobjective optimization problem,and the human-machine control command fusion is realized by an optimization solution.Based on the results of control authority allocation,the weight matrix of the human-machine optimization objective is continuously adjusted to realize the smooth fusion of human-machine control commands in the process of control authority allocation.Finally,the human-machine cooperative obstacle avoidance simulation experiment is carried out based on the Simulink-Car Sim joint simulation platform,in which the effectiveness of the proposed shared control strategy in static and dynamic obstacle avoidance scenarios is verified respectively,and the performance of the controller is evaluated based on the vehicle stability index.In the static obstacle avoidance scenario,the effectiveness of human-machine cooperative obstacle avoidance is verified when the driver is distracted or turns in panic at different speeds.In the dynamic obstacle avoidance scenario,the effectiveness of human-machine cooperative overtaking and lane changing is verified in the distracted state of the driver.The results show that the shared control strategy proposed in this paper can realize safe and stable human-machine cooperative driving in static and dynamic scenarios.