Research On Decision-Making And Control Method For Autonomous Vehicle

For reducing the traffic accidents caused by driver’s factors as well as the energy consumption caused by traffic congestion,a great deal of efforts have been contributed to the research of autonomous vehicle in this decade.As one of the key generic technology of autonomous vehicle,decision-making and control policy is mainly responsible for making reasonable driving maneuver according to environmental perception.Thereafter,optimal trajectories within the reachable area will be generated by which the control policy will control the actuator of the vehicle to follow.In this paper,the behavior of decision-making,motion planning and control policy algorithms of autonomous vehicles have been investigated.Field test experiments have been conducted to verify the feasibility and effectiveness of the algorithm we proposed.Four main parts of this paper can be classified briefly as follows:(1)Decision-making algorithm of autonomous vehicle has been proposed.A finite state machine decision for enclosed scenario driving tasks have been established.Moreover,the lane-change,leading car following and expected speed decision mechanism are analyzed.NGSIM trajectory data set are preprocessed and classified into three types of data: left lane change,right lane change under lane keeping and free driving.The feature vector is extracted on the basis of the data and ten the LSTM model can be trained.In this way,the decision-making algorithm based on LSTM neural network are established.(2)Motion planning algorithm of autonomous vehicle has been investigated.The principle of trajectory planning based on the five-order polynomial is analyzed.The influence of different trajectory prediction time on the planning trajectory are analyzed.In the curve coordinate system,the vehicle’s motion planning can be decomposed into different parts at two directions: vertical and horizontal motion planning.In this way the vertical and horizontal motion planning can be combined to generate the optimal motion trajectory.Subsequently,different scenarios such as leading car following,merging lane,parking and lane keeping have been simulated.Generally,the simulation results are consistent with the corresponding scene task requirements.Considering the potential collision risk of the generated candidate trajectory,we proposed the risk assessment algorithm for each generated candidate trajectory based on the velocity obstacle(VO)method.(3)Control policy and execution algorithm of autonomous vehicle have been demonstrated.The principle of trajectory tracking based on preview is analyzed firstly then the variable preview distance considering road curvature is proposed.On the basis of traditional preview algorithm,we proposed an improved preview control algorithm based on PID closed-loop control method.After that,the model predictive control(MPC)algorithm is generated for tracking the trajectory.For implementing the algorithm,the objective function of the linear model is established with the constraint condition and then the optimal sequence can be generated by using the rolling horizon optimization.Finally,the MPC algorithm is verified by simulation experiments,which including straight shape,‘M’ shape and circular shape trajectory tracking.(4)Field test experiments including decision-making and control policy.For validating the decision-making and control policy we proposed in this paper,field test experiments have been conducted in the enclosed test area on campus,with self-developed "Xing-Yuan" autonomous vehicle as experimental object.Different scenarios tests have been implemented such as active lane changing,static obstacle avoidance,parking and leading car following.The experimental results prove the feasibility and effectiveness of the decision-making control algorithm established in this paper.