Research On Interaction-Aware Predictive Trajectory Planning For Autonomous Vehicle In Dynamic Environments

Autonomous vehicles have been regarded as effective solutions to reduce traffic accidents caused by human error,which is an important part of the future intelligent transportation system,and research on this field is of great important in both theory and practical.As one of the key technologies of autonomous vehicles,the trajectory planning system needs to conduct an accurate risk assessment of the current driving scene of autonomous vehicle,and plan the driving trajectory based on the result of the risk assessment,so as to ensure that the autonomous vehicle can perform safe and reasonable interaction with other traffic vehicles,which requires accurate prediction of the future trajectory of other traffic vehicles in a driving scene in each planning cycle.In a real urban traffic scenario,the vehicle's driving trajectory is not only constrained by prior knowledge,such as that about the road structure,traffic signs,and traffic rules,but also by uncertain posterior knowledge,including subjective driving intentions of the driver.However,the existing prediction methods cannot fully combine the prior and posterior knowledge in a driving scene and perform well only in a specific driving scene.In addition,the previous trajectory planning methods ignore the interaction between traffic vehicles and autonomous vehicles in a driving scene.Simple collision detection cannot accurately reflect the safety of the planning state,which lead to unsafe,unreasonable and unintelligent planning results.Aiming at the deficiencies in the current research,this thesis conducts an in-depth study on interaction-aware predictive trajectory planning system.The main contents are shown below:(1)In order to solve the problem that complex and changeable driving scenes are difficult to effectively express and solve the combination explosion problem caused by high dimension of input space,this paper proposes an ontology-based driving scene modeling and multi-attribute scene evaluation method.The ontology-based semantic description method is used to describe the related concepts and relationships of the road structure,traffic signs,traffic participants and other entities in the driving scene.The multi-attribute scene evaluation method is adopted to evaluate the road entities and entity relationships in the driving scene from the three aspects of safety,legality,and rationality.The prior knowledge in driving scene is effectively expressed and the input dimension is reduced.(2)Aiming at the problem that the existing prediction models cannot fully combine the prior and posterior knowledge in a driving scene and perform well only in a specific traffic scenario,a LSTM neural network driven by knowledge is proposed for trajectory prediction.Different from the existing research method that uses absolute coordinates to train the network,the driving intention obtained by knowledge inference is fitted into a prediction reference baseline,and the Frenet coordinates based on the prediction reference baseline are used to train the LSTM network.It is not necessary to annotate the training data set manually according to the specific driving scene.The proposed network can effectively combine the prior knowledge in the driving scenario,which solves the data sparseness problem and improves the scene adaptability and long-term prediction accuracy.(3)The existing trajectory planning methods based on MPC seldom mentioned the influence of the future trajectory of other traffic vehicles in the driving scene on trajectory planning.An interaction-aware predictive trajectory planning method is proposed in this paper.The spatio-temporal grid map is introduced into the collision detection,which can make more accurate risk assessment of the spatial path set.The S-T graph is introduced into the velocity planning,which can generate speed curves for different driving strategies.Since the interaction between traffic vehicles in the driving scenario is considered,the potential safety risk of the planning results can be effectively reduced.This paper conducted a large-scale and long-distance field test to verify the proposed trajectory planning method in real urban road environment.Experimental results show that the proposed method can effectively improve the reasonableness,safety and intelligence of trajectory planning,and ensure the safe and reasonable interaction between autonomous vehicles and other traffic vehicles,and efficiently improve the intelligent level of the autonomous vehicles...