Vertical Parking Trajectory Planning With the Combination of Numerical Optimization Method and Gradient Lifting Decision Tree

Intelligent cyber-physical transportation systems (ICTS) have become the cutting-edge technology for the next generation of intelligent connected vehicle applications. Autonomous valet parking technique has significant application value in ICTS. A data-driven decision tree trajectory planning algorithm based on numerical optimization and machine learning is proposed to reduce computation time and improve the adaptability for vertical parking and enhance the transportation safety. Firstly, by learning the characteristics of vertical parking process and C-type parking constraints, a two-stage vertical parking dynamic optimization problem (DOP) is established. Accordingly, a two-stage Gaussian discretization method is proposed to solve the DOPs. Meanwhile, a trajectory dataset with 37,500 trajectories is constructed and each trajectory is verified by using the proposed posterior verification. Subsequently, the dataset is employed to drive the gradient boosting decision tree (GBDT) to establish the parking trajectory planning decision model for different types of vehicles, where 4 inputs and 1 output are considered. Simulation experiments show that the proposed method can effectively obtain the vertical parking trajectories with fast computation and good adaptability, where the calculation time is reduced by more than 99.8% when compared with multi-Gaussian pseudo-spectral method. In addition, compared with polynomial programming algorithm and hybrid A* algorithm, the computation time of the proposed method decreases by 84% on average, and trajectory planning is feasible under complex vertical parking scenarios, revealing the effectiveness of the proposed combination method.