Li, Lanxin (Shandong Jiaotong University), Pei, Wenhui (Shandong Jiaotong University), Zhang, Yu (Shandong Jiaotong University), Ma, Baosen (Shandong Jiaotong University)

Motion Planning for Automatic Lane Changing Based on Elman Neural Network and Mixed Integer Optimisation

Scheduled for presentation during the Regular session "Vehcile Control II" (ThB3), Thursday, October 31, 2024, 14:30−14:50, Room T3

7th IFAC Conference on Engine and Powertrain Control, Simulation and Modeling, Oct 30 - Nov 1, 2024, Dalian, China

This information is tentative and subject to change. Compiled on May 16, 2025

Abstract

Abstract: To address issues such as unsmooth trajectories and computational complexity in lane-changing planning for intelligent driving vehicles, this paper proposes a hierarchical learning Elamn-Mixed-Integer Optimization Programs (Elamn-MIQP) training model to enhance the performance of lane-changing planning in intelligent vehicles. First, the lane-changing problem for intelligent vehicles is mathematically modeled, introducing logical constraints to solve traditional obstacle avoidance issues. Then, a mixed-integer quadratic optimization algorithm is used to generate the lane-changing trajectories for intelligent vehicles. These generated trajectories serve as the training dataset for the lane-changing action model. This training model utilizes a cubic polynomial kernel support vector machine to classify and learn from the lane-changing trajectories. Finally, the Elamn network optimizes the trajectories to output the final lane-changing trajectory. Simulation results show that the proposed approach produces lane-changing trajectories superior to those generated by MIQP. The proposed method also demonstrates a certain level of generalization capability, effectively generating lane-changing trajectories in congested traffic scenarios and achieves a correction rate of 24% for trajectories that do not conform to the lane change scenario.