Cooperative Motion Planning and Decision-Making for CAVs at Roundabouts: A Data-efficient Learning-Based Iterative Optimization Method

The behavior of connected and autonomous vehicles (CAVs) in traffic environments is very complex. Giving efficient cooperative driving strategies in traffic intersection scenarios is still a challenge, especially at roundabout intersections. In this paper, we propose a decision-making and motion planning method for roundabouts by an iterative learning-based optimization. Time-optimal safe driving strategies are given in a hybrid architecture. Based on the crossing mode modeling, a data-efficient learning-based iterative optimization (DELIO) method for offline trajectory generation is proposed. We build the optimization problem by introducing stage and terminal costs and describing the static and dynamic constraints using Support Vector Machine (SVM) method. Iterative safety constraints are constructed to store historical data for closed-loop efficient data-driven learning. The algorithm finally converges to a time-optimal collision-free trajectory only after several iterations. In order to rationally apply the optimal trajectories, a multi-vehicle cooperative decision-making and motion planning method based on adaptive Monte Carlo tree search (AMCTS) is proposed. The algorithm eliminates inter-group conflicts and quickly converges to the decision sequence of the optimal grouping. We validate our method in a typical roundabout intersection scenario. Results show that our method enables vehicles to pass through the roundabout intersection efficiently and safely, which significantly improves the traffic efficiency.