QPNet: Lane‐changing trajectory planning combining quadratic programming and neural network under the convex optimization framework

Abstract

Lane-changing is a basic driving behaviour, which largely impacts on traffic safety and efficiency. With the development of technology, the automated lane-changing system has attracted extensive attention. Among it, the trajectory planning part is a challenging problem owing to the complexity and diversity of the driving situations. The planner requires the real-time capability to produce safe and comfortable trajectories for coping with the dynamically changing environment. Based on this, the paper proposes a lane-changing trajectory planning model in dynamic driving environments. The model constructs a neural network to predict the end position of the ego vehicle, and then adopts the mathematical programming method to solve the optimal lane-changing trajectory that guarantees the ride safety and comfort. With the assistance of the neural network, the lane-changing trajectory planning problem is converted into a quadratic programming (QP) model, thereby achieving rapid solution of the model. Moreover, to train the proposed algorithm, a novel approach for generating the scenario data is designed, which can generate rich and diverse traffic scenarios at a low cost. The simulation results show that the proposed model can plan a lane-changing trajectory quickly and effectively, and the ego vehicle can follow the planned trajectory safely and comfortably.

Dynamic Trajectory Planning and Optimization for Automated Driving on Ice and Snow Covered Road

“…In contrast to algorithms such as graph search and sampling, numerical optimization describes trajectory planning as a multi-objective optimization problem, which is solved by quadratic programming to obtain the optimal trajectory [10]. The biggest advantage of this algorithm is that the optimal solution space is continuous and there is no large jump in the optimal solution between adjacent frames [11], [12]. However, the long time required for trajectory point optimization iterations leads to a slow solution speed for some frames, which cannot meet the real-time requirements in autonomous driving scenarios.…”

Real-Time Path Planning for Obstacle Avoidance in Intelligent Driving Sightseeing Cars Using Spatial Perception

“…These algorithms encompass parameterized curves, including but not limited to, circular curves, spiral curves, trigonometric curves, and polynomial curves. Key merits of these algorithms encompass their computational efficiency, intuitive design, and high precision [22]. Such algorithms are prevalently employed in studies focused on obstacle avoidance trajectory planning for intelligent driving vehicles.…”