Abstract:

The collision avoidance path planning of artificial potential field algorithm in moving scene rarely considers the timing coupling effect with the future trajectory of the preceding vehicle， and basically regards the preceding vehicle in every planning cycle as static. A quasi-dynamic path planning is thereby carried out through the rolling update of different planning cycles， resulting in unreasonable and poor consistency of the planned path. In this paper， the prediction trajectory of the preceding vehicle is accordingly integrated into the intelligent vehicle path planning process through time-series coupling correlation. First， the attraction field and repulsion field models of the improved artificial potential field algorithm are constructed， and the position of the preceding vehicle in the repulsion field is proposed to be dynamically updated according to its prediction value in each planning cycle. Then a long-term prediction algorithm for the trajectory of the preceding vehicle by combining cluster recognition of driving intention and discrete optimization， and a short-term prediction algorithm for the trajectory of the preceding vehicle by combining cluster recognition of kinematics model and unscented Kalman filter algorithm are proposed. Subsequently， the weighted fusion is performed through the Sigmoid function to complete the prediction of the trajectory of the preceding vehicle. Finally， simulation results in scenarios such as driving out a high way and cutting in by adjacent vehicle indicate that， compared with the traditional APF algorithm， the proposed improved artificial potential field dynamic path planning algorithm can obtain more reasonable and consistent planning results.

Key words: intelligent vehicle, artificial potential field, dynamic path planning, trajectory prediction