WSiP: Wave Superposition Inspired Pooling for Dynamic Interactions-Aware Trajectory Prediction

Predicting motions of surrounding vehicles is critically important to help autonomous driving systems plan a safe path and avoid collisions. Although recent social pooling based LSTM models have achieved significant performance gains by considering the motion interactions between close each other, vehicle trajectory prediction still remains as challenging research issue due dynamic high-order in real complex scenarios. To this end, we propose wave superposition inspired (Wave-pooling for short) method dynamically aggregating from both local global neighbor vehicles. Through modeling with amplitude phase, Wave-pooling can more effectively represent states capture their superposition. By integrating Wave-pooling, an encoder-decoder learning framework named WSiP also proposed. Extensive experiments conducted on two public highway datasets NGSIM highD verify effectiveness comparison current state-of-the-art baselines. More importantly, result interpretable interaction strength be intuitively reflected phase difference. The code work publicly available at https://github.com/Chopin0123/WSiP.