Predicting Vehicle Behavior Using Automotive Radar and Recurrent Neural Networks

We present a Long Short Term Memory (LSTM) encoder-decoder architecture to anticipate the future positions of vehicles in a road network given several seconds of historical observations and associated map features. Unlike existing architectures, the proposed method incorporates and updates the surrounding vehicle information in both the encoder and decoder, making use of dynamically predicted new data for accurate prediction in longer time horizons. It seamlessly performs four tasks: the first task encodes a feature given the past observations, the second task estimates future maneuvers given the encoded state, the third task predicts the future motion given the estimated maneuvers and the initially encoded states, and the fourth task estimates future trajectory given the encoded state and the predicted maneuvers and motions. Experiments on a public benchmark and a new, publicly available radar dataset demonstrate that our approach can equal or surpass the state-of-the-art for long term trajectory prediction.

Automated summary

The proposed LSTM encoder-decoder architecture predicts future vehicle positions in a road network utilizing dynamically updated surrounding vehicle information for accurate long term predictions. Experimental results demonstrate the method can match or outperform existing state-of-the-art approaches in long term trajectory prediction.