Probabilistic Crowd GAN: Multimodal Pedestrian Trajectory Prediction Using a Graph Vehicle-Pedestrian Attention Network

Abstract

Understanding and predicting the intention of pedestrians is essential to\nenable autonomous vehicles and mobile robots to navigate crowds. This problem\nbecomes increasingly complex when we consider the uncertainty and multimodality\nof pedestrian motion, as well as the implicit interactions between members of a\ncrowd, including any response to a vehicle. Our approach, Probabilistic Crowd\nGAN, extends recent work in trajectory prediction, combining Recurrent Neural\nNetworks (RNNs) with Mixture Density Networks (MDNs) to output probabilistic\nmultimodal predictions, from which likely modal paths are found and used for\nadversarial training. We also propose the use of Graph Vehicle-Pedestrian\nAttention Network (GVAT), which models social interactions and allows input of\na shared vehicle feature, showing that inclusion of this module leads to\nimproved trajectory prediction both with and without the presence of a vehicle.\nThrough evaluation on various datasets, we demonstrate improvements on the\nexisting state of the art methods for trajectory prediction and illustrate how\nthe true multimodal and uncertain nature of crowd interactions can be directly\nmodelled.\n