Multi-Agent Tensor Fusion for Contextual Trajectory Prediction
This addresses the problem of predicting others' trajectories for autonomous driving, representing an incremental improvement.
The paper tackles trajectory prediction for autonomous driving by modeling interactions and constraints jointly with a Multi-Agent Tensor Fusion network, achieving state-of-the-art accuracy on highway and pedestrian datasets.
Accurate prediction of others' trajectories is essential for autonomous driving. Trajectory prediction is challenging because it requires reasoning about agents' past movements, social interactions among varying numbers and kinds of agents, constraints from the scene context, and the stochasticity of human behavior. Our approach models these interactions and constraints jointly within a novel Multi-Agent Tensor Fusion (MATF) network. Specifically, the model encodes multiple agents' past trajectories and the scene context into a Multi-Agent Tensor, then applies convolutional fusion to capture multiagent interactions while retaining the spatial structure of agents and the scene context. The model decodes recurrently to multiple agents' future trajectories, using adversarial loss to learn stochastic predictions. Experiments on both highway driving and pedestrian crowd datasets show that the model achieves state-of-the-art prediction accuracy.