It Is Not the Journey but the Destination: Endpoint Conditioned\n Trajectory Prediction

Human trajectory forecasting with multiple socially interacting agents is of\ncritical importance for autonomous navigation in human environments, e.g., for\nself-driving cars and social robots. In this work, we present Predicted\nEndpoint Conditioned Network (PECNet) for flexible human trajectory prediction.\nPECNet infers distant trajectory endpoints to assist in long-range multi-modal\ntrajectory prediction. A novel non-local social pooling layer enables PECNet to\ninfer diverse yet socially compliant trajectories. Additionally, we present a\nsimple "truncation-trick" for improving few-shot multi-modal trajectory\nprediction performance. We show that PECNet improves state-of-the-art\nperformance on the Stanford Drone trajectory prediction benchmark by ~20.9% and\non the ETH/UCY benchmark by ~40.8%. Project homepage:\nhttps://karttikeya.github.io/publication/htf/\n