Humans have the innate ability to read one another, and plan their future path taking into account what might happen in the future. Thus, a crucial requirement for the success of mobility applications like autonomous driving and social robots is the ability to reason about the evolution of a scene. Human trajectory forecasting refers to this task of forecasting future motion of humans in crowds.
Our recent research tackles four fundamental challenges for socially-aware trajectory forecasting (i) modelling social interactions [1], (ii) sampling multiple socially-compliant futures [2], (iii) injecting interpretability into the decision-making [3], and (iv) efficient adaptation to unseen scenarios at test-time [4].
[1] Human Trajectory Forecasting in Crowds: A Deep Learning Perspective
[2] Safety-compliant Generative Adversarial Networks for Human Trajectory Forecasting
[3] Interpretable Social Anchors for Human Trajectory Forecasting in Crowds
[4] Motion Style Transfer: Modular Low-Rank Adaptation for Deep Motion Forecasting