THOMAS: Trajectory Heatmap Output with learned Multi-Agent Sampling

Thomas Gilles (1) , Stefano Sabatini (2) , Dzmitry Tsishkou (2) , Bogdan Stanciulescu (1) , Fabien Moutarde (1)

1 CAOR - Centre de Robotique
2 Huawei Technologies France [Boulogne-Billancourt]

In this paper, we propose THOMAS, a joint multi-agent trajectory prediction framework allowing for an efficient and consistent prediction of multi-agent multimodal trajectories. We present a unified model architecture for simultaneous agent future heatmap estimation, in which we leverage hierarchical and sparse image generation for fast and memory-efficient inference. We propose a learnable trajectory recombination model that takes as input a set of predicted trajectories for each agent and outputs its consistent reordered recombination. This recombination module is able to realign the initially independent modalities so that they do no collide and are coherent with each other. We report our results on the Interaction multi-agent prediction challenge and rank 1 st on the online test leaderboard.

Domaines: Robotique [cs.RO]; Intelligence artificielle [cs.AI]; Apprentissage [cs.LG]

Thomas Gilles, Stefano Sabatini, Dzmitry Tsishkou, Bogdan Stanciulescu, Fabien Moutarde. THOMAS: Trajectory Heatmap Output with learned Multi-Agent Sampling. International Conference on Learning Representations, Apr 2022, Virtuel, France. ⟨hal-03683506⟩