А hybrid agent-centric and scene-centric approach for multi-agent trajectory prediction

Abstract

Accurately predicting the future trajectories of agents in autonomous driving is crucial for safe navigation and decision-making. Traditional trajectory prediction models have limitations when dealing with complex multi-agent interactions. In this paper, we propose a hybrid approach that leverages the strengths of both agent-centric and scene-centric models by using agent-centric normalization for dynamic agents and a scene-centric framework for static map elements.