Efficient Pedestrian Trajectory Prediction via Explicit Pose Interaction and Dynamic Multi-Level Game Theory

Pedestrian trajectory prediction is crucial for autonomous driving safety in shared spaces, but existing methods struggle with poor physical interpretability, inefficient reasoning, and kinematic inconsistencies. This paper proposes an efficient pedestrian trajectory prediction framework via explicit pose interaction and dynamic multi-level game theory. Our approach introduces three key innovations: explicit pose-based interaction encoding that directly leverages physical quantities like relative distance and heading difference; adaptive game reasoning depth that dynamically adjusts based on scene complexity; and joint optimization of B-spline trajectories and game strategies through pedestrian-specific kinematic constraints. Experiments on the Argoverse 2 dataset demonstrate superior performance compared to state-of-the-art methods while maintaining computational efficiency suitable for real-time onboard deployment.