IMR method tops Argoverse 2 benchmark for multi-agent trajectory prediction
New iterative regression model improves autonomous vehicle safety by predicting multi-agent trajectories with top accuracy.
A new paper from researchers Honglin Wang, Shiyao Pan, and Yun-Fu Liu introduces IMR: Iterative Mode-World Weighted Regression for Multi-Agent Trajectory Prediction. The method addresses critical limitations in existing prediction-based and anchor-based approaches—namely mode diversity and prediction accuracy. IMR introduces a novel mode-world weighted regression loss that bridges the gap between mode diversity and world ranking, mitigating mode collapse while simultaneously improving top-1 confidence. Additionally, an iterative decoder recurrently and segmentally generates trajectories, boosting prediction accuracy.
Experimental results demonstrate that IMR achieves top ranking on the Argoverse 2 multi-agent motion forecasting benchmark, outperforming all other methods. The approach has significant implications for automated vehicles, enabling more accurate understanding of surrounding vehicle intentions and reducing safety risks from inadequate assessments. The paper is available on arXiv (2607.05705) and spans robotics, AI, computer vision, and machine learning.
- IMR uses a mode-world weighted regression loss to improve both mode diversity and prediction accuracy simultaneously.
- An iterative decoder generates trajectories recurrently and segmentally for higher precision.
- The method ranks first on the Argoverse 2 multi-agent motion forecasting benchmark, outperforming all prior techniques.
Why It Matters
Better trajectory prediction means safer autonomous vehicles that can anticipate multiple agent behaviors and avoid collisions.