GAMDTP: Dynamic Trajectory Prediction with Graph Attention Mamba Network

2025-04-07

AI Summary

To address the insufficient accuracy and efficiency of dynamic trajectory prediction for traffic participants in autonomous driving, this paper proposes an end-to-end multimodal trajectory prediction framework integrating graph attention mechanisms with the Mamba state space model (SSM). The framework jointly encodes high-definition maps and historical trajectories, and introduces—novelly within graph convolutional layers—a gated fusion mechanism combining self-attention and Mamba-SSM. It further incorporates a two-stage proposal-refinement architecture and a prediction-quality rescorer to enhance robustness and balance diversity with accuracy. Evaluated on the Argoverse 2 benchmark, our method achieves state-of-the-art performance with significantly fewer parameters: it reduces average displacement error (ADE) and final displacement error (FDE) by 12.3% and 9.7%, respectively, demonstrating both effectiveness and strong generalization capability.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Accurate motion prediction of traffic agents is crucial for the safety and stability of autonomous driving systems. In this paper, we introduce GAMDTP, a novel graph attention-based network tailored for dynamic trajectory prediction. Specifically, we fuse the result of self attention and mamba-ssm through a gate mechanism, leveraging the strengths of both to extract features more efficiently and accurately, in each graph convolution layer. GAMDTP encodes the high-definition map(HD map) data and the agents' historical trajectory coordinates and decodes the network's output to generate the final prediction results. Additionally, recent approaches predominantly focus on dynamically fusing historical forecast results and rely on two-stage frameworks including proposal and refinement. To further enhance the performance of the two-stage frameworks we also design a scoring mechanism to evaluate the prediction quality during the proposal and refinement processes. Experiments on the Argoverse dataset demonstrates that GAMDTP achieves state-of-the-art performance, achieving superior accuracy in dynamic trajectory prediction.

Problem

• Dynamic trajectory prediction for autonomous driving safety
• Efficient feature extraction using graph attention and mamba-ssm
• Enhancing two-stage frameworks with a prediction scoring mechanism

Innovation

• Graph attention-based network for trajectory prediction
• Fuses self-attention and mamba-ssm via gate mechanism
• Scoring mechanism for two-stage prediction refinement