Accurate trajectory prediction is critical for safe and efficient decision-making in autonomous driving. However, this task remains challenging due to the inherent multimodality of agent behaviors, the requirement for long-term temporal consistency and structured interaction modeling, and strict real-time constraints. To address these challenges, this paper proposes MDTF, a novel Mamba-enhanced Diffusion-based Trajectory Forecasting framework within an end-to-end encoder-decoder architecture. At the core of the encoder is a hybrid architecture featuring a Mamba block and a gated recurrent unit (GRU), which synergistically leverages Mamba’s linear-time efficiency in modeling long-range dependencies and GRU’s proficiency in capturing local sequential dynamics. For the decoder, an anchor-based truncated diffusion model is introduced to facilitate efficient trajectory generation. This approach uniquely harnesses the strong generative power of diffusion models to produce diverse trajectories while remaining computationally efficient for real-time applications. Extensive experiments on the nuScenes dataset demonstrate that MDTF achieves superior performance in terms of minADE, minFDE, and miss rates. The proposed framework offers a robust solution for trajectory forecasting in complex multi-agent environments.
Sui et al. (Wed,) studied this question.
Synapse has enriched 4 closely related papers on similar clinical questions. Consider them for comparative context: