Collaborative-Distilled Diffusion Models (CDDM) for Accelerated and Lightweight Trajectory Prediction

2025-10-01

AI Summary

To address the challenges of large parameter counts, slow sampling, and poor real-time deployability of diffusion models in autonomous driving trajectory prediction, this paper proposes the Collaborative Distillation Diffusion Model (CDDM). CDDM employs collaborative progressive distillation to simultaneously compress model parameters and reduce sampling steps. It introduces a dual-signal regularization loss that jointly leverages teacher-model outputs and the ground-truth data distribution, thereby enhancing generalization and robustness. Evaluated on ETH-UCY and nuScenes benchmarks, CDDM achieves state-of-the-art performance with only 231K parameters and 4/2 sampling steps—retaining 96.2% of baseline ADE and 95.5% of FDE. The method yields a 161× model compression ratio, a 31× sampling speedup, and an end-to-end latency of just 9 ms, significantly advancing lightweight design and real-time inference capability for diffusion-based trajectory prediction.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Trajectory prediction is a fundamental task in Autonomous Vehicles (AVs) and Intelligent Transportation Systems (ITS), supporting efficient motion planning and real-time traffic safety management. Diffusion models have recently demonstrated strong performance in probabilistic trajectory prediction, but their large model size and slow sampling process hinder real-world deployment. This paper proposes Collaborative-Distilled Diffusion Models (CDDM), a novel method for real-time and lightweight trajectory prediction. Built upon Collaborative Progressive Distillation (CPD), CDDM progressively transfers knowledge from a high-capacity teacher diffusion model to a lightweight student model, jointly reducing both the number of sampling steps and the model size across distillation iterations. A dual-signal regularized distillation loss is further introduced to incorporate guidance from both the teacher and ground-truth data, mitigating potential overfitting and ensuring robust performance. Extensive experiments on the ETH-UCY pedestrian benchmark and the nuScenes vehicle benchmark demonstrate that CDDM achieves state-of-the-art prediction accuracy. The well-distilled CDDM retains 96.2% and 95.5% of the baseline model's ADE and FDE performance on pedestrian trajectories, while requiring only 231K parameters and 4 or 2 sampling steps, corresponding to 161x compression, 31x acceleration, and 9 ms latency. Qualitative results further show that CDDM generates diverse and accurate trajectories under dynamic agent behaviors and complex social interactions. By bridging high-performing generative models with practical deployment constraints, CDDM enables resource-efficient probabilistic prediction for AVs and ITS. Code is available at https://github.com/bingzhangw/CDDM.

Problem

• Accelerating diffusion models for real-time trajectory prediction
• Reducing model size while maintaining prediction accuracy
• Enabling lightweight probabilistic forecasting in autonomous systems

Innovation

• Progressive knowledge distillation from teacher to student model
• Dual-signal regularized loss for robust performance
• Lightweight model with reduced parameters and sampling steps