Diffusion^2: Dual Diffusion Model with Uncertainty-Aware Adaptive Noise for Momentary Trajectory Prediction

2025-10-05

AI Summary

To address the challenge of predicting pedestrian trajectories emerging from occluded regions in autonomous driving and human–machine interaction—where observations are extremely short and historical trajectory data is unavailable—this paper proposes a bidirectional diffusion-based trajectory prediction framework. The method first employs reverse diffusion to synthesize plausible historical trajectories, then applies forward diffusion to forecast future paths, thereby enabling joint “completion–prediction” modeling for the first time. Key innovations include a bidirectional diffusion architecture and a dual-head parameterization scheme that explicitly captures aleatoric uncertainty inherent in trajectory data, as well as a time-adaptive noise module that dynamically modulates diffusion intensity to enhance robustness under extreme scenarios. Evaluated on the ETH/UCY and Stanford Drone datasets, the framework achieves state-of-the-art performance, significantly improving both accuracy and safety in predicting sudden-emergence pedestrian trajectories.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Learning to Search

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Accurate pedestrian trajectory prediction is crucial for ensuring safety and efficiency in autonomous driving and human-robot interaction scenarios. Earlier studies primarily utilized sufficient observational data to predict future trajectories. However, in real-world scenarios, such as pedestrians suddenly emerging from blind spots, sufficient observational data is often unavailable (i.e. momentary trajectory), making accurate prediction challenging and increasing the risk of traffic accidents. Therefore, advancing research on pedestrian trajectory prediction under extreme scenarios is critical for enhancing traffic safety. In this work, we propose a novel framework termed Diffusion^2, tailored for momentary trajectory prediction. Diffusion^2 consists of two sequentially connected diffusion models: one for backward prediction, which generates unobserved historical trajectories, and the other for forward prediction, which forecasts future trajectories. Given that the generated unobserved historical trajectories may introduce additional noise, we propose a dual-head parameterization mechanism to estimate their aleatoric uncertainty and design a temporally adaptive noise module that dynamically modulates the noise scale in the forward diffusion process. Empirically, Diffusion^2 sets a new state-of-the-art in momentary trajectory prediction on ETH/UCY and Stanford Drone datasets.

Problem

• Predicts pedestrian trajectories with limited observational data
• Addresses momentary trajectory challenges in autonomous driving
• Reduces prediction uncertainty through dual diffusion models

Innovation

• Dual diffusion models for backward and forward prediction
• Dual-head parameterization for aleatoric uncertainty estimation
• Temporally adaptive noise module for dynamic noise scaling