Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-agent Traffic Prediction via Denoised Endpoint Distribution

Yao Liu, Ruoyu Wang, Yuanjiang Cao, Quan Z. Sheng, Lina Yao

TL;DR

The paper tackles high-speed multi-agent trajectory prediction by explicitly modeling intrinsic intent and uncertainty alongside spatio-temporal context. It introduces the Denoised Endpoint Distribution (DED) framework, which uses a diffusion process to generate endpoint distributions conditioned on spatio-temporal guidance and a Transformer to forecast agent-specific endpoints, followed by calibration against the endpoint distribution to produce accurate trajectories for all agents. Four modules—Guidance Information Extraction, Endpoint Distribution, Endpoint Prediction, and Trajectory Predictor—together yield state-of-the-art results on the NGSIM dataset while maintaining efficiency by focusing on endpoints rather than full trajectories. This approach advances real-time, multi-agent trajectory forecasting and sets the stage for future Plan-based extensions that further leverage intrinsic intent in autonomous navigation settings.

Abstract

The exploration of high-speed movement by robots or road traffic agents is crucial for autonomous driving and navigation. Trajectory prediction at high speeds requires considering historical features and interactions with surrounding entities, a complexity not as pronounced in lower-speed environments. Prior methods have assessed the spatio-temporal dynamics of agents but often neglected intrinsic intent and uncertainty, thereby limiting their effectiveness. We present the Denoised Endpoint Distribution model for trajectory prediction, which distinctively models agents' spatio-temporal features alongside their intrinsic intentions and uncertainties. By employing Diffusion and Transformer models to focus on agent endpoints rather than entire trajectories, our approach significantly reduces model complexity and enhances performance through endpoint information. Our experiments on open datasets, coupled with comparison and ablation studies, demonstrate our model's efficacy and the importance of its components. This approach advances trajectory prediction in high-speed scenarios and lays groundwork for future developments.

Multi-agent Traffic Prediction via Denoised Endpoint Distribution

TL;DR

The paper tackles high-speed multi-agent trajectory prediction by explicitly modeling intrinsic intent and uncertainty alongside spatio-temporal context. It introduces the Denoised Endpoint Distribution (DED) framework, which uses a diffusion process to generate endpoint distributions conditioned on spatio-temporal guidance and a Transformer to forecast agent-specific endpoints, followed by calibration against the endpoint distribution to produce accurate trajectories for all agents. Four modules—Guidance Information Extraction, Endpoint Distribution, Endpoint Prediction, and Trajectory Predictor—together yield state-of-the-art results on the NGSIM dataset while maintaining efficiency by focusing on endpoints rather than full trajectories. This approach advances real-time, multi-agent trajectory forecasting and sets the stage for future Plan-based extensions that further leverage intrinsic intent in autonomous navigation settings.

Abstract

The exploration of high-speed movement by robots or road traffic agents is crucial for autonomous driving and navigation. Trajectory prediction at high speeds requires considering historical features and interactions with surrounding entities, a complexity not as pronounced in lower-speed environments. Prior methods have assessed the spatio-temporal dynamics of agents but often neglected intrinsic intent and uncertainty, thereby limiting their effectiveness. We present the Denoised Endpoint Distribution model for trajectory prediction, which distinctively models agents' spatio-temporal features alongside their intrinsic intentions and uncertainties. By employing Diffusion and Transformer models to focus on agent endpoints rather than entire trajectories, our approach significantly reduces model complexity and enhances performance through endpoint information. Our experiments on open datasets, coupled with comparison and ablation studies, demonstrate our model's efficacy and the importance of its components. This approach advances trajectory prediction in high-speed scenarios and lays groundwork for future developments.
Paper Structure (20 sections, 19 equations, 5 figures, 3 tables)

This paper contains 20 sections, 19 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Sequential Models
  4. Generative Models
  5. Multimodal Prediction
  6. Methodology
  7. Problem Definition
  8. Model Overview
  9. Guidance Information Extraction
  10. Endpoint Distribution Module
  11. Endpoint Prediction Module
  12. Trajectory Predictor Module
  13. Experiments
  14. Datasets and Settings
  15. Experiments
  16. ...and 5 more sections

Figures (5)

  • Figure 1: Trajectory prediction scenarios. (a) Sequence Prediction, (b) Endpoint-based Prediction, (c) Denoised Endpoint Distribution-based Prediction (ours).
  • Figure 2: The overview of our model. Our model consists of four main modules, i.e. Guidance Information Extraction, Endpoint Distribution, Endpoint Prediction and Trajectory Prediction.
  • Figure 3: Diffusion and denoising model. (a) Diffusion and denoising process for image generation (b) Generation process for endpoint distribution.
  • Figure 4: Visualization of road vehicle trajectory prediction.
  • Figure 5: Visualization of endpoint calibration. The figure zooms in to show the endpoint information, which is in a straight line because it is in the last time interval.