From Marginal to Joint Predictions: Evaluating Scene-Consistent Trajectory Prediction Approaches for Automated Driving

2025-07-07

AI Summary

Independent (marginal) trajectory prediction in autonomous driving leads to suboptimal planning due to neglect of scene-wide consistency among agents. Method: We systematically investigate scene-consistent joint trajectory prediction, proposing the first unified evaluation framework encompassing three paradigms: post-hoc correction, end-to-end joint modeling, and generative modeling. Our framework integrates sequence modeling, interaction-aware neural architectures, and diffusion/adversarial generation techniques, and introduces novel scene-level metrics quantifying physical plausibility and social acceptability. Results: Joint prediction significantly improves the physical realism and social compliance of multi-agent motion. Empirical analysis reveals clear trade-offs across paradigms in multimodal fidelity, prediction accuracy, and inference efficiency. To our knowledge, this work is the first to quantitatively characterize the performance limits of mainstream methods on real-world traffic scenarios, establishing a reproducible benchmark and principled design guidelines for consistent trajectory forecasting.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Multiagent Systems: Adversarial Agents
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Accurate motion prediction of surrounding traffic participants is crucial for the safe and efficient operation of automated vehicles in dynamic environments. Marginal prediction models commonly forecast each agent's future trajectories independently, often leading to sub-optimal planning decisions for an automated vehicle. In contrast, joint prediction models explicitly account for the interactions between agents, yielding socially and physically consistent predictions on a scene level. However, existing approaches differ not only in their problem formulation but also in the model architectures and implementation details used, making it difficult to compare them. In this work, we systematically investigate different approaches to joint motion prediction, including post-processing of the marginal predictions, explicitly training the model for joint predictions, and framing the problem as a generative task. We evaluate each approach in terms of prediction accuracy, multi-modality, and inference efficiency, offering a comprehensive analysis of the strengths and limitations of each approach. Several prediction examples are available at https://frommarginaltojointpred.github.io/.

Problem

• Evaluating scene-consistent trajectory prediction for automated driving
• Comparing marginal vs joint prediction models' accuracy and efficiency
• Analyzing approaches for socially and physically consistent motion prediction

Innovation

• Post-processing marginal predictions for joint accuracy
• Training models explicitly for joint predictions
• Framing prediction as a generative task