Evidential Uncertainty Estimation for Multi-Modal Trajectory Prediction

2025-03-07

AI Summary

Autonomous driving multimodal trajectory prediction faces challenges in jointly modeling behavioral and perceptual uncertainties, and existing methods struggle to simultaneously quantify position-level and mode-level uncertainties. This paper proposes the first evidential deep learning framework that jointly estimates both uncertainties in a single forward pass. Our approach innovatively integrates the Normal-Inverse-Gamma distribution—used to model aleatoric uncertainty in predicted positions—with the Dirichlet distribution—employed to capture epistemic uncertainty over trajectory modes. We further introduce an uncertainty-driven importance sampling strategy to enhance training efficiency and generalization. Evaluated on Argoverse 1 and 2, our method achieves state-of-the-art prediction accuracy while delivering well-calibrated uncertainty estimates. Notably, it significantly outperforms sampling-based baselines in inference speed, enabling real-time deployment without compromising reliability or expressiveness.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Multiagent Systems: Multiagent Systems under Uncertainty

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• 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

Abstract

Accurate trajectory prediction is crucial for autonomous driving, yet uncertainty in agent behavior and perception noise makes it inherently challenging. While multi-modal trajectory prediction models generate multiple plausible future paths with associated probabilities, effectively quantifying uncertainty remains an open problem. In this work, we propose a novel multi-modal trajectory prediction approach based on evidential deep learning that estimates both positional and mode probability uncertainty in real time. Our approach leverages a Normal Inverse Gamma distribution for positional uncertainty and a Dirichlet distribution for mode uncertainty. Unlike sampling-based methods, it infers both types of uncertainty in a single forward pass, significantly improving efficiency. Additionally, we experimented with uncertainty-driven importance sampling to improve training efficiency by prioritizing underrepresented high-uncertainty samples over redundant ones. We perform extensive evaluations of our method on the Argoverse 1 and Argoverse 2 datasets, demonstrating that it provides reliable uncertainty estimates while maintaining high trajectory prediction accuracy.

Problem

• Estimates positional and mode uncertainty in multi-modal trajectory prediction.
• Improves efficiency by inferring uncertainty in a single forward pass.
• Enhances training efficiency using uncertainty-driven importance sampling.

Innovation

• Evidential deep learning for uncertainty estimation
• Normal Inverse Gamma and Dirichlet distributions
• Uncertainty-driven importance sampling for training