Uncertainty-Aware Trajectory Prediction via Rule-Regularized Heteroscedastic Deep Classification

2025-04-17

AI Summary

To address weak out-of-distribution generalization and inadequate uncertainty modeling in autonomous driving trajectory prediction, this paper proposes a rule-regularized heteroscedastic deep classification framework. It reformulates trajectory prediction as an uncertainty-aware classification task, and for the first time integrates LLM-driven retrieval-augmented generation (RAG) to extract interpretable, rule-based driving priors with spectral-normalized heteroscedastic Gaussian processes (SN-HGP) to jointly disentangle epistemic and aleatoric uncertainties. The framework explicitly separates and calibrates both uncertainty types. Evaluated on nuScenes under low-data and cross-region settings, it reduces uncertainty calibration error by 32% and average displacement error (ADE) by 21%, with particularly pronounced improvements in high-risk scenarios such as intersections. It comprehensively surpasses existing state-of-the-art methods.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Planning, Routing, and Scheduling: Model-Based Reasoning
• Reasoning under Uncertainty: Sequential Decision Making

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Deep learning-based trajectory prediction models have demonstrated promising capabilities in capturing complex interactions. However, their out-of-distribution generalization remains a significant challenge, particularly due to unbalanced data and a lack of enough data and diversity to ensure robustness and calibration. To address this, we propose SHIFT (Spectral Heteroscedastic Informed Forecasting for Trajectories), a novel framework that uniquely combines well-calibrated uncertainty modeling with informative priors derived through automated rule extraction. SHIFT reformulates trajectory prediction as a classification task and employs heteroscedastic spectral-normalized Gaussian processes to effectively disentangle epistemic and aleatoric uncertainties. We learn informative priors from training labels, which are automatically generated from natural language driving rules, such as stop rules and drivability constraints, using a retrieval-augmented generation framework powered by a large language model. Extensive evaluations over the nuScenes dataset, including challenging low-data and cross-location scenarios, demonstrate that SHIFT outperforms state-of-the-art methods, achieving substantial gains in uncertainty calibration and displacement metrics. In particular, our model excels in complex scenarios, such as intersections, where uncertainty is inherently higher. Project page: https://kumarmanas.github.io/SHIFT/.

Problem

• Improves out-of-distribution generalization in trajectory prediction
• Combines uncertainty modeling with automated rule extraction
• Enhances robustness in low-data and cross-location scenarios

Innovation

• Combines uncertainty modeling with rule-based priors
• Uses heteroscedastic Gaussian processes for uncertainties
• Automates rule extraction via large language models