Scenario-aware Uncertainty Quantification for Trajectory Prediction with Statistical Guarantees

TL;DR

This work tackles the challenge of uncertainty quantification in trajectory prediction for autonomous driving by introducing a scenario-aware conformal prediction framework operating in the Frenet reference frame. It combines online/offline CP calibration with CopulaCPTS to model temporal dependencies, and a trajectory reliability discriminator to map prediction intervals to reliable/unreliable segments across driving contexts. A risk-aware discriminator further identifies critical points by fusing longitudinal and lateral uncertainties, enabling scenario-specific, actionable reliability signals for planners. Evaluations on the nuPlan dataset show calibrated prediction intervals and meaningful segmentation of trajectories by reliability, suggesting practical benefits for uncertainty-aware planning and corner-case detection. Overall, the approach offers statistically guaranteed UQ integrated with planning-oriented reliability assessment tailored to heterogeneous driving scenarios.

Abstract

Reliable uncertainty quantification in trajectory prediction is crucial for safety-critical autonomous driving systems, yet existing deep learning predictors lack uncertainty-aware frameworks adaptable to heterogeneous real-world scenarios. To bridge this gap, we propose a novel scenario-aware uncertainty quantification framework to provide the predicted trajectories with prediction intervals and reliability assessment. To begin with, predicted trajectories from the trained predictor and their ground truth are projected onto the map-derived reference routes within the Frenet coordinate system. We then employ CopulaCPTS as the conformal calibration method to generate temporal prediction intervals for distinct scenarios as the uncertainty measure. Building upon this, within the proposed trajectory reliability discriminator (TRD), mean error and calibrated confidence intervals are synergistically analyzed to establish reliability models for different scenarios. Subsequently, the risk-aware discriminator leverages a joint risk model that integrates longitudinal and lateral prediction intervals within the Frenet coordinate to identify critical points. This enables segmentation of trajectories into reliable and unreliable segments, holding the advantage of informing downstream planning modules with actionable reliability results. We evaluated our framework using the real-world nuPlan dataset, demonstrating its effectiveness in scenario-aware uncertainty quantification and reliability assessment across diverse driving contexts.

Figures (9)

• Figure 1: Schematic of the proposed framework. The framework initially takes real-world dataset and processes it through a neural network-based trajectory predictor (left top). CP calibrator generates prediction intervals (left bottom). Scenario-based reliability model further constructs a mapping between predicted intention deviation and prediction intervals. Risk-aware discriminator then identifies critical points beyond which the trajectory segment becomes unreliable (middle). Future benefited stakeholders may include uncertainty-aware motion planning, corner case detection, and predictor evaluation (right).
• Figure 2: An illustration of coordinate transition. The reference route, extracted from the map, serves as a basis for aligning trajectories. Future ground truth (red) as well as multimodal predicted trajectories of all agents are projected onto this route, ensuring a structured representation of motion.
• Figure 3: An illustration of the critical threshold for trajectory assessment. The TRD model determines the critical point, beyond which the predicted trajectory is considered unreliable, ensuring only the usable and reliable segment will be utilized.
• Figure 4: Qualitative results for prediction intervals under different $\alpha$ and time horizons. Over longer horizons, the predicted trajectory (gradient-colored) deviates more from the ground truth (red), highlighting the difficulty in accurately capturing the vehicle’s intentions over extended periods.
• Figure 5: Scenario-aware Reliability Model for $s$ direction.