Von Mises Based Uncertainty Quantification for Closely Spaced Automotive Radar Targets

2026-06-30 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

This work investigates uncertainty-aware deep learning for direction of arrival (DOA) estimation in automotive radar, comparing a circular-statistics-based von Mises (VM) ensemble (ENS) with an evidential deep learning (EDL) framework. The ENS approach produces angular predictions parameterized by (mu, kappa), offering interpretable uncertainty aligned with directional geometry. The EDL framework uses a normal inverse gamma formulation, yielding a Student t predictive distribution. Performance was evaluated under in-distribution and multiple out-of-distribution conditions using risk coverage and ROC/AUROC analyses. Results show ENS achieves lower uncertainty nominally and stronger sensitivity to perturbations, while EDL provides smoother uncertainty variation and slightly improved ranking consistency. Crucially, ENS enables direct probabilistic integration into association modules via closed-form VM likelihoods, facilitating a unified detection tracking pipeline. This highlights a trade-off between geometric consistency and statistical generality.

Key takeaway

For automotive radar engineers designing sensor fusion systems, choosing an uncertainty quantification method requires balancing specific needs. If your system prioritizes geometric consistency and direct probabilistic integration into tracking modules, the von Mises ensemble approach offers a unified detection tracking pipeline. Conversely, if smoother uncertainty variation and statistical generality are paramount, the evidential deep learning framework might be more suitable. Evaluate both based on your specific operational environment and perturbation resilience requirements.

Key insights

Compares von Mises ensemble and evidential deep learning for uncertainty-aware automotive radar DOA estimation.

Principles

• VM ensemble provides interpretable uncertainty aligned with directional geometry.
• Evidential deep learning offers smoother uncertainty variation.

Method

Compares a circular-statistics-based von Mises ensemble (parameterized by (mu, kappa)) with an evidential deep learning framework (normal inverse gamma yielding Student t distribution) for DOA estimation.

In practice

• VM ensemble enables direct probabilistic integration into association modules.
• Facilitates unified detection tracking pipelines.

Topics

• Automotive Radar
• Direction of Arrival
• Uncertainty Quantification
• Deep Learning
• von Mises Ensemble
• Evidential Deep Learning
• Sensor Fusion

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• Von Mises Based Uncertainty Quantification for Closely Spaced Automotive Radar Targets — Von Mises ensembles offer geometrically consistent uncertainty for automotive radar DOA, balancing against statistical generality.
• Introduction to Deep Evidential Regression for Uncertainty Quantification — Evidential Deep Learning quantifies both epistemic and aleatoric uncertainty in a single inference pass.
• Architecture-Adaptive Uncertainty Fusion for Deepfake Detection — Correlation-Optimized Fusion (COF) adaptively combines multiple uncertainty sources to predict deepfake detection errors, improving cross-domain reliability.
• Diffusion-Based Stochastic Operator Networks for Uncertainty Quantification in Stochastic Partial Differential Equations — SON combines DeepONet with SNNs and a two-phase training for efficient SPDE uncertainty quantification.
• Uncertainty-Aware Pedestrian Attribute Recognition via Evidential Deep Learning — UAPAR uses Evidential Deep Learning to quantify prediction uncertainty in pedestrian attribute recognition, enhancing robustness.
• [D] Conformal Prediction vs naive thresholding to represent uncertainty — Conformal Prediction offers theoretical guarantees for uncertainty quantification, unlike naive thresholding.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.