Uncertainty-Aware Hybrid Machine Learning in Virtual Sensors for Vehicle Sideslip Angle Estimation

TL;DR

This work targets accurate estimation of the vehicle sideslip angle $eta$ under cost constraints by developing a Uncertainty-Aware Hybrid Learning (UAHL) virtual sensor that fuses a transformer-based ML model with vehicle motion models, each with uncertainty quantification. The ML component uses an Informer*-style time-series predictor to forecast $eta$ and its uncertainty, while two vehicle dynamics models provide complementary estimates; three fusion strategies—Expert Fusion, Deep Fusion, and Gaussian Regression Fusion—combine these sources using quantified uncertainties. A novel Real-world Vehicle State Estimation Dataset (ReV-StED) is introduced to support evaluation, including 0.9 million synchronized samples from 5 drivers with diverse maneuvers. Across extensive experiments and ablations, the UAHL variants, particularly the Deep Fusion approach, demonstrate superior accuracy and reliability in VSA estimation, including under challenging dynamic conditions, highlighting the approach’s potential for enhancing active safety in autonomous vehicles.

Abstract

Precise vehicle state estimation is crucial for safe and reliable autonomous driving. The number of measurable states and their precision offered by the onboard vehicle sensor system are often constrained by cost. For instance, measuring critical quantities such as the Vehicle Sideslip Angle (VSA) poses significant commercial challenges using current optical sensors. This paper addresses these limitations by focusing on the development of high-performance virtual sensors to enhance vehicle state estimation for active safety. The proposed Uncertainty-Aware Hybrid Learning (UAHL) architecture integrates a machine learning model with vehicle motion models to estimate VSA directly from onboard sensor data. A key aspect of the UAHL architecture is its focus on uncertainty quantification for individual model estimates and hybrid fusion. These mechanisms enable the dynamic weighting of uncertainty-aware predictions from machine learning and vehicle motion models to produce accurate and reliable hybrid VSA estimates. This work also presents a novel dataset named Real-world Vehicle State Estimation Dataset (ReV-StED), comprising synchronized measurements from advanced vehicle dynamic sensors. The experimental results demonstrate the superior performance of the proposed method for VSA estimation, highlighting UAHL as a promising architecture for advancing virtual sensors and enhancing active safety in autonomous vehicles.

Figures (5)

• Figure 1: Overview of the proposed architecture for virtual sensor.
• Figure 2: The architecture consists of three main components: A model, , and an uncertainty-aware hybrid fusion module. The onboard sensor data ${\bm{x}}_t$ is the input for both and . The model, predicts $\beta_{\text{ml}}$ with model uncertainty $\delta_{\text{ml}}$ using the Uncertainty Quantification (UQ) block. The output their predictions, $\beta_{\text{vm1}}$ and $\beta_{\text{vm2}}$ with their respective uncertainties, $\delta_{\text{vm1}}$ and $\delta_{\text{vm2}}$. Finally, the hybrid fusion module leverages uncertainty-aware inputs to estimate the for the virtual sensor.
• Figure 3: of yaw rate residual vs residual for .
• Figure 4: Mean absolute error of estimates, calculated for each bin of 0.125° along the ground truth $\beta$.
• Figure 5: Virtual sensor estimates for in "figure-eight" maneuver.