Autonomous Vehicles: Vehicle Parameter Estimation Using Variational Bayes and Kinematics

Abstract

On-board sensory systems in autonomous vehicles make it possible to acquire information about the vehicle itself and about its relevant surroundings. With this information the vehicle actuators are able to follow the corresponding control commands and behave accordingly. Localization is thus a critical feature in autonomous driving to define trajectories to follow and enable maneuvers. Localization approaches using sensor data are mainly based on Bayes filters. Whitebox models that are used to this end use kinematics and vehicle parameters, such as wheel radii, to interfere the vehicle’s movement. As a consequence, faulty vehicle parameters lead to poor localization results. On the other hand, blackbox models use motion data to model vehicle behavior without relying on vehicle parameters. Due to their high non-linearity, blackbox approaches outperform whitebox models but faulty behaviour such as overfitting is hardly identifiable without intensive experiments. In this paper, we extend blackbox models using kinematics, by inferring vehicle parameters and then transforming blackbox models into whitebox models. The probabilistic perspective of vehicle movement is extended using random variables representing vehicle parameters. We validated our approach, acquiring and analyzing simulated noisy movement data from mobile robots and vehicles. Results show that it is possible to estimate vehicle parameters with few kinematic assumptions.

Chair for Intelligent Transportation Systems–Sustainable Transport Logistics 4.0

“…Furthermore, the need for explainable machine learning solutions for safety-critical systems motivates the research into algorithms for motion models and image processing. To identify unknown behaviors and causes that lead to particular predictions by artificial neuronal networks, approaches relying on Gaussian processes are investigated [16]. Nonparametric and nonlinear methods for image processing using Gaussian process latent variable models for street sign feature extraction are also considered [17], [18].…”

Feasibility of Formulating Ecosystem Biogeochemical Models From Established Physical Rules

“…The successes of computer‐aided design software in industry and numerical models in weather forecasting are founded on mathematical models formulated according to physical rules (see Table 1 for examples of these rules). The performance of these models can be continuously improved by including new essential processes (Zhou et al., 2022), adopting more robust and effective numerical solution strategies (Candel et al., 1999; Lin & Rood, 1996; Liu et al., 2019), utilizing better constrained parameters (Kotsuki et al., 2018; Wöber et al., 2020), implementing more accurate initial and boundary conditions (Saredi et al., 2021; Xiao et al., 2007), and increasing spatiotemporal resolution with more computational resources (Caldwell et al., 2021). Such a trajectory allows for the realization of “the unreasonable effectiveness of mathematics” wherein simple equations can accurately describe complex real‐world phenomena (Wigner, 1960).…”

Autonomous Vehicle Calibration via Linear Optimization

“…In addition, the mobile robotic community proposes methodologies for motion estimation based on probability theory known as probabilistic robotics [9]. Based on those models, machine learning is used to obtain motion from sensor data [10], [11]. Nevertheless, due to limited explainablitiy of recent machine learning models [12]- [14] classic models relying on deterministic kinematic properties are still used in IVs [15].…”