On the benefit of smart tyre technology on vehicle state estimation

A Two-Stage Bayesian optimisation for Automatic Tuning of an Unscented Kalman Filter for Vehicle Sideslip Angle Estimation

“…Despite the extensive use of UKF for vehicle sideslip angle estimation, the tuning of the process and observation noise parameters is rarely investigated in the literature. The observation noise parameters are usually tuned through statistical analysis, obtained after sensor calibration tests [5]. The only exception is when the observation noise parameters are tuned together with the process noise parameters [9].…”

“…The idea is to perform a grid search in the parameter space and reach the best tuning iteratively. Despite the oversimplicity of the concept, it is still very often used, especially for vehicle sideslip angle estimation [10], because the performance of Kalman filters is stable for a range of process noise parameters settings [5]. The main drawback is the inefficiency of the approach.…”

“…Metaheuristics are procedures that can provide an acceptable solution to an optimisation problem with incomplete information about the cost function. Examples of metaheuristic algorithms are: the already cited downhill simplex algorithm [12], GA [13] or the Multi-Objective GA [5], SA [9] and FFO [4]. A comparison of different metaheuristic algorithms for Kalman filter tuning [4] shows that GA and FFO found the best cost function optimum.…”

“…Furthermore, only stochastic optimisation algorithms have been compared that can be inefficient if the UKF takes a long simulation time. For vehicle sideslip angle estimation, large-scale training sets representing vehicle behaviour [5] make the cost function evaluation very time-consuming. In different engineering applications, several solutions have been proposed, e.g.…”

“…The methodology framework is represented in Figure 1, a more detailed explanation will be presented in Section IV. The observation noise parameters are tuned by performing a statistical analysis of the vehicle sensor measurements [5]. The process noise parameters' tuning is based on optimising a cost function defined as the sum of the Normalized Root Mean Squared Errors (NRMSE) of the vehicle states and measurements for 8 manoeuvres.…”