Road angle estimation for a vehicle-trailer with machine learning and system model-based approaches

Abstract

This paper proposes two different approaches for estimating grade and bank angles for arbitrary vehicle-trailer configurations independent from road friction conditions: model-based and Machine Learning (ML) approaches. The model-based method employs unknown input observers on a vehicle-trailer roll/pitch dynamic model with fault thresholds. In the proposed ML approach, a Recurrent Neural Network (RNN) with long-short term memory gates is designed to estimate the road angles. The inputs of the RNN have been selected based on the vehicle-trailer roll and pitch dynamic models, and are normalised by the vehicle wheel-base, mass, and centre of gravity height so that the network is modularly applicable to different trailer types. The simulation and experimental test results justify the performance of the proposed road-bank and grade-angle estimation scheme in various cases and demonstrate that both bank and grade angles can be estimated with high accuracy.