Neural Observer for Nonlinear State and Input Estimation in a Truck-Semitrailer Combination. Issue 2 (2020)

• Record Type: Journal Article
• Authors: Jahn, Tim; Ziaukas, Zygimantas; Kobler, Jan-Philipp; Wielitzka, Mark; Ortmaier, Tobias
• Is Part Of: IFAC-PapersOnLine. Volume 53:Issue 2(2020)
• Page Start: 14306
• Page End: 14311
• Publication Date: 2020
• Subjects: Feedforward Neural Network -- Nonlinear Autoregressive Exogenous Neural Network -- Extended Kalman Filter -- State -- Input Estimation -- Truck-Semitrailer Combination; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2020.12.1372
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: Driver assistance systems have become an indispensable part of today's vehicles technology. Especially in the commercial vehicle sector, the challenges in obtaining information increase with rising system complexity. Compared to trucks, trailers for commercial vehicle combinations are sparsely equipped with electronic components. This leads to difficulties in implementation of intelligent systems for the trailer as necessary information is not provided. Reasons for this can be an insufficient sensor equipment due to uneconomical costs or a missing communication channel between the two vehicle units, preventing the transmission of required truck related information to the trailer. A possible model-based method to obtain unmeasured states is the Extended Kalman Filter. However, this approach requires elaborate preliminary work steps of high complexity and a sophisticated domain knowledge. Alternatively, this paper proposes the applicability of Neural Networks for estimating the required state and input variables, namely the articulation angle and the truck's steering angle. Two different network types are used: the Feedforward Neural Network and the Nonlinear Autoregressive Exogenous Neural Network. The measured input variables for the networks, in accordance with the inputs of the Extended Kalman Filter in a previous publication, are merely trailer yaw rate and longitudinal speed. In conclusion, a comparison between the results of the Neural Networks and those of the Extended Kalman Filter is drawn.