Authors:
Matthias Bahr
;
Sebastian Reicherts
;
Philipp Maximilian Sieberg
;
Luca Morss
and
Dieter Schramm
Affiliation:
Chair of Mechatronics, University of Duisburg-Essen, Lotharstraße 1, 47057 Duisburg and Germany
Keyword(s):
Artificial Neural Network, Active Roll Control, Neural Controller, Reinforcement Learning, Actor-critic, Driving Maneuvers, Vehicle Dynamics, Machine Learning.
Related
Ontology
Subjects/Areas/Topics:
Application Domains
;
Automotive Industry
;
Formal Methods
;
Mobile Software and Services
;
Neural Nets and Fuzzy Systems
;
Simulation and Modeling
;
Telecommunications
;
Wireless Information Networks and Systems
Abstract:
This paper deals with the application of machine learning for active roll control of motor vehicles. For this purpose, a special learning method based on reinforcement learning with an actor-critic model is used. It discusses and elaborates the basic design of the neural controller and its optimization for a fast and stable training. The methods mentioned are then validated. Both the training and the validation data are simulatively generated with the software environments MATLAB / Simulink and IPG CarMaker, while the architecture and training of the artificial neural network used is realized with the framework TensorFlow.