Actor-Critic Reinforcement Learning to Traction Control of an Electrical Vehicle

Abstract

Direitos para esta edição cedidos à Atena Editora pelos autores. Todo o conteúdo deste livro está licenciado sob uma Licença de Atribuição Creative Commons. Atribuição 4.0 Internacional (CC BY 4.0). O conteúdo dos artigos e seus dados em sua forma, correção e confiabilidade são de responsabilidade exclusiva dos autores, inclusive não representam necessariamente a posição oficial da Atena Editora. Permitido o download da obra e o compartilhamento desde que sejam atribuídos créditos aos autores, mas sem a possibilidade de alterá-la de nenhuma forma ou utilizá-la para fins comerciais. A Atena Editora não se responsabiliza por eventuais mudanças ocorridas nos endereços convencionais ou eletrônicos citados nesta obra. Todos os manuscritos foram previamente submetidos à avaliação cega pelos pares, membros do Conselho Editorial desta Editora, tendo sido aprovados para a publicação.

Actor-Critic Traction Control Based on Reinforcement Learning with Open-Loop Training

“…For the proposed controller, which avoids the inadequate sleep of the wheels, the predicted future rewards do not have a significant influence on the current action. Furthermore, the evaluations of the discount factor researched by Funk Drechsler et al [1] indicate better behavior of myopic training processes for this specific implementation.…”

“…The implemented critic is composed of a network with two hidden layers and twenty nodes in each layer, while the actor has the same number of hidden layers and just twelve hidden nodes in each layer. The network architecture including input and output data is the same applied by Funk Drechsler et al [1]. The critic and actor networks use the Tangent-Sigmoid activation function in the hidden layers and linear function in the output layer.…”

“…An actor-critic controller also presented adequate behavior during the traction control of an electric vehicle in a simulated environment. In this research, the actor-critic algorithm was able to avoid the inadequate slip-on different floors with a single control network [1]. The actor-critic RL implementation also overcame the challenges of traction control development as the nonlinearities of tires and the simplicity necessary to real-time applications [1].…”

“…A preliminary study of the actor-critic reinforcement learning algorithm to control the traction of an electrical vehicle had been presented at the conference XXII ENMC-Encontro Nacional de Modelagem Computacional at Juiz de Fora in Minas Gerais, Brazil [1]. In the previous research, a closed-loop approach had been used to train the neural network, while in the current paper, an open-loop method is proposed to permit a future dataset collection and the controller implementation on the real vehicle.…”

“…The fine-tuning of the controller is required due to the simplification and limited representation of the plant models in the simulation, which cause different behaviors of the controller when controlling the virtual model or the real vehicle. However, as appointed by Funk Drechsler et al [1], even the finetuning can be challenging due to iterative effort, limited processing power inside the vehicle, and the possible generation of unsafe maneuvers as cited before.…”