@InProceedings{MarquesChag:2020:ReLeAp,
               author = "Marques, Wilson de S{\'a} and Chagas, Ronan Arraes Jardim",
          affiliation = "{Instituto Nacional de Pesquisas Espaciais (INPE)} and {Instituto 
                         Nacional de Pesquisas Espaciais (INPE)}",
                title = "Reinforcement learning applied to the control of the pitch-axis of 
                         a satellite",
            booktitle = "Anais... do 11º Workshop em Engenharia e Tecnologia Espaciais",
                 year = "2020",
               editor = "Rodrigues, Aline Castilho and Barbosa, Anderson Luis and Teixeira, 
                         Andr{\'e} Ferreira and Batista, Carlos Leandro Gomes and 
                         Oliveira, Caroline Mateus de and Cuellar, Dairo Antonio",
         organization = "Workshop em Engenharia e Tecnologia Espaciais, 11. (WETE)",
            publisher = "Instituto Nacional de Pesquisas Espaciais (INPE)",
              address = "S{\~a}o Jos{\'e} dos Campos",
             keywords = "Satellite, Attitude control, Reinforcement learning.",
             abstract = "In this paper, we assess the control of the pitch-axis of a 
                         satellite with the use of reinforcement learning techniques. The 
                         main goal of this work is to show the feasibility of this approach 
                         and to compare its performance with a traditional method of 
                         control design from the satellite literature. The state-of-the-art 
                         Twin Delayed Deep Deterministic Policy Gradient (TD3) 
                         reinforcement learning algorithm was used. Results show that the 
                         attained optimal policy can have similar performance to a 
                         classical PD control law, and it is able to adapt its output 
                         accordingly in order to control a different inertia that the one 
                         it has been trained with. In fact, the same policy trained to 
                         control a nanosatellite inertia was capable of controlling the 
                         inertia of a medium-sized satellite, assuming different initial 
                         conditions. While the PD law was also capable of controlling a 
                         different inertia from the one it has been originally tuned for, 
                         its performance dropped considerably, and it presented a 
                         high-frequency control signal, very difficult to be implemented by 
                         a typical satellite actuator.",
  conference-location = "S{\~a}o Jos{\'e} dos Campos",
      conference-year = "18 -20 ago. 2020",
                 issn = "2177-3114",
             language = "pt",
         organisation = "Instituto Nacional de Pesquisas Espaciais (INPE)",
                  ibi = "8JMKD3MGPDW34R/433LLG2",
                  url = "http://urlib.net/ibi/8JMKD3MGPDW34R/433LLG2",
           targetfile = "35 - [ARTIGO][INPE] Wilson de Sa Marques.pdf",
                 type = "CMC",
        urlaccessdate = "24 abr. 2024"
}