Eventos Anais de eventos
ENEBI 2022
VII Encontro Nacional de Engenharia Biomecânica
Bipedal walking using deep reinforcement learning and proximal policy optimization
Submission Author:
Luca Borgonovi , SP
Co-Authors:
Jhon Paul Feliciano Charaja Casas, Luca Borgonovi, Adriano Siqueira
Presenter: Jhon Paul Feliciano Charaja Casas
doi://10.26678/ABCM.ENEBI2022.EEB22-0007
Abstract
Mathematical models have been developed to describe the dynamic of bipedal walking activity. Some of these models are formulated with a simple physical interpretation of the system. Others consider many nonlinear relationships and kinematic constraints to guarantee high precision and reliability. However, in some cases, using the more detailed models can be a challenging activity due to the number of parameters to be tuned. Likewise, it also does not guarantee the stability of a dynamic system. For this reason, this work focuses on performing bipedal walking using a machine learning approach that does not require a complex mathematical model or control formulation. On the one hand, the MuJoCo dynamic simulator will be used to simulate the dynamics of a two-legged robot. On the other hand, deep reinforced learning with the proximal policy optimization algorithm will be used for the robot to learn to walk.
Keywords
Two-legged Robot, Reinforcement Learning, proximal policy optimization
DOWNLOAD PDF VIEW PRESENTATION
