Author(s): Gonçalves, Fernando ; Ribeiro, Tiago ; Ribeiro, A. Fernando ; Lopes, Gil ; Flores, Paulo
Date: 2024
Persistent ID: https://hdl.handle.net/1822/91489
Origin: RepositóriUM - Universidade do Minho
Author(s): Gonçalves, Fernando ; Ribeiro, Tiago ; Ribeiro, A. Fernando ; Lopes, Gil ; Flores, Paulo
Date: 2024
Persistent ID: https://hdl.handle.net/1822/91489
Origin: RepositóriUM - Universidade do Minho
[Excerpt] 1. INTRODUCTION The Laboratory of Automation and Robotics of the University of Minho has developed CHARMIE, a human-inspired robot designed to assist humans in domestic tasks [1] (See Figure 1). One of the challenges in developing a complex mobile manipulator robot is testing and optimizing its control. Damages to the physical prototype, or possible hazardous situations, can be avoided by testing control solutions in a virtual environment. This is even more relevant when using Reinforcement Learning as a control strategy, where the neural network requires testing a set of inaccurate and inadequate solutions before the reward function allows the system to converge into more optimal strategies. This work deals with the development of a multibody simulator that represents the key kinematics and dynamics of CHARMIE’s bodies and joints.[...]