Sistema de controle para um Andador Robótico inteligente
This work is about creating a kinematic control to the smart robotic walker, the system must be able to stabilize the robot in a desired position in the environment without obstacles. This work aims to study and evaluate machine learning algorithms on creation of the kinematic model. First, a li...
Na minha lista:
| Autor principal: | |
|---|---|
| Outros Autores: | |
| Formato: | bachelorThesis |
| Idioma: | pt_BR |
| Publicado em: |
Universidade Federal do Rio Grande do Norte
|
| Assuntos: | |
| Endereço do item: | https://repositorio.ufrn.br/handle/123456789/50464 |
| Tags: |
Adicionar Tag
Sem tags, seja o primeiro a adicionar uma tag!
|
| Resumo: | This work is about creating a kinematic control to the smart robotic walker,
the system must be able to stabilize the robot in a desired position
in the environment without obstacles. This work aims to study and evaluate
machine learning algorithms
on creation of the kinematic model. First, a literature review about
machine learning in context of kinematic control was made. Second, the
simulation of the robot and the environment was builded with The robotics
simulator CoppeliaSim. Third, the control system was created.
Two kinematic models were created, both models obtained using supervised learning,
the parameters of the neural networks are the kinematics parameters or
inspired by the kinematic constraints. In order to training these networks
an algorithm to collect data from simulation
and another algorithm to preprocessing this data was made. The controller
is a feedback control with two proportional-integral-derivate (PID) controllers
with the parameters obtained empirically. The models were evaluated using the
analytical model and a test dataset. In order to evaluate the kinematic
control system, was observed the robot in your position stabilization
task where during the task the distance and angle between the robot and
the goal were measured, this task was executed four times with four different
desired positions.
The results of mean square error of the models in the test dataset and
the graphs of distance and angle shows that the models are equivalent.
The parameters of neural network
that have the kinematic equations, show that the wheel radius and the
distance between the wheels are proximately to the analytical solution, but
the machine learning model found another combination of wheel angles.
This work concluded in the literature review that classic approach
with PIDs is better than machine learning models, because of memory
and processing usage. This work concluded that creating a kinematic
control system though the analytical solution is more simpler than
with machine learning, and creating a neural network with the parameters
inspired by the kinematic constraints produces a model equivalent to
the analytical model. |
|---|