Vision Based Robot Control Using Machine Learning
No Thumbnail Available
Date
2019-06
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
A new simpler vision based robot control system is proposed characterized with position specific
artificial neural network (ANN) and end-effecter integrated camera system. Position specific ANN
avoids the difficulty of covering the whole joint space with changing parameters using one set of
ANN, and end-effecter integrated camera system makes the image of an object consistent when
the end-effecter approaches the object. The object coordinate can be directly used as feedback.
Most vision-based robot positioning techniques rely on analytical formulations of the relationship
between the robot pose and the projected image coordinates of several geometric features of the
observed scene. Feature matching algorithms, camera calibration, models of the camera geometry
and object feature relationships are also necessary for pose determination. These steps are often
computationally intensive and error-prone, and the complexity of the resulting formulations often
limit the number of controllable degrees of freedom.
This thesis presents controlling mechanism of a parallel robot based on deep neural learning and
position based visual servoing that overcomes many of these limitations. ROS/Gazebo simulator
is used to model delta 3 parallel robot. From the model training data set is collected and a multilayer
feed
forward
deep
neural
network
is
used
to
learn
the
complex
implicit
relationship
between
the
pose
displacements
of
the
delta
3
robot
and
joint
angles.
Three
networks
with
three
hidden
layers
but different number of neurons per hidden layer were trained and their performance is
evaluated. Based on the simulation result it is shown that a network with higher number of neurons
per hidden layer shows better performance.
The trained network may then be used to move the robot from arbitrary initial positions to a desired
pose with respect to the observed scene with MSE less than 0.05. Simulation result shows that the
system works smoothly, and converges in limited steps. The algorithm simplifies the model of
vision based robot manipulator control system, and improves the control accuracy and response
time.
Description
Keywords
Artificial Intelligence, Machine Learning, Artificial Neural Networks, Deep Learning, Deep Neural Networks, Feed-forward neural Network, Rectified Linear Units, ROS/Gazebo, Supervised Learning, Visual Servoing