Authors:
Pablo Quintía
;
José E. Domenech
;
Cristina Gamallo
and
Carlos V. Regueiro
Affiliation:
Facultad de Informática, Universidad de A Coruña, Spain
Keyword(s):
Reinforcement Learning, Mobile Robotics, Artificial Vision, Visual Reactive Behaviours, Motivation.
Related
Ontology
Subjects/Areas/Topics:
Image Processing
;
Informatics in Control, Automation and Robotics
;
Mobile Robots and Autonomous Systems
;
Robotics and Automation
Abstract:
This article describes the development of a methodology for the learning of visual and reactive behaviours using reinforcement learning. With the use of artificial vision the environment is perceived in 3D, and it is possible to avoid obstacles that are invisible to other sensors that are more common in mobile robotics. Reinforcement learning reduces the need for intervention in behaviour design, and simplifies its adjustment to the environment, the robot and the task. The designed methodology is intended to be general; thus, in order to change the desired behaviour, only the reinforcement and the filtering of the image need to be changed. For the definition of the reinforcement a laser sensor is used, and for the definition of the states a fixed 3x3 grid is used. The behaviours learned were wall following, object following, corridor following and platform following. Results are presented with a Pioneer 2 AT. A Gazebo 3D simulator was used for the Learning and testing phase, and a te
st of the wall following behaviour was carried out in a real environment.
(More)