Expert Guided Autonomous Mobile Robot Learning

Expert Guided Autonomous Mobile Robot Learning

Gintautas Narvydas (Kaunas University of Technology, Lithuania), Vidas Raudonis (Kaunas University of Technology, Lithuania) and Rimvydas Simutis (Kaunas University of Technology, Lithuania)
DOI: 10.4018/978-1-61692-811-7.ch011
OnDemand PDF Download:
$37.50

Abstract

In the control of autonomous mobile robots there exist two types of control: global control and local control. The requirement to solve global and local tasks arises respectively. This chapter concentrates on local tasks and shows that robots can learn to cope with some local tasks within minutes. The main idea of the chapter is to show that, while creating intelligent control systems for autonomous mobile robots, the beginning is most important as we have to transfer as much as possible human knowledge and human expert-operator skills into the intelligent control system. Successful transfer ensures fast and good results. One of the most advanced techniques in robotics is an autonomous mobile robot on-line learning from the experts’ demonstrations. Further, the latter technique is briefly described in this chapter. As an example of local task the wall following is taken. The main goal of our experiment is to teach the autonomous mobile robot within 10 minutes to follow the wall of the maze as fast and as precisely as it is possible. This task also can be transformed to the obstacle circuit on the left or on the right. The main part of the suggested control system is a small Feed-Forward Artificial Neural Network. In some particular cases – critical situations – “If-Then” rules undertake the control, but our goal is to minimize possibility that these rules would start controlling the robot. The aim of the experiment is to implement the proposed technique on the real robot. This technique enables to reach desirable capabilities in control much faster than they would be reached using Evolutionary or Genetic Algorithms, or trying to create the control systems by hand using “If-Then” rules or Fuzzy Logic. In order to evaluate the quality of the intelligent control system to control an autonomous mobile robot we calculate objective function values and the percentage of the robot work loops when “If-Then” rules control the robot.
Chapter Preview
Top

Background

In the control of AMR there exist two types of control: global control and local control. The requirement to solve global and local tasks arises respectively. In the figure 1 the situation when an AMR – an unmanned ground vehicle must navigate from the start position to the finish position is represented. The first task is a global path planning. To solve this global task GPS tracking devices or/and GPS auto navigator should be used. It depends where a car must drive: on road or off-road terrain. For example, in the 2005 DARPA Grand Challenge project no global path planning was required (Thrun et al., 2006). The route definition data format defined the approximate route that robots would take. What concerns a local task, Sebastian Thurn et al. write: “When a faster robot overtook a slower one, the slower robot was paused by DARPA officials, allowing the second robot to pass the first as if it were a static obstacle. This eliminated the need for robots to handle the case of dynamic passing“. These facts show that there still are a lot of unsolved problems in local and global control and encourage roboticists to continue their work on these tasks.

Figure 1.

Global and local tasks and the control of an AMR

In the figure 1 it is shown that AMR must solve several local tasks: recognize moving objects, detect the pits and obstacles on the road, see road marking and signs, understand traffic-light signals, and cope with prediction of the movement of other cars, people, and even segways or horsemen.

Complete Chapter List

Search this Book:
Reset