|  | |||||||||
MINNI: Micromouse Incorporating Neural Network Intelligence
Jondarr Gibb
Computing Department
Macquarie University
NSW 2109
Len Hamey
Computing Department
Macquarie University
NSW 2109
Abstract
MINNI is a system whereby a back propagation neural network is used to control the steering of a micromouse (small robot) in following a straight path. The neural network must be minimised to run on the hardware/software platform available on the Macquarie Micromouse. The development of the network follows intensive trials of algorithm and architecture variations in MATLAB. The implementation is programmed in ADA.
Keywords Neural networks, robotics.
1 Introduction
The IEEE International Micromouse Competition runs events pitting robots from many universities at state, national, and international levels to promote the application of real time control in university computing and engineering departments. The Rodentronics team from Macquarie has known some success in the past.
The micromouse itself has undergone many changes over the years. Starting with the simple kit used as the basis of most competition entrants, controlled by C code, the group has designed and developed a more sophisticated model which now relies on an ADA run-time support system (Alsys 286 DOS Ada Vers 4.2) and an Arcom Control Systems SCIM88 80C188 CPU board. As far as is known, MADAM (the Macquarie ADA Mouse, figure 1) is the first of its kind, perhaps in the world.
This development has allowed the micromouse to be used in the teaching of an Advanced ADA course, wherein students program the robot to follow a simple path in the competition maze, as well as being used as the basis of some research projects. Normally, some form of feedback control technique is used to control the micromouse. Fuzzy logic has been applied to the steering control for cornering [11], but neural networks had not come into the picture.
Proceedings of the 20th Australasian Computer Science Conference, Sydney, Australia, February 5{7 1997.
Figure 1: The Macquarie ADA Mouse.
Initially, the purpose of the research was to investigate the application of neural networks to control the micromouse when following a straight path. This is a non-trivial problem, due to factors such as variable slippage across the two driving wheels, differing motor wear (the wheels have independent control), initial orientation or position of the micromouse, and the varying sensitivity of sensors.
This project could be considered a scaled-down version of the ALVINN project [12, 8]. In that system, a large amount of computing power was used to steer a truck using video input. The training of that back propagation neural network required some filtering of the video signal, and manipulation of data patterns to extend the training set to gain generalisation (through modifying the input video data to indicate differently-shaped roads ahead). With sufficient computational power, the network was able to steer at quite high speeds and over great distances, reacting quickly to the environment. It holds the distance record for an autonomous vehicle on a public road.
This research will be using simple sensors instead of video input, and relying only on the immediate position of the robot relative to the path, rather than the shape of the road ahead. A back propagation neural network is used for simplicity of implementation, requiring very little