Controlling a Robot with a Neural Network

1
Controlling a Robot with a Neural Network

• CS/PY 231 Lab Presentation 9
• March 30, 2005
• Mount Union College

2
The Problem

• To illustrate pattern recognition by a Neural
  Network, well train a network to recognize when
  a robot is on a predefined path, and to move it
  back on the path if it strays to one side
• Input signals will come from light sensors
• Output signals will control the wheels of the
  robot

3
LEGO MindStorms Robots

• Constructed from LEGO pieces
• Yellow Brick contains a processor
• Programs are loaded into the Brick via an
  infrared transmitter connected to a conventional
  computer
• write a traditional program
• upload to the robot
• robot runs the program when RUN button is pushed

4
Programming Language

• NQC
• Not Quite C
• Similar to standard programming languages
  (collection of instructions)
• How to interface with a TLearn network?
• Write a program that reads Tlearn-formatted files
• Ive done this for you. (youre welcome)

5
Possible Inputs for this problem

• Input Signals come from Light Sensors
• 1 light, 0 dark
• Case 1 0 0
• Both sensors on the road
• Case 2 0 1
• Right sensor is off the road
• Case 3 1 0
• Left sensor is off the road
• Case 4 1 1
• Both sensors off the road
• distributed
• 4
• 0.5 0.6 0.4
• Choose Network/Testing Options, then
• choose a specific weight file from the original
  project

6
Corresponding outputs for this problem

• Two output signals
• One for each wheel (1 move, 0 stop)
• Case 1
• Keep driving straight (both wheels move at same
  speed)
• Case 2
• What should we do (were off the road to the
  right)?

7
Corresponding outputs for this problem

• Case 3
• What should we do (were off the road to the
  left)?
• Case 4
• What should we do (wheres the road)?

8
Network Configuration

• 2 input nodes, 2 output nodes
• Rest is up to you to define!
• Break into 2 person teams, choose a team name,
  define the .CF file for your network, then go to
  lab to train your network
• Next time, Robot Testing!!!

9
Controlling a Robot with a Neural Network

• CS/PY 231 Lab Presentation 9
• March 30, 2005
• Mount Union College