Analysis of Artificial Neural Network used to control a Crawling Robot

In the previous video I showed how an ANN evolved to produce forward and backward movements in a crawling robot. In this video I will show how the sensor inputs propagate through the net and come out of the outputs to control the robot's servos. The sensor inputs propagate through the net in patterns that repeat over and over and produce cyclical behavior. It is this cyclical behavior that is the key to the locomotion.

Category:

Science & Technology

Tags:

• artificial
• neural
• network
• crawling
• robot
• senior
• design
• engineering
• swarthmore
• college
• locomotion

[HyppoStudios]

What are the inputs you use?

[shaggyman87]

Starting from the left: Inputs 1 through 3 receive values relative to the position (angle) of the servos which control the motion of the arm and rotation of the arm in relation to the base. Servo 3 wasn't really used. Input 4 was the force sensor at the end of the arm. Input 5 was the odometer at the beginning of the chassis. And input six was the contact sensor at the bottom of the chassis. The ANN evolved just using the changing of the contact sensor to coordinate the articulation of the arm.

[amrosik]

what is this programm

[shaggyman87]

This was down in using MATLAB's gui and plot functions. I wrote a script which plotted the location of the robot and the neural net as lines and circles, and then in the next time step I just moved the lines/dots/circles. So half the code worked with robot geometry and the learning algorithm and the other have just plotted and animated it.

[dt28469]

This is a good example of how the human brain became inteligent through millions of years but i still dont understand how an individual neural network becomes trained through the back propogation of error.

ie. taking your first steps or learning to play guitar.

[plmqas]

was that Neat neural network ?

[amrosik]

ok---

[zelsenussi]

WODERFUL JOB..