ROBOTIC EXPERIMENTS IN

SYNTHETIC PSYCHOLOGY

KREST Institute Summer Research

June 27, 2008

Project Leader: Dr. Pedro Diaz-Gomez

Research Group

Benoit Tufeu Cora James Judy Kula Terri Godman

First: What is a Robot?

It is an autonomous system. It must exist in the physical world. It must be able to sense its environment. It can act based on the sensor. It must achieve a goal.

Pg. 2 The Robotics Primer, Maja J. Matarić

Is this a robot?

Problem:

To design a Braitenberg style robot simulation that would accurately touch a light and then seek the next.

A Braitenberg Robot

Trial and Error

Four Robots

TerriBenoit

JudyCora

Four Environments

Environment 1Environment 2Environment 3Environment 4

Four by Four sets of data

Statistics on the Four Robots

Dr. Diaz’s Analysis of Variables

Parameters of the selected robot: Benoit’s Robot B

+10

-2

+ (variable)

Hypothesis

A robot simulation that includes a central sensor with a small positive bias will be more accurate in acquiring and hitting lights then one with a central sensor with a high positive bias.

Note: Bias is a value that controls the speed of the engine based on the intensity of the light.

Experiments

Run Robot B through training Environment 4 with 10 trials, varying the bias on the central sensor from 0.0 to 1.0 (in 0.1 intervals).

Run Robot B through a new 14 light test environment with 10 trials, varying the bias in the same manner.

New 14 Light Test Environment

Varying the Bias in Robot B – Training Environment 4

Varying the Bias in Robot B – Test Environment 5

Initial Evaluations

After reviewing the data, we found that there was little difference in the accuracy at low biases of 0 to 1.

Further tests were run at biases of 5, 10, 15, and 20 in order to have a wider range of data.

Distribution of Training Environment 4

Distribution of Test Environment 5

Results of ANOVA test

All data from biases 0 to 20 on the central sensor was compared to the number of lights hit.

This showed statistically that bias has an effect on accuracy.

Results of ANOVA test for training environment 4

Results of ANOVA test for Test Environment 5

Analysis of ANOVA tests

The very small P-values statistically show that the bias value of the central sensor has a significant effect on the accuracy of robot performance.

To support our hypothesis that a low bias is more accurate, a KS (Kolmogorov-Smirnov) test was run to compare biases under one to biases greater than one.

General Statistics on the Additional Experiments.

Training Environment 4Bias 0.0 to 0.9

Training Environment 4Bias 1.0 to 20.0

TestEnvironment 5Bias 0.0 to 0.9

TestEnvironment 5Bias 1.0 to 20.0

Mean 7.185 3.640 11.67 10.08

Standard deviation

1.77 2.45 .958 2.17

P value 0.000 0.000 0.000 0.000

1st Quartile 8.00 2.00 12.0 8.00

2nd Quartile 8.00 3.00 12.0 11.00

3rd Quartile 8.00 4.75 12.0 12.0

Maximum 8.00 8.00 12.0 13.0

Minimum 2.00 0.00 12.0 6.00

K-S test results for Training Environment 4

Higher bias

Lower bias

K-S test results for Test Environment 5

Higher bias

Lower bias

Conclusions

Based on the results of the KS tests for both environments, a bias below 1 on the central sensor is more accurate then a bias above 1. This is consistent with our hypothesis.

Future Research Could Include

a larger trial population in order to be statistically more significant.

more test environments. more variation of biases. investigations on biases of the other

sensors. changes in the positions of sensors. programming that allows for evaluation

of environmental conditions

BUILDING ROBOTSApplication

Building BYO-bots with Dr. Miller

The Handy Board: the Brain and Power of the Robot

Our First Working Robot

Our First Robot in Action

A New Prototype

THE ENDThank You,

Gracias, Merci,

Dr. Diaz