COSI for Credit Fall 2009 (MP151)

There was no real defined method to my madness of changing, testing and debugging my social robotics program, but I did take screenshots along the way. The original purpose was to create visualizations of different variables within the socialization process. The first of which was going to be the conformity levels, or the degree of how much the child robot wishes to mimic the parent robot.

The problem that was discovered is that from iterations 0-1000, there is a pattern where the parent conformity level moves from quadrant 4 to other quadrants. However, when the program was expanded to go from 0-infinity iterations, the pattern was essentially lost. This is important since the parent conformity level determines how much the child robot decides to mimic the parent robot.

Below is a visual walk through time of this process, with annotations and descriptions!

---

This is a screenshot of the updated control GUI of the social robotics program. There are no longer many sliders and buttons, which makes the program run faster and longer!

1. Iterations can now go from 0-infinity instead of 0-1000
2. Buttons include stepping 1 iteration, automatically iterating, time delay (for automatic iterations), and log data

This is a screenshot of when the visualization program first worked! Woohoo!
Bubble size is the total conformity level, X position of the bubble is parent conformity level, Y position of the bubble is action conformity level

1. Not many iterations
2. The transperency of the bubbles will have to be lowered in order to see more changes over time

This is a screenshot of when the visualization program was modified a tad to see the data over many iterations

1. Bubbles are transperent and red, so that many layers can be seen
2. Iterating for a numerous amount allows us to see where most of the data lands
3. Labels are the number of averages taken, where averages are taken every 5 iterations… this allows us to see where most of the averages are

Showing us more iterations

1. Many more iterations
2. Distribution of bubbles seems to be always in the same area
3. Distribution of the averages also seems to be in the same area

Modified the transparency of the visualization program

1. Distribution still is in the same place, all the time! (This is not good)
2. Many iterations…

I wanted to add more colour to the visualization, just checking to see what it would look like

1. 499137 iterations
2. Bubbles are coloured, still located in the same quandrant
3. Average labels still in the same place

I modified the algorithm that changes the X position (parent conformity level) of the bubble in attempt to limit where the bubbles venture. In essence, it did that, but not exactly what we were looking for…

1. We can at least see an average that made it out of quandrant 4, but the rest seem to be around the same area.
2. Many iterations – 1384197
3. X position of the bubbles is more clumped together

The visualization program was modified to remove the random colours (sort of distracting after a while), but also draw lines from the bubbles

1. Lines are drawn from bubble to bubble
2. Not many iterations into the program…
3. …but the x positions seem to be wandering! (this is good)

Removed the line layers and averages just to see the general location…

1. X values are all in quandrant 4 (this is not good)
2. Not many iterations, we’ll leave it to iterate for a bit

Instead of labels being the averages, red dots are. This way we can clearly see the distribution…

1. X values of the bubbles do not appear to be widely dispersed
2. Many iterations – 394039
3. Red dots show that X values are all in the 4th quandrant

An age component was added– whereby the older the robot gets, the more converged the x-values will be (in theory)

1. X values seem to converge (too much)
2. Evaluated many iterations – 182028

As a final hope, I did a time trial of when the robot would go through its different age categories

1. The X values are scattered for a little while at the beginning — but not scattered enough
2. Time trial would last 2 hours, 5 minutes.
3. Just a few iterations in! (496)

1. X values are all around the same quandrant, that’s not what we were aiming for!
2. 2 hours 5 minutes of time is up — robot went through all its age categories
3. 84752 iterations

After this time trial was when I started to wonder if there would be any other methods to overcome this. After looking into it, neural nets will be the best answer! The only problem is that there are several inputs that make up the parent conformity level, and the way that I would neural net everything would entail that there will be about 10 hidden layers (yikes!!).

Tags: bubble, child robot, conformity, debugging, experiment, iterations, madness, neural nets, quadrant, screenshots, social robot, Social Robotics, socialization process