finished version #1 of the software block diagram. also broke the E-MAXX’s front-left suspension arms, front-left shock shafts, and short front-left shock caps by driving it into a curb.

Here’s the software block diagram:

The fuzzy logic parts are the neural networks.

set up hotmail account at GCWinners@hotmail.com for Chris and I to use to communicate about the Grand Challenge neural networks.

added a graph on the neural network software that shows the percent of the pictures that it gets correct during the learning process

went to my cousin’s house for Easter. chris and I are going to be on a team. he’s building the vehicle (a Volkswagen) and I’m doing the computer stuff. the E-MAXX might be used as a test vehicle.

I had decided not to use the Ni-MH batteries that I had bought because they would be too hard to build into battery packs, so I ordered two Ni-MH 8.4 V 3000 mAh battery packs and two chargers for them my dad said that I could give him the AA batteries and he’d buy the packs if I’d pay the difference, which is $20. Shipping and handling was $5.42.

bought batteries for E-MAXX. also made it so that there is an input layer of neurons that can have a different number of neurons than the hidden layers. Bob agreed to give me $500 if I filled out a quad chart that told my basic approach

I bought fourteen AA Ni-MH batteries for the E-MAXX. Each of them has 1.2 V and is rated at 1700 mAh. They were $2.33 each. I only have one battery charger that can charge Ni-MH batteries, and it can only charge four at a time. It takes eight hours. That’s twenty-four hours to charge twelve of them. (If somebody is awake to switch them whenever they’re done charging)

changed learning process in ANN software a little bit and made it so that the user can limit The number of inputs that each neuron has

This is how the learning process works: First, a random synapse, or connection between one neuron and another neuron (or an input), is selected. Next, a number called the TestChange is added to it. I have been using the number 0.1. After the network gets a new output, the difference between the current error level and what the error level was before the synapse was changed is found. It then subtracts the TestChange from the synapse, undoing everything that it just did. If the error level had gone down, another number is added to that synapse. If the error level had gone up, another number is subtracted from that synapse. The amount that it adds or subtracts used to be found by multiplying the difference in the error level by a number called the LearnAmount, which was 200. This made it so that the more the synapse effected the output of the network, the larger the change was made to it. I did this because it was suggested to me. Later, I decided that it would be better if the amount that it added or subtracted was the same no matter how much of a difference the test change made. I changed it so that it equaled the sign of the difference times LearnAmount, which I changed to 0.5. Once the network is done with that synapse it checks if its output is close enough to the correct output. If it isn’t it repeats the process, using a new randomly selected synapse. The network learning process works better with the new change.

got E-MAXX in mail

I got the E-MAXX in the mail today. I put the Logitech QuickCam on it and took some pictures of it.

tried to make program faster, but failed (mostly). Taught network to recognize people

There are two things that are taking a long time to do. The first one is pixelizing the image and converting it to black and white. I can’t find any other solution than the one I’m using, which involves adding up values of pixels and dividing by the total number to get averages. I tried using PaintPicture to paint it into a very small spot and then paint it onto a big spot, but that doesn’t work the same. I took out a DoEvents that I had in there and it goes two seconds faster on a 50x50 image. The other thing that’s taking forever is giving the information from the picture to the first layer of neurons.

Later, I put the Logitech QuickCam on top of the computer monitor and took a lot of snapshots, some with me sitting in front of the computer and some without. I made an information packet, or whatever you want to call it, taught a 30 x 3 network, and saved it as Person-NoPerson.ANN. After that I made a new Visual Basic program and made it load a neural network with the Person-NoPerson.ANN file. It had a label that would say either “Person” or “No Person”. It was cool!

changes to ANN software’s user interface

I made the program a little neater and easier to use. (hopefully)