Yes and No

[DirtyRoboto]

I have been having a play with Robobasic and had some really good results. What I have been doing is simulating AI in that the RN will learn from my input and modify its actions.

My first attempt used 10 routines from the overall template. I RND the selection of an routine and the RN performs it and then beeps. I input a yes or no via the remote and the RN logs this, plays an up tone or down tone and adds or subtracts from the value for that routine.

After 50 of these interactions the RN goes into autonomous mode and will select actions based on the previous input given with various time delays between routines.
At various points the RN beeps to request remote override (I can modify the original bias table by running moves via remote) so that I can interrupt its exploration.

As of now i'm starting with 50 routines and 130 interactions. Most of the routines are basic (and minimal) actions. For example, one move may be "left arm up 25 degrees" another may be " bend forward 10 degrees". I do throw in a few complex routines for measure.

Next month I will bring in combat specific routines to the program and see how he can refine user defined guidance to hone his method of battle.

I hope this opens a new avenue for you.

Marcus.

[PaulP]

That sounds a bit like pattern learning in OCR. When the OCR is run you have the opportunity of telling the software which characters it got right and wrong. The wrong ones you tell what it should have been and the next time it runs it will compare any uncertains against its past corrections.

As time progresses the OCR gets more and more accurate.

[DirtyRoboto]

Try to think of that sort of function within RoboBasic and you will start to understand the extent of the coding I have done.
I am just using a bias table, although it is the 7th son of the bastard 7th son of any table you may have seen (this is RoboBasic y'know), It functions as I expected it too function. I have ended up with various incarnations of Takeshi. I wish I could figure a way of saving particular personalities and reloading them. I have yet to figure that.

[Pev]

Perhaps a method of saving the personality data would be to write the bias table data out over the serial comms on the processor board (ETX/ERX). You could then reload it the same way. A simple vb.net application could do it and write the personality data to a file on a PC.

Just a thought

Pev

[DirtyRoboto]

But a good thought!

I am retarded at the higher level aspects of the I/O of RN. Like all the protocols and that. I have quite alot of catch up in this area.

[PaulP]

If you are going to use VB then VB6 is very friendly...

VB.NET on the other hard has some quirks in the serialport implementation. It expects to use Unicode so everything is 2 bytes long..

If thats the route you go then let me know, I have a codepage fix that gets round it...

[Pev]

Hi PaulP

What version of VB.Net are you using? I have been using VS 2005 and the serialport object. I haven't seen any issue with sending single bytes or receiving them. Would be really interested in your issues and the codepage fix

Thanks

Pev

[Humanoido]

A more direct approach is to use data statements and just read assigned values for behavioral routines. Since RoboBasic does not have the Data and Read statements, the same could be accomplished with Peek and Poke by reading and writing data to and from the controller RAM. The template could load up the values at the beginning for AI assignments, then access at any time during the program. This is the same approach used in the 1970s with Integer Basic. Many of those routines and techniques are quite useful with RoboBasic, and various tutorials still exist on the web.

humanoido