RoboSavvy Forum

AI

2008-03-12T23:12:36+01:00

I'm behind the times again.

Excellent work Dan !!!

I've just read your article in December issue of Servo. Its Fascinating, although I don't understand fully yet

Any more progress news? I am especially interested in the work you are doing with the 5 axis IMU (sparkfun S-SEN-00741 ?) you mentioned.

Statistics: Posted by NovaOne — Wed Mar 12, 2008 11:12 pm

AI

2007-06-03T04:41:45+01:00

Dan, that's remarkable what you've accomplished so far! Your MOOSE operating system sounds pretty cool! Especially that you're taking suggestions for an AI section. Thanks for posting in this thread. I think to answer the question about AI and finding some simple applications of real AI, we probably have to look at some of the definitions.

This one is very interesting.
http://www.dobrev.com/AI/definition.html
But that was beyond my means.

So I found this, and quote it here.

artificial intelligence
The branch of computer science concerned with making computers behave like humans. The term was coined in 1956 by John McCarthy at the Massachusetts Institute of Technology. Artificial intelligence includes
# games playing: programming computers to play games such as chess and checkers
# expert systems : programming computers to make decisions in real-life situations (for example, some expert systems help doctors diagnose diseases based on symptoms)
# natural language : programming computers to understand natural human languages
# neural networks : Systems that simulate intelligence by attempting to reproduce the types of physical connections that occur in animal brains
# robotics : programming computers to see and hear and react to other sensory stimuli

Currently, no computers exhibit full artificial intelligence (that is, are able to simulate human behavior). The greatest advances have occurred in the field of games playing. The best computer chess programs are now capable of beating humans. In May, 1997, an IBM super-computer called Deep Blue defeated world chess champion Gary Kasparov in a chess match.

http://www.webopedia.com/TERM/A/artific ... gence.html

humanoido

Statistics: Posted by Humanoido — Sun Jun 03, 2007 4:41 am

AI

2007-06-02T06:45:18+01:00

Hi guys,
Sorry for the late entry on this thread. I've been off-line for a few months working for a living.

With robogames coming up I have been working on my replacement OS for the RN. Here's what I have so far:

I've named it MOOSE (My Own Operating System Executive)

I have fixed the FreeLoader.exe program..It can now upload MOOSE or any ATmega128 hex file to RN and it works!..Duh.. Why didn't I check my messages from i-bot about the botched XOR code 4 months ago.

MOOSE has I2C support. It can read and write to the external EEPROM chip.

It can play sounds in the background. (An ISR routine, not a blocked call)

The servo generating ISR is very very fast. Also ISR based. Lots of time to do other stuff.

Freeloader can load Positions and Sequences of Positions into the EEPROM.

The Positions are read by FreeLoader from a text file. They are in the same format as RoboBasic Moves except they are preceded with an up to 10 character name for the move and a speed byte from 0 to 0xF. 0xF being the fastest. There are a lot of great moves out there and I wanted to stay compatible.

IE.
PlCDcd,0,85,71,152,91,112,16,100,40,80,0,254,254,100,40,80,254,254,254,112,76,145,93,92,60

Sequences are lists of positions, variable in length, and can be looped (as in walking) and crossed into other sequences when a mutual move exists in both sequences. IE if WALK and TURNRIGHT both use Position 4 then you can transition from one sequence to the other dynamically without having to finish the WALK sequence.

I have been putting in some RoboBasic command emulation. I currently only support F0, F2 and AF. These are used by RoboBasic to download the RoboBasic OS if you don't like MOOSE.

Both USART ports actual work so strings of commands can be sent.

I have loaded two different RoboNovas and had them walk. I then converted them back to RoboBasic. No harm, no foul.

What's left to do;

LOTS

There is no trim function for the servos.
There is no servo position read.
PTP moves are by default but I may allow non-PTP moves at some point.
A larger RoboBasic command set would be nice to be able to use the existing programs to generate moves.
An I2C API would be nice for those that want to add external devices.
Analog/Digital converters and Gyro/Accelerometer/IMU support is important.

AND OF COURSE THE REASON FOR BEING IN THIS THREAD!

Some sort of AI for autonomous operations.

Here's the nice part..There is plenty of space left for code and tons of EEPROM available.

My push right now is to get the RN soccerbots ready for RobOlympics in two weeks. (Nothing like a deadline!)

Then I am taking requests for features and I will be releasing MOOSE to the public.

Please think about how YOU would like to see things done.
I am always open to suggestions.

Thankx,,,,
Dan

Statistics: Posted by DanAlbert — Sat Jun 02, 2007 6:45 am

AI

2007-06-02T04:39:45+01:00

AI for RN
What is the Most Simple AI?

I've obtained the powerhouse BS2px microcontroller and just now going over the programming and ports. I got a cable for programming from the local IT department, and best of all it's usb to usb so it can work with any of the newer NBs.

Anyone care to suggest some very simple AI applications that can be developed in my lifetime?

Well, I'm using the stock RN-1 so that means using motion, and sound. Plus I think some add-on vision would be possible. But how would you make simple vision work with AI? Any ideas? I think I can get RN to react to faces, and moving objects. Also some IR LEDs and receiver is quite common and easy to wire up for vision, plus some photo cells that measure light and changes in light could be very useful.

I think this is the question - what is the most simple AI?

humanoido

Statistics: Posted by Humanoido — Sat Jun 02, 2007 4:39 am

AI

2007-05-29T01:47:13+01:00

More on Linux and AI

http://linuxgazette.net/issue19/ai.html
http://www.linux.com/howtos/AI-Alife-HOWTO-2.shtml

I had also talked about Open Source AI and then
found this site!

http://openai.sourceforge.net/

humanoido

Statistics: Posted by Humanoido — Tue May 29, 2007 1:47 am

AI

2007-05-22T00:53:56+01:00

Linux Robotics: programming smarter robots By D. Jay Newman

I just bought this book. Has a lot of source code in the book or for download. It also looks like you can read most of the book online

http://books.google.com/books?id=mxHKcq ... =1#PPP1,M1

Statistics: Posted by Gort — Tue May 22, 2007 12:53 am

AI

2007-05-21T20:17:39+01:00

Pev, this is a brilliant idea. I've found a BS2px processor that has the power I'm looking for and it would make an ideal test bed. It has both serial and I2C. I'm looking into the interface wiring at this moment, and if the I2C interface can be pinned, then I plan to use it for AI.

Here are some specs on the board:

Technical Specifications
Processor Speed 32 MHz Turbo
Program Execution Speed ~19,000 instructions/sec.
RAM Size 38 Bytes (12 I/O, 26 Variable)
Scratch Pad RAM 128 Bytes
EEPROM (Program) Size 8 x 2K Bytes ~4,000 Instructions
I/O Pins 16+2 Dedicated Serial
Voltage Requirements 5 - 12 vdc
Current Draw at 5V 55 mA Run / 450 μA Sleep
PBASIC Commands 61
Size 1.2"x0.6"x0.4"

http://www.parallax.com/detail.asp?product_id=BS2PX-IC

Another processor I'm looking at is the propeller. It could be much more challenging to program. The one chip has eight 32-bit processors in one chip for real simultaneous multi-processing! (I can envision software dedicated to each processor, such as vision interpretation, interpretive speech, continuous speech recognition, ideas and cognition, and sensor arrays.) It took eight years to develop. It looks something like this:

Propeller Specifications
Package Types 40-pin DIP, 44-pin LQFP, 44-pin QFN
Model Number P8X32A
Parallax Stock Codes P8X32A-D40 - DIP package
P8X32A-Q44 - QFP package
P8X32A-M44 - QFN package
Power Requirements 3.3 volts DC
External Clock Speed DC to 80 MHz
(4 MHz to 8 MHz with Clock PLL running)
Internal RC Oscillator 12 MHz or 20 KHz
System Clock Speed DC to 80 MHz
Global RAM/ROM 64 K bytes; 32K RAM / 32 K ROM
Processor RAM 2 K bytes each
RAM/ROM Organization 32 bits (4 bytes or 1 long)
I/O Pins 32
Current Source/Sink per I/O 40 mA

http://www.parallax.com/propeller/index.asp

Previously I didn't want to tie up the RN serial port, thinking about adding Bluetooth. But now, there's no need for BT when an AI package is installed. It can make the decisions, not the large computer.

Pev, I would highly encourage you to go ahead with the circuit you described using the parts and board you have. I will continue along the same venue using the parts I have. It looks like this could be one powerful AI system for all RNs, and perhaps all humanoids.

If we get a powerful workable AI hardware system platform developed, one idea would be to open source the software. Considering the complexity of AI, it's going to take a lot of human minds working together to create such a vast system.

humanoido

Statistics: Posted by Humanoido — Mon May 21, 2007 8:17 pm

AI

2007-05-19T20:32:10+01:00

Humanoido

As the MR-C3024 is not by default an I2C slave how about this:

1. Add one (initailly) additional AI processor with both serial and I2C interaces.
2. Slave the MR-C3024 to the AI processor over the serial connection for calling motions.
3. Slave sensors and additional sub-processors to the AI processor over I2C.

Thus the 'intelligence' can utilise the sensors etc to make descisions and then call the appropriate motion.

I Think I have enough stuff knocking about to try this, certainly have an MR-8 board and couple of I2C sensors knocking about.

Pev

Statistics: Posted by Pev — Sat May 19, 2007 8:32 pm

AI

2007-05-19T12:37:23+01:00

Very Important Key Quote:
Pev Wrote:

So that has the code for making the MR-C3024 an I2C master. The tricky bit would be making it a slave so that it's "functions" could be called over I2C so for example a motion could be called by writing a motion number (11 for walk forward) to an I2C slave register.

Then a Master Processor (the Basic X24 for example) could act a the primary and call the motions and also communicate with other I2C processors or units such as sonar LCDs etc etc

All very theoretical in reality other than the fact I have made an MR-8 CPU Board acting as a slave I2C device for another project driven by an AMTEL processor. The Software for that system was all written in BASCOM basic.

The Key to Developing Robonova AI
Indeed, for AI, I think we are on to something here that can be very useful. The key is to develop an interface to go from the MR-C3024 to another MCU, using I2C. As I understand, this would be ideal for accessing numerous sensors capable of I2C communications. Say we use the MR-C3024 as the dedicated motion board. And say we use an add-on processor, be it the MR-8 CPU board, Basic Stamp IIpx, Basic X24, or other board. This opens up an entire new door for very powerful AI processing.
--------------------
I'm starting to look for and evaluate microprocessor boards with I2C built in for AI add-on to Robonova. If anyone has any suggestions, and/or is constructing some simple systems to illustrate the concept, I'd be glad to hear about it. Also some references to "working I2C systems" with multiple processors would be helpful.

humanoido

Statistics: Posted by Humanoido — Sat May 19, 2007 12:37 pm

AI

2007-05-16T19:29:09+01:00

Humanoido wrote:
So what would be involved in using this approach with the MR-C3024, I2C interface, and Basic X24? (hardware and software) or basic stamp, etc...

humanoido

Now there is a leading question. Firstly I found these post from KurtE:

*********************************************************
I finally had some success with bit banging out an I2C interface in RoboBasic. I ended up not using the SDA/SCL pins on the board as they are used by an EEPROM on the board and it is unclear in RoboBasic what IOPORTS are assigned to thise signals. Currenty I am using the ATOI 1 (33) and ATOI 2 (34) pins for SCL and SDA.

I have now can do pings on a Devantech SRF10 Range Finder.

*********************************************************
I believe a copy of the source code is still up at:
http://robosavvy.com/Builders/KurtE/I2C_Test.bas

The ATI pins are up upper right hand side of the board. At the point I did the post I was using ATI 1 and 2 as 0 was used for a GYRO. Currently I am using ATI 2 and 3 and I am now playing with two gyrsos and they use 0 and 1.

I hope this helps.
**********************************************************

So that has the code for making the MR-C3024 an I2C master. The tricky bit would be making it a slave so that it's "functions" could be called over I2C so for example a motion could be called by writing a motion number (11 for walk forward) to an I2C slave register.

Then a Master Processor ( the Basic X24 for example) could act a the primary and call the motions and also communicate with other I2C processors or units such as sonar LCDs etc etc

All very theoretical in reality other than the fact I have made an MR-8 CPU Board acting as a slave I2C device for another project driven by an AMTEL processor.

The Software for that system was all written in BASCOM basic.

Pev

Statistics: Posted by Pev — Wed May 16, 2007 7:29 pm

AI

2007-05-16T17:00:39+01:00

So what would be involved in using this approach with the MR-C3024, I2C interface, and Basic X24? (hardware and software) or basic stamp, etc...

humanoido

Statistics: Posted by Humanoido — Wed May 16, 2007 5:00 pm

AI

2007-05-15T16:51:27+01:00

I also remember a post ages ago about bit banging one of the other ports to support an I2C interface. I think it was KurtE who posted it but I may be mistaken

Pev

Statistics: Posted by Pev — Tue May 15, 2007 4:51 pm

AI

2007-05-15T15:17:53+01:00

Th MR-C3024 uses the 2 wire interface to connect to the program EEPROM. On power up it searches for the EEPROM in the range 0 to 3. The board EEPROM is at address 3. I use this to plug in extra memories at address 0 and know this works.

If you appear to the controller as a slave and respond to adddress 0 as a slave, then you could send byte code instructions like group moves for execution. There are some timeouts, so likely you would have to use an interrupt routine. Just send a NOP (FF), if you dont want any actions

Statistics: Posted by i-Bot — Tue May 15, 2007 3:17 pm

AI

2007-05-15T14:38:09+01:00

Humanoido wrote:
Pev wrote:
Nested/Dirstributed processing has been an approach used for many years for complex robotics control. I have used it myself on many occasions. With regard to the RN 1 I believe the dustributed approach is more than vali but would suggest looking at the I2C bus for communications rather than the ETX/ERX ports.

You are absolutely correct about nested distributed processing. This technique works well for extremely powerful microprocessors that can handle nesting and produce fast results. Being that we are at the hobby level, at least in my tiny lab, pretty much the only thing affordable would be some small dedicated hardware processors, like the Basic Stamp and those previously mentioned. IC2 is a great idea for communications. It has implementations on most microcontrollers.

humanoido

Sorry I obviously wasn't clear. I have built nested/distributed processing systems using ATMega8 processors and I2C with a Basic X24 as the master controller. The ATMega 8 processors are less than £2 each and the BASCOM compiler is available as a free demo (up to 4k of code I think)

Pev

Statistics: Posted by Pev — Tue May 15, 2007 2:38 pm

AI

2007-05-15T11:05:44+01:00

There are a lot of AI programs posted on the net for download. Many of these involve conversations that can remember data during the dialog. Some various good sources for this and other AI info are at:

http://www.simtel.net/category.php%5Bid ... simtel.net
http://www.alanturing.net/turing_archiv ... %20AI.html
http://www.developer.com/java/other/article.php/1546201
http://www.agt.net/public/bmarshal/aiparts/readme.htm
http://www.markwatson.com/
http://norvig.com/paip.html
http://library.thinkquest.org/18242/essays.shtml
http://blogs.msdn.com/devschool/archive ... gence.aspx

humanoido

Statistics: Posted by Humanoido — Tue May 15, 2007 11:05 am