Subject:
|
Re: AI in RCX
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Sat, 4 Nov 2000 17:23:06 GMT
|
Original-From:
|
Rama Hoetzlein <RCH8@saynotospamCORNELL.EDU>
|
Viewed:
|
717 times
|
| |
|
|
You picked a challenging project.
It is not possible for the RCX to do Neural Networks alone, since the RCX
does not have the command set, and is not a fast enough processor to
calculate the weights of the Neural Network. Also, the RCX doesn't have
ultra-sonic sensors, so you would need to provide these yourself.
If you >really< want to do what you said, the >only< approach is to take
input from the RCX and the sensors, bring it into the PC, do Neural
Network calculations on the PC, and send commands back the the RCX. You
would need a >LOT< of customized code on the PC to do the following:
- receive input from the RCX via IR tower. (serial port programming in
C/C++ perhaps)
- receive sensor data from the ultra-sonic sensors. (how are they
connected to PC?)
- train the Neural Network on PC (train the correct responses for given
sensor inputs)
- query the Neural Network on PC (determine how to act based on new
sensors inputs)
- send motor commands back to RCX
Because you want the RCX to respond to objects in real-time, all code must
run in real-time except for the training (which can be done on the PC
before hand). This is possible, but very very difficult.
I would suggest, instead, that you focus on one aspect of the project or
another. Either discard the RCX and try and do a project just with Neural
Networks (like bringing sensor data into PC, and doing Neural Network
object recognition on PC)...
Or... discard Neural Networks, and try and get the RCX to do something
simple like a Finite State Machine. A good source for Finite State Machine
AI is "Tricks of the Game Programming Gurus" (vol. I and II). This is a
very good starting point for robot AI, and you won't have to solve neural
network issues and ultra-sonic sensor issues...
If this is your first project programming the RCX without using LegoRCX
code, don't expect too much at first. Focus on simplest problems first,
then build up.
Rama Hoetzlein
rch8@cornell.edu
Yiu wrote:
> I want to program AI into RCX
> that can avoid objects by the readings of 3 three ultra-sonic sensors.
>
> I am supposed to use Neural Network.
> But I difficult as the programming is limited by the RCX
> (even RCX 2.0 is used)
>
> Would you be kind enough to give me a piece of advice?
>
> Please give me any info or link that could be useful
>
> Thanks
>
> Yiu
>
|
|
Message is in Reply To:
 | | AI in RCX
|
| I want to program AI into RCX that can avoid objects by the readings of 3 three ultra-sonic sensors. I am supposed to use Neural Network. But I difficult as the programming is limited by the RCX (even RCX 2.0 is used) Would you be kind enough to (...) (24 years ago, 4-Nov-00, to lugnet.robotics)
|
4 Messages in This Thread:
 
 
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
