Subject:
|
Re: Analysis of robot behaviours?
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Sun, 1 May 2005 17:42:24 GMT
|
Viewed:
|
1071 times
|
| |
|
|
Hi Brian,
Thx alot for the reply!!
I´m sorry. I was so........ unclear. So I will try to explain the problem again.
But first little info about the robots. There are two robots, one is the prey (3
bump sensors, 1 rotation sensor) that just drive around and try to avoid things
and it has a lamp on it. The other one is the hunter(2 bump sensors and 2 light
sensors). It avoid things and try to follow the prey.
The firmware I use is BrickOS and the programs runs only on the RCX units.
To teach the neural network the tobot do an action and after that it evaluate
the result of that action by measuring the speed, check if all the bumper
sensors are off and if both light sensors exceeds a threshold value.
Exactly what you suggested!
My question is really how do I evaluate the robots behaviour after I´ve run the
program, so I can see if it helps for example if I change the learning rate for
the neural network or if I change the sensors position.
This most be something people that are studying ethnology most know about.
Measuring the difference of effectiveness of a hehavioar between animals or
individuals in refernce to some specific value function, for example energy. Is
there a standardize method for this?
> Or, more interestingly, have the light sensors positioned on a structure
> that the robot can adjust, and have the neural network optimize this too.
That is really a cool idea!
Cheers,
Johan
|
|
Message has 2 Replies:
 | | Re: Analysis of robot behaviours?
|
| On Sun, May 1, 2005 12:42 pm, Johan Hjertman said: (...) ... (...) I would think you'll want to run some controlled tests. For example, dis-connect the motors on the prey, and see how long it takes the hunter to "catch" it. Try several positions, (...) (19 years ago, 2-May-05, to lugnet.robotics)
|
Message is in Reply To:
| | Re: Analysis of robot behaviours?
|
| (...) Generally, with a neural network, just write an evaluation function that generates a number and use that as your "standard" (...) What is considered important here for a "successful" result? Speed ("the faster a find a bright light the (...) (19 years ago, 1-May-05, to lugnet.robotics)
|
6 Messages in This Thread:
  
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
This Message and its Replies on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
