[Home]Katana

Robo Home | Changes | Preferences | AllPages

Version History

Author

UnderDark

Extends

AdvancedRobot (my first)

What's special about it?

Its my first comp bot the extends AR

Great, I want to try it. Where can I download it?

http://www.robocoderepository.com/BotDetail.jsp?id=2498

How competitive is it?

Unkown

How does it move?

AntiGravityMovement

How does it fire?

CircularTargeting

How does it dodge bullets?

AntiGravityMovement

How does the melee strategy differ from one-on-one strategy?

It doesn't

How does it select a target to attack/avoid in melee?

It doesn't

What does it save between rounds and matches?

Learning Multi-Mode System Data

Where did you get the name?

From the Japanese sword

Can I use your code?

No

What other robot(s) is it based on?


Comments, questions, feedback:

what kind of NN do you use, and what is it used for. it doesn't seem to be used for the movement or gun. -- rozu

It a basic Punish/Reward? system. The NN is actually been set in this release to not make use of the MultiMode features. This is do to problems that have come from porting the Robot code to AdvancedRobot. To see my NN in action, you'll have to look at UnderDark3 right now. -- UnderDark

Ah ok, I probably wouldn't call it a Neural Net in this case. If I understand you right it's normal multimode stuff like: if round won then plus 1 otherwise minus 1, next round take best mode... rozu

Yes, and also for Bullet Hits, Misses, Rams, and Time as well. I call it a NN for the fact that it does learn what works, and what doesn't against other bots. If my definition is innacurate, please tell me so. -- UnderDark

A Library to build a NeuralNetwork can be found here: NeuraLib. I wouldn't call it a NN becaus you don't train a network in this case. I don't know a lot about NNs, but I think there you train the weights between the nodes in the network. Anyway, multiple movements have mostly been a hard thing to implement (a working one would be for example the MusashiTrick). It's mostly better to work on one movement that works fine then on many of them. -- rozu

Rozu is right NeuralNetworks uses a simplified simulation of how bilogical brains learn. It's about caving out the right paths in the neural synapse system in order to fulfill a particular task. So, unless you have neurons and such involved it's not NN. Though it can still be AI and certainly it can be a learning system. If you're interested in trying out real NN learning you can check out my NN bots, Orca, OrcaM and GB which all are NN bots. They use a free NN library. And they are all failures though. I think maybe I have not understood NN enough... -- PEZ

I see, as such, I have changed the data title to "Learning Multi-Mode System". Thanks for this insight --UnderDark


Robo Home | Changes | Preferences | AllPages
Edit text of this page | View other revisions
Last edited January 30, 2005 18:12 EST by UnderDark (diff)
Search: