Image Image Image




Post new topic Reply to topic  [ 19 posts ] 
Author Message
 Post subject: RoboCup
PostPosted: Wed Jan 16, 2013 12:29 pm 
Online
Junior member
User avatar

Posts: 35
Favourite Bot: Working on it...
Hello,

This isn't poker based but it is quite AI based.
I'm doing a team project at uni on RoboCub Football, basically a heads up match.
I'm hoping to focus on the strategy of the game.

I was wondering if any has any experience or ideas for strategy?
The main and probably one that will be used is an A* search.

The idea of the project is to build an autonomous lego robot that will play against an opponent.

If you're bored or need a break from poker ai it would be nice to hear some ideas!

Thanks


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Thu Jan 17, 2013 12:26 am 
Offline
PokerAI fellow
User avatar

Posts: 1115
Favourite Bot: Johnny #5
Football = soccer, right?

Is the robot going to be bipedal? Is there a standard "body" that is used among all contestants? What kind of sensory mechanisms are employed? Cameras? Radar? Will it have ficken lasers on it's head?


Top
 Profile  
 
 Post subject: Re: RoboCup
PostPosted: Thu Jan 17, 2013 2:44 pm 
Online
Junior member
User avatar

Posts: 35
Favourite Bot: Working on it...
Oops, soccer of course!
Here's an example from previous years:
https://www.youtube.com/watch?v=e4M29WvsVxU

It's generally a block of lego with wheels, bipedal would be cool but not quite our level.
It's possible to use a set of holonomic wheels but the design will be a lot simpler with standard uni-directional wheels.
There's a camera set up above the pitch, the images can then be translated to locations of ball/goal/opponent/self and then the strategy does it's part and finally the instructions are passed onto robot, e.g. move forward 100cm, kick ball etc...

Sensors can be used, mostly touch, to sense when the robot is in contact with the ball, or the wall.

It's quite simple and it seems as if most strategies are relatively basic too. I was thinking about adding some opponent modelling but since the strategies in 1vs1 don't seem very technical, it might be overkill!


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Thu Jan 17, 2013 3:29 pm 
Offline
PokerAI fellow
User avatar

Posts: 1239
Favourite Bot: my bot
if oppo between ball and oppo goal
then bounce ball off wall

Is the robot standard and you just play with controller?
Where are the rules?


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Thu Jan 17, 2013 3:39 pm 
Online
Junior member
User avatar

Posts: 35
Favourite Bot: Working on it...
The robots will be fully autonomous, i've attached the rules in a pdf.
That's definitely a strategy that's been used in previous years, the harder part tends to be calculating the angle and making sure the robot hits the ball with enough speed and correctly.


Attachments:
football-rules.pdf [42.18 KB]
Downloaded 10 times
Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Thu Jan 17, 2013 4:19 pm 
Offline
PokerAI fellow
User avatar

Posts: 1239
Favourite Bot: my bot
- It looks as though whoever gets to the ball earliest after kickoff has a big advantage.
- After a goal is score, are the ball and machines repositioned? Where?
- If the robot had a protruding bumper at the rear the robot could use this to scoop the ball to one side in the initial face off
- If the front bumper could be rotated around the vertical axis, then even if the robots collided head on the ball could be squeezed out to one side. Better still a wheel rotating around a vertical axis could push the ball out to either side
- When I said kick the ball into wall I just meant as a tactic to get round oppo, not to shoot at goal
- If the two robots are the physically the same then this problem become one of game theory. If they just run towards the ball from the goal line they just block one another. If they try to play the ball round oppo they become exploitable. To avoid that you could dummy one one way and then play the other. When you know the physical properties of the robots you might be able to simulate this and find nash equilibria and best responses


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Thu Jan 17, 2013 7:30 pm 
Online
Junior member
User avatar

Posts: 35
Favourite Bot: Working on it...
Quote:
- It looks as though whoever gets to the ball earliest after kickoff has a big advantage.

Yep, that's true. The games are a few minutes long so hopefully it shouldn't be too big a problem, as long as the bot has a good defensive strategy.

Quote:
- After a goal is score, are the ball and machines repositioned? Where?

Both robots are repositioned to their goal lines and the ball is placed in the center again.

Quote:
- If the robot had a protruding bumper at the rear the robot could use this to scoop the ball to one side in the initial face off
- If the front bumper could be rotated around the vertical axis, then even if the robots collided head on the ball could be squeezed out to one side. Better still a wheel rotating around a vertical axis could push the ball out to either side

I like this idea, i've not seen it in any of the previous designs and does make a lot of sense. There are quite a few situations were the typical designs are stuck where this function would help.

Quote:
- If the two robots are the physically the same then this problem become one of game theory. If they just run towards the ball from the goal line they just block one another. If they try to play the ball round oppo they become exploitable. To avoid that you could dummy one one way and then play the other. When you know the physical properties of the robots you might be able to simulate this and find nash equilibria and best responses

Good idea as well, we do currently have a simulator but more basic strategies, e.g. A* search and a bezier curve around blocking object. I think they're quite simple though and in practice none seem to work particularly well.
I'll definitely have a look into nash equilibria and best responses, i've not used them in my poker bot so would be a good way to learn.
Speed may be an issue but it's worth trying.

Thanks for the interest, i'll post if there are any interesting developments.


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Thu Jan 17, 2013 9:42 pm 
Offline
PokerAI fellow
User avatar

Posts: 1239
Favourite Bot: my bot
I've been looking at holonomic wheels. It seems to me you break out of head to head jams quite effectively with them. Also if the bot could turn on it's own axis and had a circular base you could use that to squeeze the ball out to one side.

http://www.youtube.com/watch?v=KXbTorONqT0


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Thu Jan 17, 2013 11:27 pm 
Offline
PokerAI fellow
User avatar

Posts: 1239
Favourite Bot: my bot
I'm getting the impression that the lego processor isn't very fast. If so maybe you could develop a bunch of rules using a simulator. Then transfer those rules to the bot that has a simple engine that can interpret them. You'd have to interpolate some positions, but that is not so hard. Seems like you could do a lot more in simulation than on the real thing.


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Fri Jan 18, 2013 11:38 am 
Offline
PokerAI fellow
User avatar

Posts: 1115
Favourite Bot: Johnny #5
Did I misunderstand or did you say that the bot uses the overhead camera to discern the game state?

I wonder if GAs wouldn't shine in a game like this?


Top
 Profile  
 
 Post subject: Re: RoboCup
PostPosted: Sat Jan 19, 2013 11:40 am 
Online
Junior member
User avatar

Posts: 35
Favourite Bot: Working on it...
Quote:
I've been looking at holonomic wheels.

We have a choice of holonomic wheels but unfortunately they lab only has a few sets and we'd have to order in from elsewhere - which could take 2/3 weeks. We'd have to decide soon but chances are it won't be worth the lost time in waiting and figuring out how to use them in the future. Annoyingly..

Quote:
I'm getting the impression that the lego processor isn't very fast. If so maybe you could develop a bunch of rules using a simulator. Then transfer those rules to the bot that has a simple engine that can interpret them. You'd have to interpolate some positions, but that is not so hard. Seems like you could do a lot more in simulation than on the real thing.

As the lego processor isn't fast, most of the computing is performed on a desktop computer and the instructions bluetoothed to the robot. The main consideration here would be time, if code has to be exectuted very quickly, will the lag in the bluetooth connection be too long? We've not tested it yet though.
I do feel that a simple strategy could work just as well in this case as it is only 1vs1.. but working on something ai based is more fun :)

Quote:
Did I misunderstand or did you say that the bot uses the overhead camera to discern the game state?

Yep, it does. It's a relatively simply procedure, implementing it isn't so simple. Each robot has a different colour, along with the ball. The picture is split into R/G/B and then darker areas are detected and the centroids are set as the locations.

Quote:
I wonder if GAs wouldn't shine in a game like this?

I hadn't even thought it about it, but now looking only there is actually a fair amount of information on GAs and NNs solving path finding problems, thanks for the idea!
One study found that the GAs were up to 1000% faster than an A* search, could be worth looking into.


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Sat Jan 19, 2013 1:00 pm 
Offline
PokerAI fellow
User avatar

Posts: 1115
Favourite Bot: Johnny #5
Not just path finding but strategy development as well. You could create a 'virtual soccer field' and play out simulations (offline) to develop your AI... Maybe use a basket of hand-written tactics for the GA to choose from? The biggest problem with this approach, as I see it, would be translating the virtual world strategies to live play. Maybe you could add in various kinds of noise to the virtual world, for example, to reflect things like ball english, etc.


Top
 Profile  
 
 Post subject: Re: RoboCup
PostPosted: Sat Jan 19, 2013 1:15 pm 
Offline
PokerAI fellow
User avatar

Posts: 1239
Favourite Bot: my bot
http://lejos.sourceforge.net/forum/view ... f=6&t=3696


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Sat Jan 19, 2013 1:27 pm 
Offline
PokerAI fellow
User avatar

Posts: 1239
Favourite Bot: my bot
- So most of the intelligence is on the PC? What language are you using?
- Do you have a simulation framework in place/mind?
- When is the competition?
- Robot assembly doesn't take long does it? Most of the time will be in software development?


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Sat Jan 19, 2013 1:54 pm 
Online
Junior member
User avatar

Posts: 35
Favourite Bot: Working on it...
Quote:
You could create a 'virtual soccer field' and play out simulations (offline) to develop your AI...
There's a basic simulator that's been implemented from a previous year for testing strategies, i'm sure it wouldn't be too hard to figure out and use to run simulations. I don't know much about GAs but like said above, i'm definitely interested in figuring them out.

Quote:
- So most of the intelligence is on the PC? What language are you using?

Yep. Any language can be used, but we're currently using Python for the vision and Java for the rest (strategy and control etc..).

Quote:
- When is the competition?

We have several milestones which need completed every 2/3 weeks. However, they shouldn't be too difficult. The final day is on 3rd April and the main tournament is hosted then. We will play friendlies before hand though for practice.

Quote:
- Robot assembly doesn't take long does it? Most of the time will be in software development?

In theory, we have a team of 10 people and it feels like there is slightly more focus on the robot building instead of the software development. This will probably change, but I think that most time should be spent on strategy.
All previous years have had a decent robot designs, but the majority have flawed strategies...

I guess it's almost similar to poker botting in the sense that when lots begin they want to have a stealthy interface with the client, before they even begin the ai, and then only after realise that the strategy is what takes most time (definitely not something i did :xx19 )

PS thanks for the link, i've messaged them so we'll see if anything comes from it!


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Sat Jan 19, 2013 2:06 pm 
Offline
PokerAI fellow
User avatar

Posts: 1115
Favourite Bot: Johnny #5
I wonder if, using the plethora of videos online, you could model past opponent's strategies and come up with counter strategies?

How is a strategy discretized (in the framework)? How are tactical utilities discerned?

How many people are working on the software aspect?


Top
 Profile  
 
 Post subject: Re: RoboCup
PostPosted: Sat Jan 19, 2013 3:02 pm 
Offline
PokerAI fellow
User avatar

Posts: 1239
Favourite Bot: my bot
Can you post the simulation framework


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Mon Jan 21, 2013 12:44 am 
Online
Junior member
User avatar

Posts: 35
Favourite Bot: Working on it...
Quote:
I wonder if, using the plethora of videos online, you could model past opponent's strategies and come up with counter strategies?

It's definitely a possibility, however It think there are lots of dud strategies that would have to get filtered out as they would probably be counteractive.
Once filtered out it could have some potential.

Quote:
How is a strategy discretized (in the framework)? How are tactical utilities discerned?

At the moment two of us are working through the code and writing up some documentation/refactoring what we can. So (If i understand you correctly) as far as I can understand there are several strategies which all solve independent areas of the game.
There is a main game strategy which incorporates the smaller strategies.
Basically, although the code is pretty messy, it looks more or less like a bunch of if-then-else rules, which decide which minor strategy to use.

The minor strategies all have a step function, this, as it sounds, decides the next step/movement for the robot to make.

For example, at the beginning of a game, the initial strategy is to approach the ball at an angle and try to hit it around the opponent. This uses InitialBezierStrategy, which in turns takes a snapshot of the game, looks at the robots current position, calculates the path and then sends a message to the bluetooth controller to move the robot in the path direction.

Hopefully that makes sense...

Quote:
How many people are working on the software aspect?


At the moment, two on vision, two on the rest. But we'll have two more from tomorrow onwards who should be working on software too.

Quote:
Can you post the simulation framework


As far as I know it shouldn't be a problem, all the code is available online for previous years.
I'll pm you both the link, just in case :?:


Top
 Profile E-mail  
 
 Post subject: Re: RoboCup
PostPosted: Mon Jan 21, 2013 9:54 am 
Offline
PokerAI fellow
User avatar

Posts: 1239
Favourite Bot: my bot
Code:
nfo rotacaster info@rotacaster.com.au via gmail.com
21:51 (11 hours ago)

to Form, paul.bart.spea.
Hi paul,

Thanks for your enquiry.
The delivery time to the UK is +/-4 Days.

Kind regards
Carly


On Sat, Jan 19, 2013 at 11:14 PM, Form Submission at the Webconsole <Contact_Us@rotacaster.com.au> wrote:
The following email was sent from the Contact Us form on http://www.rotacaster.com.au/;

Submission Amount:   :     507
First Name   :     paul
Last Name   :     spears
Company Name   :     
Email Address   :     paul.bart.spears@gmail.com
Country   :     United Kingdom
Phone Number (incl. Area Code)   :     12345678
How did you hear about us?   :     Google Search
Message   :     What is the delivery time for to the uk 48mm and 125mm Rotacaster lego robotics wheels?
I wish to subscribe to the Rotacaster Newsletter to receive product updates, news and events   :     No Thanks


I'm expecting a response from their uk distributors http://www.varley.co.uk/ today


Top
 Profile E-mail  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 19 posts ] 


Who is online

Users browsing this forum: No registered users and 4 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to: