computerized analysis of team tactic

[FatMan]

I don't know if this is deep blue material - there no physics or hardware in the usual sense - only computers, and tactics - but its still pretty technical, so you decide.

A large entity that shall remain nameless is funding us to do research to the tune $150K a year or so. What we want to do is build a computerized system to analyze small team combat tactics. We plan to demonstrate the system using paintball.

The heart of this thing is a computer vision system using multiple cameras to build a very detailed real-time map of an area and its occupancy. IOW, where stuff is. We will have to augment it to identify players (but that won't be hard). We also want to measure things like heart rate and respiration and so on.

Using this, we will be able to record games, not just optically, but being able to see around objects, and see where things were and how they were moving at any give time. We will try to develop the ability to measure 3D data selectively on the players (too much data to do the whole court).

Now, here is what I'm NOT sure how to do - and what the great minds of deep blue might be able to help us with: Once we have this information, how can we use it to develop training for the players and actually improve their game? This is, ultimately, the goal - and while we know how to gather the data for evalutation, the experience to use that may be harder to find.

Any ideas? Don't be shy, we're open to anything!

FatMan

[Brak]

you got a damn-sweet job. it took me a few minutes to think, and this got the rusty cogs of my head moving.

since its measuring and keeping track of most of the aspects of players and such, look at what happens when someone makes a mistake. if someone makes a bad move, look at their heart rate. after enough data is collected, you can see the optimal heart rate for being able to fuel the physical stress or paintball, and "panicking" for lack of a better word.

you can also look as one team the enemy, and another the good guys. and the enemy will have set parameters in terms of movement, strategy, and basically behavior. then try different ways of the good guys to eliminate the enemy. after trying different "good guy" strategies, you will find the most effecient, casualty-free strategy to win in that situation.

so what i think you should do is run different strategies and behaviors, and see what happens. if one person if left in front, measure the heart rate and body temperature. obviously it will go up, but see what kind of effects it had on their performance. and while doing that, orchestrate different situations and find out what people "should do" in those situtaions

if i have the wrong idea, you gotta excuse me, im a moron. but im a moron that wants to help.

[osiris]

Interesting idea. Very interesting. Likely not a paintball vendor, or marker vendor funding the effort...

OK... So you have all sorts of telemetry running plus cameras of the field, etc... etc...

The question to be answered, "How does one do CQB optimally?" Difficult question because there are multiple good solutions. This is, after all is said and done, going to be a hill climbing problem. You will need to be able to define "acceptable losses", preferrably as a percentage.

The problem is going to take a bunch of computing power, so start looking into Beowolf clusters. Try to parallelize as much as you are able.

Try to break the problem down into multiple zones of cooperation. What I mean is, you will have to figure out what is optimal with one, two, three, ..., and N players, in your game. You can't simply break it down by team. Everything matters... unfortunately.

Try to think of each player as a vessel, with it's own space and it's own overlap with another (some other) vessel's space. That's going to be the only approach which will work in the end.

I say that becauce I've helped solve similar problems in a different problem space.

You will need to discover the underlying principals which enhance individual survivability and the rules which govern when those principals do not apply... altruism IS what wins CQB. How will you train that?

One must take chances on behalf of his fellows, can that be trained into the individual? Probably, but is is morally bankrupt to train soldiers to sacrifice themselves on behalf of their friends? The answer to that question is a resounding NO, it is not, because when you see a young man who has NOT been trained to sacrifice himself AT THE RIGHT TIME, jump upon a grenade that he did not have to jump upon... well you begin to understand training helps, even when that training teaches a soldier when to sacrifice himself.

The things you are considering are beyond gravely serious. I do hope you realize that. If you don't, you will not satisfy the contract.

-m-

[robertjuric]

You asked for opinions/ideas.

But I dont see how it will help apply to other players. I think it would be kind of cool to know, but seriously how is knowing heartrate and all that stuff going to affect to people snappin it out??

[Redkey]

what about...

I assume you have thought of the following.

You will need to have baseline data for all the instrumented players. Otherwise your data might be hard to interpet since everyone is going to have different fitness levels. This way you could measure the percentage increase over normal operating conditions for certain events on the field.

You should include audio on the players as well. That way what they said to their teammates and heard from their teammates could be factored into the game time line. For instance... 7 man team loses two players on the right line, everyone but the left front guy hears the headcount call... data shows four guys with bumped heart rate while the guy who didn't hear remains normal.

Instrument the guns so you know when and how fast they are shooting, when they are reloaded, or even cleaned/breakdown and better yet... where they are pointed... you could probably accomplish this with each gun using a IR laser with it's own individual pulsing sequence. Any ccd video camera should be able to record the IR illumination.

You could even use the laser system on a guys mask to see which way he is looking.

You might also want to instrument the players with a GPS that uses a local transmitter (or whatever) to track the players position.

What to do with the info? You would be developing a very detailed step by step time line of the game that could be used by people to evaluate their teams performance, that of their opponents and how one set of tactics work against another.

Not sure what would be done with the players bio-signs though. Everyone is different... perhaps, over time, you could use it to track players personal development. Cross reference their psych profile with their improvement from their first tourny experience to one a couple years later.
With experince I'm sure you will see reductions in heart rates and respirations as the person gains experience.

you can paypal me my portion of the 150K for my input.

yuk yuk yuk...

[Smoking Nun]

conclusions

Conclusions will be much easier with more data...

In other words, collect as many different kinds of quantitative data as possible when doing this. Things like times, angles, shots fired, light, weather, etc. Collect as much numerical data as possible. Collect as many samples of this data as possible.

Then, don't just make observations and conclusions from the descriptive data (means, standard deviations, etc.) Make your conclusions about data through statistical testing. Only make conclusions when you find significant results. You can do statistical testing to test differences in means, to test for relationship, to test for predictability. It can all get very complex very quickly. But, to really get the useable results you can present with complete confidence - get someone good at statistical analysis to test your data once you have it. They will show you some amazing stuff you would never even think about and will prove wrong some of the most basic and seemingly obvious conclusions you might make.

See my post in this thread for more information. However, if you've already thought about statistically testing your data, then you are a step ahead of me. In which case, you will know that conclusions become much easier to make once you have statistically significant results. It leaves much less up for guessing and theorizing.

Feel free to contact me for more information.

Nun

[FatMan]

Thanks!

Wow, great feedback! Thanks you guys!

Here are some thoughts:

instrumenting the guns - we thought about that. The first thing is, that's been done before, so we get no research benefit from doing it BUT - it may indeed get us more important data. The real problem is instrumenting the guns has other issues - most teams won't want to have their expensive electros wired up like that. We might think about building a set of instrumented guns just for the project though. We can GPS the players easy enough(though the vid system is much more precise), and there is a device that will tell us the direction (in 3D) its pointing. I'm thinking an audio device can detect the firing rate ... maybe.

The heart rate stuff is to measure stress - that's pretty easy to do, and yeah it has to be calibrated to each player.

As for statistics - we're academics in engineering ... we do loads of statistics. We will certainly be paying attention to that.

So, some of the things *I* have been thinking of include things like evaluating the safety of a player's position by finding the angle needed to hit him in the bunker and comparing it to angles on opponents - and then looking at how well a team mate protects a front position by keeping opponents out of good shooting angles. And stuff like timing how players move within their bunker, and possibly leave open an opportunity to make a run and bunker move.

Ever see a guy jump up and run down the field and take out 3 or 4 players? I'm not talking recball, but in the pros and upper Ams. How do they do that? Can we see how timing and coordination add up to make that happen? Can we teach someone to read the signs to do it? Or is it just luck?

To be honest, I don't know if this will work ... and there may be other things we could measure that is more important.

Thanks again for the input, and keep the comments coming. I'm thinking more and more about how to instrument the guns - now I just need some guinea pigs to work on!

FatMan

[SlartyBartFast]

Re: Thanks!

Don't over do the data collection.

Probably don't need more than position of each player, the direction they-re looking, and the direction their marker is pointing (and that only when fired).

Have define what you absolutely want to analyse and the minimum requirements first....

[AGD]

Well first of all (trying not to sound redundant) we have already done real time EKG data acq. on police officers in paintball shooting scenarios. It's not too interesting since the heart rate goes up with mental stress not physical. They rev up when the realize it's conflict time.

I would like to see the following analysis on your data. In my mind you are trying to answer questions about team tactics, what works and what doesn't for proper training.

I would first group the scenarios into winners and loosers. Next I would run a spacial distance analysis to determine if physical distance between the team members showed any postive correlation. This would tell you if you should stay close or distant from your people and would require different communication and training styles.

Next I would look at movement patterns the same way. Does lots of short moves win or does a few long ones?

Then I would examine movement on the left and right. Is pushing a side more effective?

Lastly I would look at shooting patterns, fewer short bursts or lots of cover fire?

Answering these questions would give you clear directions on how to train a group of warriors, paintball or otherwise. It would indicate what types of equipment to implement. For instance if close group movement is important you would train hand signals. If you should spread out, radio communication would be the most important.

Very cool stuff, keep me posted Fatman.

AGD

[Miscue]

Perhaps you could keep track of everyone's field of vision and blind spots. How about taking the FOV of the team as a whole? Maybe looking for tunnel vision... how long people will focus on one area.

How often should I check both sides of my bunker? Statistically, what are my chances of bunkering someone who has not poked his head out in x seconds from a y distance?

After an opponent makes a move, how long does it take for the team to communicate this... and react?

How about keeping track of spacial advantage? Does being able to cover more of the field statistically give an advantage? What bunkers should we control to have a statistical advantage? Which bunkers should we avoid? Which opponent bunkers should we most worry about? I'd like to see a calculation of bunker value.

What are the odds of being hit at x range?

What moves can I statistically get away with?

[FatMan]

AGD, Miscue,

Those are both really good suggestions, and those are the kind of thing I was trying to find. Right now I'm not sure of the value of the heartrate stuff either. One approach we are considering is using this system real-time during training, by having a trainer who can watch the system, and communicate with the "practice dummies" directing when and how they move. In this case a trainer might want to decide which side to push or when to push by looking at the stress of the players being trained.

Anyway, its an idea. I think the stuff you guys were talking about is the most likely to be beneficial. I am hoping we can capture at least a rough idea of where players are looking, and where guns are shooting so we can do the kind of analysis Miscue suggests. Certainly positional information will be the first data available.

FatMan

[Miscue]

Maybe it would be good to keep track of each player's 'knowledge' of the field. Where are opponents I've seen? Where are opponents that I cannot see but I know are there... either through team communication or deduction. Where do I think they are? How valuable is a sneaky player that goes undetected? What disadvantage is there to being detected? Or does it matter?

What sort of an advantage does a team with more accurate knowledge have over one w/o it? Maybe with info like this, a team will decide to work on communication moreso than snapshooting during practice... as it's shown to be more valuable than what was thought. Then again, maybe not.

I'm thinking, that if you do a really good job with the analysis... you can construct rules for a computer simulation. Go to a new field... put stuff in the computer... it tells you what bunkers to play and such. Maybe figure out probabilities of what bunkers the opponents will take... something like this. The idea being that, you have some idea of what the game is going to look like before even playing it.

[rahimi_saeed]

Computerized analysis system is required

Dear FatMan,

I am going to be the paintball match analysor in Iran and am looking for an appropriate software for that,i would be thankful if you help me and show me the clues.




saeed

[latches109]

this already exists, it was done in the early 90's. I am sure it will be better this time around, edit:I'll see if I can get you a reference.

[Miscue]

Eh... old thread resurrection!

Forgot about this completely... but had some more thoughts.

I don't think that such a study would be fruitful. The variables are insane... I don't know how you could possibly create any kind of meaningful prediction on this stuff.

It would be easier to do a statistical analysis on a football game, trying to come up with a winning strategy. The field is always the same size and is 2-dimensional. The number of participants is always the same. There are no equipment differences to account for.

There are not any rules or standards in combat/warfare.

You'd have to focus on something very specific... limit the scope of what you're trying to develop a simulation for. But then what use would that be?

Time would be better spent researching more effective ways to kill people with technology. What do you need strategy for if you can easily over-power?