Madden 17 Deep Dive - A Look Inside CPU Playcalling

[timhere1970]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

Playbooks should be constantly updated or users should be able to use custom playbooks for all teams.

[toneloc129]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

Wow! Once again, great critique with excellent solutions offered and Once again, EA should hire you ASAP

[D81SKINS]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

Awesome post! I really hope EA looks into this, they could make the CPU A.I. so much more competitive just by implementing more team specific plays/more variety on 3rd dwns and 2 mins drills in the star system.and not have EVERY team become so predictable.

Just look at how lame the 2min drill is for all teams, it's like they just copy pasted it for the whole league, smh

[SteelD34KC]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

Can you just modify the default playbooks and then the AI will automatically use them? IE KC Andy Ried offense and defense, modify those then the AI of KC will use them?

[NDAlum]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

No you need to create a CPB and load it for the CPU. It sucks.

[Chocolate_Milk]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

Maybe I am just incredibly naive but this is just downright embarrassing from EA that this is an issue when the fix on their side seems very easy. It's like this isn't anyone's department or job to consider ai playcalling.

[wellred]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

I dont think they just leave some playcalls off the gameplan for the hell of it. The ai is just better at executing certain plays rather than others. This isnt anything new to madden, guys who played the old NES games can confirm that. The plays that have zero stars I suspect have that rating for a reason. This is evident when calling coach suggested plays yourself on offense. They usually work very well. Thats the same gameplan the cpu runs.

It would be nice to have the cpu have more variety in playcalls but I dont really care what they call as long as they can move the ball.


People really shouldnt compare it to basketball or other sports games. Football has more moving parts and hundreds of little thing happening on any given play. Tech wise were a probably 30-50 years short of emulating bill walsh adaptive ai on consoles.

[timhere1970]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

When I make a playbook for offense. I rate every play in the book but for only one circumstance. This keeps me from using the same play repetively. I also go to the playbook and change a few star ratings for each circumstance before each game. This changes what plays are recommended, not necessarily recommending the higher star ones more as it should though.

[DeuceDouglas]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

I dont think they just leave some playcalls off the gameplan for the hell of it. The ai is just better at executing certain plays rather than others. This isnt anything new to madden, guys who played the old NES games can confirm that. The plays that have zero stars I suspect have that rating for a reason. This is evident when calling coach suggested plays yourself on offense. They usually work very well. Thats the same gameplan the cpu runs.

I disagree to a certain extent because a lot of plays they do choose to put into the GamePlan don't really benefit the AI from an execution standpoint. There are a lot of plays that can pretty much be hard countered by certain defenses and end up in sacks a good majority of the time because the AI can't handle them if they're covered. Curl flats, four verticals and corner strike are good examples of this. It doesn't mean they won't execute them properly sometimes, I've just noticed that a lot of plays where baffling things happen are when the CPU chooses to run these plays.

[DeuceDouglas]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

Trying to think about more solutions, I think they'd really benefit from moving the UI into CFM and having it be more of a franchise oriented feature like it originally was. With the emphasis on playing games quicker with Play the Moment, they could essentially bring back the original premise of GamePlan which was to have it call plays for you based on how you create your GamePlan and have it coincide with, ironically, GamePlanning. The first step would be having the tendency tracking actually work for the CPU teams but since we're more concerned with CPU playcalling here that doesn't matter too much.

But basically the idea would be that within each playbook, it could have, let's just say five, unique GamePlans. They would be Run Heavy, Run balanced, Balanced, Pass Balanced, Pass Heavy which are all fairly self explanatory. Each week based on who you were playing and how you were playing during a given season would change how teams call plays against you. Each team would have a preferred GamePlan which would essentially be their base offense but let's say you were 1st against the run and 32nd against the pass, a team might shift to their pass balanced or pass heavy game plan for the week to try and attack your weaknesses. Or maybe they're a good running team instead of going pass heavy they go from a run balanced game plan to a balanced game plan when playing you just to throw in more passes. Conversely, each week you'd practice a couple concepts to attack your opponents weaknesses and be made aware of what plays should be more effective versus them during the game you'd be suggested those plays more often.

I'm making that sound a lot more complicated than it is but the idea still remains to have the CPU be able to change and adapt their playstyle and more importantly utilize more of their playbook over the course of multiple seasons. As it currently stands, if you play the CPU using an Andy Reid playbook in 2016 or 2056, it's going to play pretty much the same. You're going to see the same plays called and not much is going to really change. All the potential is already implemented for them to potentially blow this out of the water, it's just a matter of dedicating time and utilizing it to it's fullest potential.

[wellred]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

I disagree to a certain extent because a lot of plays they do choose to put into the GamePlan don't really benefit the AI from an execution standpoint. There are a lot of plays that can pretty much be hard countered by certain defenses and end up in sacks a good majority of the time because the AI can't handle them if they're covered. Curl flats, four verticals and corner strike are good examples of this. It doesn't mean they won't execute them properly sometimes, I've just noticed that a lot of plays where baffling things happen are when the CPU chooses to run these plays.

Of course they can be countered with the right defense, that goes for all play designs, but can the cpu effectively execute those plays? Yes. Can they execute other plays as well? When Im calling the 'no star' type plays I notice theyre prone to be a lot more unpredictable and chaotic as they unfold. I just dont think the madden ai works like 'this guy is open so Ill throw it to him'. I think the gameplan is just a crutch to hide its pretty obvious deficencies.

Ill put it this way, when you snap the ball youre scanning the whole field. Ive playded so much madden Im pretty well aware of where every guy on the field is. Its like a chess player thinking 8 moves ahead. The cpu ai doesnt work like that. It knows what D you called and it waits for certain criteria to be met and it just reacts. Hence why theyre prone to stand in the pocket while guys are wide open in the flat for a 15 yard gain, they wont look at it until a defensive player is so close to them. A human player is feasting on that as soon as they see it. Just my thoughts.

[goalieump413]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

Great post and discussion everyone! I really hadn't been paying attention to this, but you're right. This is an issue that deserves some degree of attention as it seems that CPU playcalling is so narrow in scope that AI teams don't even need large playbooks.

But I had a thought regarding fixes that EA could implement in a patch...

Variation of formation, motion, and hot routes based on human controlled defensive alignments. I know the AI audibles, but how many times has anyone seen a receiver go in motion, or an AI qb bark a hot route call to a WR, or a formation shift at the LOS? I see what look like audible calls at the LOS, but not the other three. This alone could produce enough variation in play results that the bigger issue could be postponed until next year's game. I generally don't like the idea of sidestepping the true issue though.

The idea is just to do more with what you have. If EA doesn't implement a change the the volume of plays it actually uses per game/season, then at the very least, introduce non-vanilla options to every play that is called. For me, when I run a 4 vert play, I routinely hot route one of the WR's to runt a slant or out to give me more options.

[NimitsTexan]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

I really doubt EA can implement AI hot routes in a patch, at least in any way that would be helpful (rather than completely random) to the AI.

[HailTheSkins]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

Man I hope EA looks at this and listens to you! Great post

[GrayDawg]

Re: Madden 17 Deep Dive - A Look Inside CPU Playcalling

Deuce great post! I also have been wanting to see some changes to the playbook system. Since jumping into the custom playbooks using the 2nd controller, I have really taken a look at team playbooks and many of them are not even close to what some teams are running this year even from just a formation standpoint.

Also, the 3rd and long plays are really bad as well as 4th and long. I really wish they had a situation geared for screen plays and gaining better field position. For example, what you would call on 3rd and 12 could be very different from 3rd and 20.

Additionally, part of the problem as it relates to EA is that there is so much focus on online user vs user that offline has been ignored in some key areas over the past few years but this hopefully will change with the current folks in charge. We saw this last year with very little effort being put into CFM.

When looking for customized playbooks at least so far, many are geared towards using playbooks themselves with very little emphasis on game planning. So hopefully that changes in the near future. Also, want to thank those of you who have been working hard on making good playbooks. You have given me some good ideas moving forward.