R.E.A.L. AI in FIGHT NIGHT ROUND 4

Hey everyone, I’m the AI Lead for Fight Night Round 4, and I’d like to take you on a tour of the work my team has done and give you a glimpse of what you’ll be enjoying come June 25th!  A few months ago, when our producers started taking the stage and unveiling massive feature sets and unprecedented visuals, one of the announcements was that we’re going to be featuring REAL AI.  I’d like to give you a glimpse of how this new technology will bring our extensive boxer roster, past and present, to life in Fight Night 4.

Styles make fights

One of our biggest features for Round 4 has got to be the addition of in-fighting through the use of physics, not just because we’re able to give our fans the most realistic boxing experience ever, but it gives my AI team a bigger canvas to work on.  It gives us the opportunity to use contrasting styles to highlight what boxing strategy is all about.  With the addition of the inside fighter, we can challenge the user to stay at jab’s length if they’re playing as Ali, or challenge them to get in past the jab if they’re fighting against him.  You really do need to play differently based on your reach and your opponents reach.

Positioning in the ring is also very important as a boxer.  Working an outside fighter AI into a corner can take some skill, you have to learn to cut off the ring and use your punches to overwhelm the opponent, while avoiding the jab flurry.  Once you have him in the corner, make your shots count or else the tables could turn and the hunter can become the hunted!
We feature 7 main boxing styles: Conventional, Boxer Puncher, Brawler, Inside, Outside, Counter Puncher and Unconventional.   But the differentiation doesn’t end there, our team watched a lot of classic footage over the entire project, and we’ve been adjusting the AI to represent the uniqueness of each boxer style and punch output.

R.E.A.L. AI

R.E.A.L. stands for Record, Evaluate, Adapt and Learn.   These actions are the key to the AI Technology that has been developed over the past few years and is really on the cutting edge of how AI in games is made.  One of the major differences in the AI in Fight Night is that it is now completely data driven.  In Round 3, each boxer style actually had hard coded punch combinations which made the game very repetitive.  Using R.E.A.L. AI technology, our production and testing teams taught the AI System how to move, punch, block and counter by playing the game as the boxer they wanted to create.  We’re able to reproduce all the fluid movements available to us from the Physics engine.  In Round 4, each boxer has thousands and thousands of individually recorded sequences, making sure no two fights are ever the same!

If that wasn’t enough, the R.E.A.L. AI technology allows us to adjust the specific boxer styles for specific boxers, because even within boxing styles boxers are unique.  Ali hardly throws to the body, Frazier’s sweeping hook combinations, Roy Jones Jr’s tendencies to throw leading rights and the list goes on and on.

Difficulty Levels

Fight Night 4 features 4 different difficulty levels that have been tuned to fit all player styles.  Amateur is great for new users wanting to experience Fight Night for the first time, Pro satisfies most of our regular gamers.  For those users that have been playing Fight Night for a while, Champion difficulty level will give you a good run for your money.  During production we looked at all of our user feedback and we wanted to give those who love the series something to keep them challenged. We added a fourth difficulty level just for them!  If you’re looking for the fight of your life, challenge Tyson to a heavyweight fight on G.O.A.T. (Greatest Of All Time), you might just be able to get an achievement or trophy out of it, (just watch out for that uppercut!!).

Data Driven

The change to use a fully data driven system allows us to do things on a scale never before possible.  Traditionally, AI is typically a set of rules that the CPU follows based on a set of conditions.  IF this, THEN do this.  You can imagine that then the AI is limited to only the conditions listed by the programmer.  When you move to a data driven system, like REAL AI, you’re able to make connections from game states to actions that is difficult to comprehend.  I made the image above a few months ago of what Mike Tyson’s AI ‘Brain’ looks like as trained by our experts.  Each connection represents the relationships between actions and game situations, tens of thousands in total.  Isn’t that so cool?