The date is January 23rd, 2014 and it’s roughly five months before GRID Autosport will be in the hands of players around the world.
It’s gone 6pm and the office is slowly winding down, the sound of the cleaners and their trusty Henry vacuums are resonating throughout the building and the motion sensor lights within the corridors rarely illuminate, blackness ensues. The office is a strange place after working hours as the emptiness of it all has a strange feeling about it, almost eerie in a way.
While some of the team are still hard at work, the vast majority are already on their way home, it’s another day of work complete and we’re one day closer to announcing GRID Autosport to the world, a date that is creeping up on us at an alarming pace.
Located upon the second floor of our HQ is a meeting room called Neon and from within, a discussion is taking place, one that would shape the very foundations of GRID Autosport.
…get it right and everyone’s happy, get it wrong and…well…your Community lets you know.
The subject of the discussion is handling, an area within racing games that can quite literally make or break the game, get it right and everyone’s happy, get it wrong and…well…your Community lets you know.
Those around the table are faced with the classic “stick or twist” dilemma. Everything we’ve been building towards rests on the choice we make in that room. The way the cars handle when you’re playing the game will be based on that very decision. No pressure then.
On the one hand we have a solid, authentic handling model. On the other hand, we have a new improved handling model, which we created following feedback from our community, and which we’ve tested with Autosport magazine journalists, members of the team and racing drivers. We love it, it’s a big improvement. So, simple decision right? Just pull the trigger and make it happen?
Well, to understand why this decision is creating the wringing of hands and tense glances around the room, you’d need to know a bit more about the way in which our handling model is linked to the rest of the game.
Let’s start with the change itself first. We want to change the way then grip falls away as you approach the slip angle and as wheelspin occurs. We’ve traditionally had quite a ‘cliff’ here, with grip falling away very quickly. Based on some of the observations of our community playing the game, we decided that if there was a slightly more gradual falloff you’d have the chance to correct mistakes and more room to push to the limit with each car.
Making this change means first changing the grip falloff graph across over 80 cars, and then going back through each of those cars and tuning the handling all over again to make sure we haven’t unbalanced the delicate setup of each, and that we’ve remained true to their character.
Some cars are relatively ok, whereas some, such as the Open Wheel cars and some of the American Muscle cars, need a complete rethink with this new model.
However, the story goes deeper than that.
It’s no secret that we take huge pride in our AI within Racing Studio, and the process involved in making human-like AI on track is extremely labour-intensive.
It’s no secret that we take huge pride in our AI within Racing Studio, and the process involved in making human-like AI on track is extremely labour-intensive. Firstly, we have dedicated members of the game design team who do nothing but set benchmark times for the AI to use.
That means playing every car on every track in the game to make sure that we have a human benchmark for the AI to hit.
The AI then uses the information created by the level design team on where they are allowed to drive, what line they should use, where they should brake for corners and so on. With the benchmark time and the level design information combined, we then have everything we need to run our special in-house training tool on the AI, making it drive each track hundreds of times to fine-tune its driving line to try and hit the human-set time.
Finally, only when all of that is completed, our AI designer, Neil Atkinson, and our AI programmer, Pete Edwards, can use this solid data to start creating all the behaviours you see in race, ranging from the propensity to make mistakes, all the way through to the basic race pace the AI will use in at each skill level.
At the point of making the handling decision, we are already well underway setting the AI benchmark times, with about 30% coverage of all the career tracks. Making the handling change means the cars will drive differently, which, in turn resets the clock on the AI benchmarking work.
Yet, even though this will mean re-doing weeks and weeks of work, Joe Smith, setting our benchmark times, is up for the challenge. He’s played the new handling, he can see its way better, and he’s up for going again.
So we look carefully at the schedules, which are really, really tight, and we make the decision to go for it. We pushed hard to get the changes rolled out and the AI times set again and balanced. The result is the game you see now, and the handling model that you soon will be able to play.
We hope you’ve enjoyed this little glimpse behind the curtain of racing development at Codemasters, and hopefully this gives you an idea of the hand craft that goes into making each of our games. Let us know if you’d like more articles like this in the future!