How the computer-controlled cars or bikes race is one of the most important aspects of any racing game. Do they just follow each other on the racing line, or are they able to battle and fight? Will they pressure you when behind and defend when ahead? Are they likely to jump out of the way to a divebomb or will they defend and punish you for recklessness?
As the genre of racing games grows so do the ways in which their program their racers. MotoGP 19 has a new concept for racers this year; Neural AI.
What is Neural AI?
Machine learning is exciting, but it might as well be blockchain or compensating for gravometric interference for all that most people would care or understand. To put it simply in this instance, it’s a computer that has been presented with the challenge of racing a bike under the watchful eye of Milestone and Orobix engineers. Unlike the Drivatars of Forza, these AI don’t learn their racing craft from real world players or references from real races, it’s all figured out from scratch.
Milestone claim their use of neural AI and machine learning enables the computer-controlled riders to exploit your mistakes. Keep scrubbing off a little too much speed at the end of the back straight at Autódromo Termas de Río Hondo? The AI will figure that out and ride round the outside of you.
It’s fascinating to see different iterations of the neural AI as it literally learns how to ride a bike. Luckily, it’s like riding a bike and the AI never forgets, and so through round after round of competing with different AI routines constantly trying to improve itself, it goes from the motorbike equivalent of a fawn that barely knows what legs are to something that looks as smooth and fast as real world racers.
According to Milestone, the training phase based on the reinforcement learning concept lasted several months, with more than 200K races simulated daily. The result is, supposedly, an exceedingly realistic and natural driving behaviour that will be ‘aggressive but fair’. At any rate, MotoGP 19 will also feature customizable difficulty levels to cater to both casual and hardcore players.
What does it feel like to race them?
They feel almost indescribably more alive than the last few MotoGP games. The opening laps have all the riders jostling for position, swarming through corners and making it easy to win as many places on one corner as you’ll lose the next. They’re not perfect – I did have one bike practically fly past me into the gravel trap as I went through a corner – but I found them enjoyable and Milestone freely admit that they’re just the first iteration of this exciting technology, while also promising to update the AI after launch.
How does it feel compared to past MotoGP games?
MotoGP games in the past have been accused of being stale in both challenge and competition. They designed this AI with reinvigorating the formula in mind. This translates to the way the riders race. Instead of being scalextric bikes on one racing line and you forcing your way past, this game the riders feel more real in the way they attack and defend. If you go wide, which you will early on as everyone knows how difficult these games are to get right. The AI will be there to stuff one up the inside.
Overall, compared to other motorbike games there is much better progress. However, we will have a better picture once everyone's AI's have learnt more about the racing. In a few month's time we will see if those who put the hours in are rewarded with an AI that races a lot closer because it has had more time to learn.
Is this technology revolutionary though?
In all honesty, no. It looks great on paper and it may prove me wrong by those who play day in day out. Those that can see the AI improve, but right now especially on the career mode the field spread is too large. When you watch a Moto 3 race, the field is separated by maybe 10-15 seconds. On the game the top 5-6 can become detached from the rest and get stale very quickly.
The reason for this is that if you fall off and crash, you only lose 1-2 positions rather than the penalty you should pay which should drop you further down the field. It also means if you're fighting in the midfield on a difficulty you enjoy the top guys are just unattainable which is something no racer wants to see. Right now I would argue the AI on F1 is slightly better. That all being said though, compared to other racing simulators like Forza 7 for example this is more realistic. The concept is clearly there, just needs a bit more time to refine.
What do you think of the Neural AI? Let us know in the comments below!