Controversial opinion: Introducing (optional and adjustable) rubberband AI would significantly improve the single player experience of AMS2. Modern racing sims usually suffer from the AI performing wildly differently on different tracks, even once you account for the player's own skills at those individual tracks. Now, I'm not saying Reiza isn't a competent company that's consistently improving the experience, but it's a tweaking hole that has no bottom, and it will literally never get to a point where it's perfect, or maybe even really good. A little real-time number fudging would go a long way to gloss over its deficiencies, though. I have no idea how the game works from a technical standpoint, so none of the following might be actually feasible, but I'm just throwing it out there to illustrate what kind of rubberbanding I mean. Let's add three numbers alongside the difficulty setting: Target position - the race (and/or quali) position where the difficulty is exactly as you set it. Magnitude - The maximum allowed deviation from the set difficulty Range - How far from the target position you need to be to reach the maximum difficulty deviation So, for example, let's say we have difficulty at 100, and we want to aim to be in the upper third of a 20-car field, which is rougly 7th place. Let's set some simple, easy to calculate parameters of (Position=7, Magnitude=4, Range=4). This would change the AI difficulty live based on your race position like so: ... 2nd: 104 3rd: 104 4th: 103 5th: 102 6th: 101 7th: 100 8th: 99 9th: 98 10th: 97 11th: 96 12th: 96 and so on Does that seem like too extreme of an adjustment? Or too little? It's up to you. Tighter ranges and higher magnitudes give more obvious banding. But that's just nonsense I made up on the spot for illustrative purposes. The real point I'm making here is one that I was going to make even before that Empty Box video dropped, funnily enough, and it goes like this: The illusion of the race is more important than the simulation of it. It's not something I expected to say myself, as someone who's really into dynamic simulations, but it's true. Honestly, it's really weird that the Codemasters F1 games are the only sims on the market that do even a token effort of making you feel like you're part of an actual race weekend and a race, instead of kinda just existing on track with a bunch of bots. Reiza and other companies have the capacity to do better, it's just a matter of readjusting priorities. Anyway, rant over.
When sims begin to have real intelligent AI using deep learning and neural network in the AI bots - there will be no reason to use such sidestepping "intelligence" illusions. The first move into this direction can be seen in the Anna and Sophy algoritms.
When do you think that is and are you optimistic enough to think that AMS2 will ever see such a solution?
I hate having to share a jimmer video on the reiza forums, but this is relevant to why we dont want rubberbanding:
Actually I dont know. I have only watched from the sidefence - skimming news about mainly Sophy. But it sounds like it is rather close to be used in the higher ends AI bots. Check out this easily understandable article from a guy who tried to compete against some of Sophys different personalities Racing Against GT Sophy: Reflecting on Gran Turismo 7's new AI Driver Conserning Reiza then it will probably take a looong time - if ever
That's really the key thing here. Just because of the complexity of the sim, we're going to have a fiddly singleplayer experience for a long time - possibly forever - unless a more heavy-handed approach is taken in an effort to make the AI experience more consistent.
Neural network AI doesn't make the bots more fun to race against, it just makes them inhumanly strong and tedious to race against (Sophy is infamous for pushing humans out of the way). Just like trying to play Chess against a computer, it sucks.
Instead of just being stubborn negative then check as example the link out - and/or ask yourself if you consider the un-Intelligent AI bots ai AMS2 more state of the art than this new wave in real AI is now. Or aspire to be in a reasonable future.
But its true. The AI optimises its performance to try to win at GT7. It turns out driving like a maniac in GT7 is how to win at that game, because its not punished for bad behaviour.
Oh yeah if you say so. Hehe this sounds exactly how human sim drivers does mostly behave online But if you read up a tiny bit of how Neural Networks can be implemented then its obvious that there allways will be some trial and errors in the beginning of the use of this pretty human brain-ish way of controlling virtual cars. And do you honestly think that such as you describe some variants(!) of Sophy is the end of the development process? But anyway instead of rigid konfrontation then follow this link to an explanation of what neural networks is and can be. Examples Of Dynamic Networks In Deep Learning? A dynamic network is a neural network that can change its structure during the learning process. This allows the network to adapt to the data it is processing, which can improve performance. Dynamic networks are typically more difficult to train than static networks, but they can be more powerful. What Is Dynamic Neural Network – Surfactants
I think allready they have ironed this behavior out of Sophy - eventhough "she" set a new WR on Nord driving this way
