Automobilista 2 September 2024 Development Update - The Road to V1.6 Pt2

Um, Geoff Crammonds Grand Prix.

 

Simulation genius, I see it as Adrian Newey of F1 simulators, I followed all of the World Circuit, the first version on the Comodore Amiga and then released for PC, it was a watershed, I followed this evolution and I can't understand a company like Reiza not being able to make a decent AI, you see the AI drivers unable to pass, even on long straights with better cars, I hope one day they can.

 

So you are saying Geoff Crammonds Grand Prix, a game with 16-17 tracks is the same as AMS2, a game with 183 track variants? You are right. Obviously it is. How could I not see it. Makes perfect sense now. Dammit, they should have had flag waving marshals in AMS2 already in 1991. What's taken them 33 years!?

 

The ai in his games were genius, a bit like the ai in Halo. That just blew me away.

 

It's very difficult to balance the AI with so many cars and tracks. Reiza will solve it, and it will be amazing. It’s true that for its time, Geoff Crammond's work is impressive, and in my opinion, it has the best AI by far. Plus, the level of immersion a game like GP2 from 1996 offers is incredible

 

The sim has great ai for me as long as I do a quali session, and it's going to be even better in 1.6. And visual marshalls with flags instead of a pop up onscreen would be the icing on the cake for me in vr.

 

This is one of the issues the audio devs are working on. Agree, it is not pleasant to have only one sound for all types of curbs.

 

GP4 is one of my favorite games of all time and the AI is incredibly good, but it had its flaws. They were overly aggressive and optimistic, often ruining your race because of some very erratic dive bombs and massive swerves when defending, plus they had a massive advantage on wet to dry races due to being impossibly accurate driving on the dry line.

Were they have no equal yet is in how they simulated perfectly the overall flow of a real F1 race of the time. Their strategies, pace, mistakes, retirements, consistency, how they became more error prone in wet conditions, etc. Made for incredibly immersive and dynamic races.

AMS2 is IMHO the closest of the modern sims in recreating that immersion and dynamism, it's not yet there, specially in the strategies as the AI only uses the soft tires, but the races can be pretty immersive, and you always have different results. Also I enjoy the wheel to wheel against the AI more in AMS2 than GP4, even more so in the beta.

 

I suspect if we had dozens of streaming videos of GP4, the memories of how amazing it was might come into a slightly clearer focus. It was the best in its day, but I don't think the AI in it could cope with what we ask of the cars in AMS 2 (various dynamic conditions across a wide array of divergent series and vehicles).

 

A lot easier to dial in Ai for one class of cars for 16 tracks, than it is for 183 track variants, a crapton of cars with various different performance levels, multiclass, ovals, dirt, a variety of wet conditions, puddles, etc.

 

I think the point was that even back then they had working flags, don't really think track count comes into it.

 

Reiza have said that track side marshals are coming, just have patience.

 

Then again is that not a self-imposed challenge ? It obviously is easier to do 16 tasks than 183, but if you set to do 183 then they should eventually be made equal quality or better because that was the end goal. Or at least as best as possible quality.

 

I'm (almost) sure this only happens with paid content (DLC) so, most likely, is a content verification problem.
Free content stays selected as it should.

While at sounds, please Reiza, do change the mix of the other cars sounds.
Just for the sake of argument, the graph below shows in red what AMS2 does and in blue what LMU does. I'm won't say what LMU does is real and AMS2 is not.
I don't mind if it's fake, it seems more natural to my ears, it makes the other car much "more present" beside me.

ps - another thing: many times, a car is louder when approaching than when it is moving away from us - given most of a car's noise is projected to the rear, I would expect the opposite

AMS2 Ai has short-sightedness. Yesterday I launched LMU and when I stay stationary in the middle of the track (on purpose ), they conduct evasive maneuvers long way before they reachs us. In the same conditions, AMS2 Ai often collide with us, even with spare space on both sides.
From the videos I saw, AI in v1.6 still does this.

Reconfiguring AI for all those tracks/car combinations every time something changes, is not just an herculean task but also a frustating one.
A sim like AMS2 does need a better way to do this or else further developments become increasingly difficult until it is just economically impossible.

I would like options to remove all text/artificial information from the screen. Just plain driver in car view.
My rig has everything I need and CrewChief is a better way to inform otherwise.

 

I missed that , That’s good news. I’ve had too many race ending or ruining accidents because of a car under no yellow flag sitting in a precarious position. Normally Nordschleife or Bathurst where there’s little room to avoid contact.

 

Fewer tracks, simpler rules and a visual and animation standard that's isn't anywhere near the current one. Didn't marshalls stand on the track then too (last one I played was GP3)? Apples and oranges somewhat.

 

"I would like options to remove all text/artificial information from the screen. Just plain driver in car view.
My rig has everything I need and CrewChief is a better way to inform otherwise."

That's in the game from the beginning, you just have to assign a button to the hud cycle, one of its positions is nothing on screen.

 

While I agree that AMS2 generally gives the feeling that it operates with more limited predictive/forward thinking strategy (not sure if that could be implemented in AMS2 but I can see how beneficial at times that could be) I would also say that we should refrain from glorifying LMU AI as benchmark for everything.
For instance, I was casually watching the yesterday the stream of a well known RF2/LMU streamer/fan that I will not name doing a 24 HR race against AI (saving every pitstop, hopefully something AMS2 will allow us in a not too distant future ) and AIs consistently managed to stick their nose on the inside of the player doing half a....ed attempts at passing him and resulting more often than not with the AI spinning alone just because they exceeded the available grip in the attempt with no contact in the process (especially at Tertre Rouge).
That is to say that:

1. Every AI can be improved, there is none of them perfect. In the end it is an AI and by nature it will be imperfect and exploitable
2. If we select one as the "be-all-end-all" trait of an AI (be it the capability to overtake or the consistency of their driving without spinning while attempting an overtake) everything can be blown out of proportion and then driven to a conclusion that AI is unplayable

Apart from this, I would also like to see AIs following each other (and the player for that sake) at a bit more than 1 inch as it was shown in Yannick's video as that feels slightly not plausible apart from increasing the possibility of crashes with the player. Surely Reiza is not oblivious of that situation.
But again, based on my experience of the Beta, saying that AI isn't showing any changes from v1.5 is plainly and factually not true (generic comment to some of the posts in this thread, nothing specific or personal intended). No matter how we turn the words it's just not the truth.

 

Can you tell me what makes the AI so good ? I am curious by what metrics it is judged, what it can do or what range of situations it can be in.

 

At least in part thats in RL a thing because of the Doppler effect - higher frequencies are usually better (and stronger) to hear than lower ones - I will not explain the Doppler effect - inform yourself