im now have the very old UE 3D3 LowLevel fatal Error Bug , thats the Pain in the A** for UE folks since Years in games and in the UE SDK
Oh , yes the great famous UE not the uncomplicated Biggest engine from what many spoken (seeing the cubic explosion from Mechs ingame)
only Solution ,start the mission with most low Seetings (all without thats will have a Restart) and set up the Seetings in the Mission =no Problem , after mission ,im set the seetings down or its crashed with the next mission.
Edited by MW Waldorf Statler, 30 November 2019 - 03:50 AM.
It's actually more complicated to make the AI *miss* the player than it is to land a hit, since you have to give it a bunch of random parameters to offset the shot etc. It's pretty simple for an AI to hit you with 100% accuracy all the time, but it's not very fun. A bad AI would track and hit you with 100% accuracy all the time.
Just about every single player game I've ever played its pretty easy if you hide somewhere at the back of the level and snipe at stuff.
*edit*
That said, where *is* the demo? I can't see anything on Epic store or MW5 homepage.
I have also found that the enemies you engage repeatedly are likely to pursue you in return, and the ones you haven't engaged don't come after you until you do shoot at them. Thus.. the AI is reacting to what you do and what your lance mates are doing (whether they're co-op humans or not).
That's the problem, you have gone too low. You picked the easiest mission availabe (probably designed for noobs), with mostly recruit or green rated enemies. What do you expect from recruit enemies, of cause they are supposed to miss a lot. Your veteran "doppelganger" Awesome however hit a lot of his shots, just as expected. I can imagine this Awesome to kill you if you try this mission unarmored, and i can imagine a mission filled with such veterans to be more challenging.
The only really bad AI thing i have seen in your video is the locust not realizing that his friends are attacked. On the other hand, later on that locust tried to flank you while you were fighting the Archer, that's not too shabby.
In the end i'm pretty sure the demo does not feature the most difficult AI the game can field against you. So for example i have hope that light mechs on higher difficulties will stand still less often, that's what bothered me most in the demo AI. If your locust would have been on the move all the time, you would have had a very hard time hitting it.
Also i think in the demo enemy weapons simply deal less damage as they should to give us a smooth ride. This could also be very different in the final product.
Edit: Regarding B33Fs clip: It's very obvious that the AI uses an aggro system like MMORPGs do, so they attack the most dangerous threat. Since B33f did not fire, he was no threat and the AI focused on his much more dangerous buddies. That's way smarter than most MWO players who keep firing the unarmed sticks to pad up their KD ratios...
However this behaviour can probably be exploited, if they don't do proxyimity checks. I will try to do this mission with a backstabbing loadout and see if they let me do that to them if i approach them without firing...
Your arguments are ridicilous.
I can solo pretty much any mission on the list. The assassination is actually one of the harder missions, not the easiest. And I don't consider myself a very good player even when I try.
Also, like I have said a dozen times: accuracy isn't an indication of good AI. Accuracy can easily be changed. Even with 100% accuracy, the AI is still bad. Stop using this as some sort of proof that the "final AI will probably be good". You don't know that. We have ZERO proof that the final AI will be any better than this.
Regarding TheB33fs clip: there hasn't been a single MW-game where AI hasn't shot at the player if the player hasn't shot at the AI. It's pretty obvious there is some sort of aggro system, but it's a BAD aggro system. There is a stationary assault mech on the field, but the AI is more concerned of a Commando running in circles (lol at the crappy lancemate AI as well). This is NOT good AI, stop pretending everything is fine.
It's actually more complicated to make the AI *miss* the player than it is to land a hit, since you have to give it a bunch of random parameters to offset the shot etc. It's pretty simple for an AI to hit you with 100% accuracy all the time, but it's not very fun. A bad AI would track and hit you with 100% accuracy all the time.
Just about every single player game I've ever played its pretty easy if you hide somewhere at the back of the level and snipe at stuff.
*edit*
That said, where *is* the demo? I can't see anything on Epic store or MW5 homepage.
It's not very hard to make an AI that occasionally misses players, and it's even easier to make an AI that never misses players. Accuracy is not an indication of good AI, except when it's ridicilously done (such as the tank that misses like 7/10 shots in point blank range).
However, accuracy is not what I have my issues with. The poor actions and reactions of the AI are. It's not even on 1990's level of AI design. Period.
The demo is not available to anyone but those who pre-ordered the game in the start of the year. If you've done that, you should have a redeem code in your profile on mw5mercs.com
I think what we see in B33Fs video is the Ai prioritising targets after how dangerouse they are. A mech that does nothing but standing still isn't a threat compared to the ones shooting.
Someone else allready noted that behavior when you make a run for your dropship. The Ai lets you escape as long as you don't look or fire at them. The moment you turn around and open fire it will focus back on you.
So it's prioritizing a Commando that's pretty much just running in circles next to B33f, maybe shooting a single laser occationally instead of removing a completely stationary assault mech from the field? Really good AI!!1
Quote
So the Ai follows a very simple pattern that is very benifitial for the player. Is that bad or broken? Well depends on the person you ask I think. Some will say its good because it can make the game easier. Like when you are heavly damaged and limp back to your drop, that way you have a chance to get there instead of beeing forced to repeate the mission or load a savegame because you where shut down on the last meter. Others will say it takes away the challange of a coordinated retreat or choosing the best path back to your ship. Like looking for cover and so on.
To me the question comes down to a design decision and that was, I think, in favor of a more accessable gameplay then hardcore simulation.
Yes, and that's what I'm criticizing. It's bad design. If you want to make a game more accessible, you put difficulty options in it - you don't craft the whole game for the lowest common denominator. It's very easy to make a capable AI easier to defeat by reducing variables such as accuracy or damage, but making a useless AI harder is a nightmare, because increasing these variables will only create a frustrating experience.
If you've made stupid decisions in your mission, and have to limp all the way back to extraction (there isn't even limping by the way, as getting legged doesn't even slow you down except for a moment... so), and then manage to die because you make even more stupid decisions, you deserve it. That's called challenge, and it's what makes games fun.
It's funny how now "this isn't supposed to be a hardcore simulation" is being used to defend abysmally bad AI, when previously people were whining that they want a more hardcore simulation experience because "MWO is too much of a twitch fps for kids". So, I guess some people only want to stomp around in big mechs with explosions around them and dominate every other vehicle on the field by just existing to satisfy some childhood power fantasy or something. For me, it's not doing it.
I'm here trying to raise enough concern to perhaps get PGI to rethink their attitude towards the state of their game. They have obviously had a too big of hardon for their procedural maps instead of concentrating on basic gameplay issues. If enough people would actually support me on this, we might have a good game on our hands.
But no, instead you're all cynically attacking ME for wanting a good game. Why?
I'm here trying to raise enough concern to perhaps get PGI to rethink their attitude towards the state of their game. They have obviously had a too big of hardon for their procedural maps instead of concentrating on basic gameplay issues. If enough people would actually support me on this, we might have a good game on our hands.
But no, instead you're all cynically attacking ME for wanting a good game. Why?
LOL. people have been trying to to that to PGI about MWO for years and years and it hasn't worked. Why do you think it will work for MW5?
Monkey Lover, on 29 November 2019 - 10:27 PM, said:
It's set on easy mode. I see nothing wrong with the AI. They could set it to kill you in seconds if they wanted. When the full game comes out it should have options for this.
Show me videos on the AI getting stuck behind a building, not shooting at you or something if you want to call it broken.
You are aware you can't set the difficulty...there is just the difficulty of the mission and on warzone it's easier to survive at difficulty 90 than at difficulty 40.
If you can set it I've seen no way to do so. Each mission has its own difficulty which corresponds to the number of simultaneous enemies and nothing more. The more enemies there are the slower they shoot and the less accurate they are and the more often they kill each other.
To note difficulty 40 is the setting on another mission.
I'm here trying to raise enough concern to perhaps get PGI to rethink their attitude towards the state of their game.
I think pgi knows about it. The squirrels likely tested it which probably led to the past delays. Problem is they are also yes men.
Overall the real issue is the game is too late in the cycle to fix the issue. At best you can hope for a day one or week one patch.
However as I have learned ai in ue4 is easy as **** to program. Far easier than it was when I worked on my ff7-2 mod for deus ex and tried to create a Skyrim esque open world without ever knowing about the elder scrolls series until 2006 (which was when I lost the hard drive and most of my work).
The main issue is the need for foresight for situations and knowledge of how the game is played.
Both are things pgi either lacks or does not fully understand. (Players tend to do this. Why do players do this? Uh, I don't know they just do.)
LTC Kilgore, on 30 November 2019 - 08:33 AM, said:
LOL. people have been trying to to that to PGI about MWO for years and years and it hasn't worked. Why do you think it will work for MW5?
So? One of the reasons it's not working is because (some) people are white knighting the developer for crappy practices and blaming anyone who has criticisms instead of coming together.
Still doesn't mean I won't try.
This is a new game, and it's a single player experience which they put on EGS to get visibility. They know they can't just sell mech packs to loyal little poodles to make MW5 profitable. So they might actually care what kind of first impression youtube- and twitch-videos from us who played the demo leave on potential new customers. As in, they might actually care enough to patch the AI if we tell them the current iteration is not going to make them money.
And rest assured, the poor AI WILL be the center of attention in reviews unless it's much better in release version.
edit: Oh, and by the way. Back in MWO beta they tried to push an even crappeir micro transaction model for paints. I made a multiple option vote against it, and the overwhelming majority of players voted for a different kind of model. It showed CLEARLY what players were ready to pay for. After that they changed that model. So yea, we can change things.
There isn't limping but being legged slows you down to a percentage of your speed without allowing you to go back to full. Though another bug I've found is that under unknown conditions sometimes you can get 100% speed back inexplicably. However this has been an exception not the norm. (I've broken numerous legs trying to DFA mechs.)
(as a light against heavier ones)
If your max speed is less than 40 you won't be affected at all.
I'm here trying to raise enough concern to perhaps get PGI to rethink their attitude towards the state of their game. They have obviously had a too big of hardon for their procedural maps instead of concentrating on basic gameplay issues. If enough people would actually support me on this, we might have a good game on our hands.
But no, instead you're all cynically attacking ME for wanting a good game. Why?
I had often wondered what people where talking about when they described MW4
then I realized i played the game less then 10 hrs
I could complain all I wanted but it would not have done any good
so basically you pay your money and take your chances unless you can get a refund your wasting your breath
oh I can give you a long list of games that I played less then 4hrs (Strike Eagle III cost $65 and sucked)
I stopped buying PC games in the 1990's because 95% of them sucked
If you've made stupid decisions in your mission, and have to limp all the way back to extraction (there isn't even limping by the way, as getting legged doesn't even slow you down except for a moment... so), and then manage to die because you make even more stupid decisions, you deserve it. That's called challenge, and it's what makes games fun.
Limping, legging, lets not split hair here. You are slower then normal.
Beside that, getting your leg shot off dosn't need to be part of makeing a bad decision but can also come down to blind luck...or bad luck, depending on your point of view.
Walk around a corner of a building and catch and AC20 to the knee, legs of and whos fault is it? No ones because its a random event and don't come me with "you should have spotet" or "a good player can see that comming". A random event like that can happen to the best.
So not everything is because of "players are stupid and deserve to die".
Quote
It's funny how now "this isn't supposed to be a hardcore simulation" is being used to defend abysmally bad AI, when previously people were whining that they want a more hardcore simulation experience because "MWO is too much of a twitch fps for kids". So, I guess some people only want to stomp around in big mechs with explosions around them and dominate every other vehicle on the field by just existing to satisfy some childhood power fantasy or something. For me, it's not doing it.
There have allways been and most likely will allways be the two kinds of players, those who like it easy and those who prefer it as hard as it gets...thats what she said......
Anyway when you look at the groups of players out in the market, I would say that the "GG EZ" people are in the majority. As a dev you mostly try to cater to that groups because..well money.
But don't worry. I think there are others here that agree with your point of view and I think some might even come up with a mod that might takle the Ai.
Personaly I find the Ai not shooting a AFK player not the biggest problem.
Things I would like to see more is something that Konivig suggested, an Ai Battleplanner that coordinates the single units but with a wider perspective of the battle.
Also tanks and VTOLs to use formations to concentrate fire. That way they would become a bigger threat as well as act more belivable.
Problems with building need to be adressed like the Ai not beeing able to destinguish if a bulding is still there or if a player hides inside one.
Konivig has made a lot more observations and I hope he will be able to adress some of them.
In the mean time the Ai is good enough for some entertainment and bridgeing the gap will someone comes up with some mods and hopefully better Ai.
I have beta tested several games where the AI has been toned down in the final release. Turns out single-player games players do not want AI that is really good, and those players that do, they go play online multiplayer games instead.
Really good AI (in the tactical sense) is not something that is worth investing in for offline single player games. They just need for the AI not to do things like getting stuck, fall off ledges, go into a loop, or not do anything. Tactically smart AI is not something many players care about or even welcome.
I have also found that the enemies you engage repeatedly are likely to pursue you in return, and the ones you haven't engaged don't come after you until you do shoot at them. Thus.. the AI is reacting to what you do and what your lance mates are doing (whether they're co-op humans or not).
This is true, they are reacting to what you do and there are specific things that you can do in order to change their behavior.
The issue isn't so much how they react within reason (though honestly if you're there to fight them and by ignoring them they'll basically leave you alone...that's not realistic, it should be harder to trigger them losing interest than simply ignoring them), but the fact that they have no concept of each other's existence and see each other as objects that can and should be destroyed if in between you and them.
What it should be: "My friend is in the way of the enemy. Maybe I shouldn't shoot."
What it is: "There's a weird moving object in my way, SHOOT IT!!! SHOOT IT DEAD! RUN THROUGH IT!!! KILL KILL KILL!!!! Oh it's out of the way now, good." / "It's destroyed, good."
Then there's friendly fire with allies.
Your allies have the same notion about your existence as an obstacle to THEIR targets, if you're in the way they will shoot right through you making YOUR ALLIES a bigger threat to you than the ENEMY.
And if you friendly fire them... If you're lucky you'll get a warning about it when they are almost dead.
And nobody gives a single damn if you kill your entire team.
These are very simple things to have not happen but they were completely overlooked.
And while in games like HBS BT, the fact that the enemy mechs have no concept of each other and instead each work to the benefit of itself and that works out... in a first person game with long time to kill compared to most shooters the AI has to at least be reasonably passable.
At present its an autoaimbot instructed with RNG for miss and for body part selection as well as low engagement/pursuit ranges if not attacked, all slapped together with the ability to follow a navmesh, and a few special circumstances for how it moves when in combat or its distance to target preference.
To be honest, the AI in MW2 is basically the same with the exception that in MW2 it was more aggressive once aggro'd and a fair bit more direct. But we're decades apart, the AI Behavioral Tree is so easy a 12 year old can work with it, and to be honest we're reliant on this AI to provide us with the gameplay as most of the game is fighting in generated sandboxes so it was important.
This said I'll probably have it significantly improved in a couple of weeks. (The primary issue of not recognizing each other can likely be solved in under two days with most of that dedicated to testing)
The hardest element would be either figuring out how to use a behavioral tree to run an "HQ" AI or plugging in a Finite State Machine to manage the individual AIs with an overall commander. Afterall, having a dozen or so smart enemies in the field isn't really practical for lower-end machines, but a commander AI controlling them like a real time strategy player is far more resource efficient and can create the illusion of greater intelligence, as well as present expanded mission opportunities like Mechwarrior 3050 for the SNES where destroying the radar system would stop the AI from knowing exactly where you are, or destroying the communication system would cut off the enemies from their HQ, and the addition of "Command" mechs that could also serve as the embodiment of that functionality (as such destroying those things can disable the command AI orchestrating them or disable many of its abilities to simulate the teamwork then resorting to radio communication as opposed to being directed by a MechCommander.)
Adding that might take a couple of dedicated months. Less with help.