DM-Deck16][ improved paths

[sektor2111]

Thanks for images but I'll use my version VERY suitable for high skilled Bots - MBots here.
Then, Redeemer reachable with Jumping or Translocator has a bigger AIRATING than clumsy ripper slow and inaccurate and hitting themselves. I'm not paying for PathNodes so I could add 200 nodes in a few minutes. Bots are doing serpent movements under threat and high skill. DnyAI which I'm using in server makes their moves also less predictable. Imagine that Node moving slow of faster in circle and bot Moving at it. It will not go in straight line - so I don't see a lot of prediction, and hey, even if you have 40 nodes in spot you know where they are about to move - for ripper, and you can release a few grenades for fun in advance. Be serious, it's only a load for Engine without a lot of improvements in high skill. Then, map has some paths which Editor won't do itself, they gain shield fast, Udamage, when I'm reaching in the middle from lower area, they already have "PowerUp-s" loaded, and it's my time to step away because they are not kidding.

To summarize they follow the same shortest routes out of enemy on sight - more nodes are not helping unless you are TOGGLING paths on them for less movement prediction. If not, all that charge has no purpose and then... I'm fine with 500+ reachspecs - but I think I can reduce them even more. I see spots still loaded and paths going to end-points or multiple routes to the same items. If Polge said: minimize Network Nodes number and cover map properly, then this is what I did - and high skilled Bots at 150 ping defeated me - I'm not going to load 6000 paths just for the sake of BotPathing.
Bots3 map has useless reachspecs not registered into navigation, it's a charge without any reason. I reduced them without even to touch any node just doing moves with cleaning and nothing else. I understand to give them some other options but not this way. Also can you see PrunedPaths ? Another load...

Master piece ? Not a master but... a piece. Smarter Bots are pruning routes by themselves and even translocating at random - try predict that... Not a single time they evaded from damage leaving me with eyes at the roof... (because Deck16 doesn't have sky). Another custom DeathmatchPlus does another moves. Then map can be played extended, exactly as it is. If plain DeathPatchPlus is not having such a great DeathMatch that's not a concern for me.

there's nothing worse than leading a small group of bots in a particular direction and finding them going off to pick up stuff) - MBots may not even pick up health when they need it and it is indeed difficult to get them to do so sometimes.

And there is nothing worst than Bot (not MBot_G series) going for an item in a rat-hole unable to recover back the navigation because they are ALWAYS busy with items. MBots here have changes in their deals - only in state roaming. The most of maps have a crap-ton of items in place and there is the bugging spot in X cases. MBot it's just getting over trashed area moving ahead - more or less randomly. Also this Pawn is capable to do a random lock to a heavy to reach route and then it sorts another way, as for bad angled paths perhaps it can manage to not sit there all day long - more other habits are coming in XC_Engine presence. I still might be willing to adjust little things at this Pawn but today... it's time to play.


To repeat then: I don't have a single reason for an overloaded map.

[Buggie]

I do paths for default bots. not for custom ones. Of course custom bots can use different rules, so big network can be useless here.

Let see what default bots do.
On Beta 3:

beta3_1.gif

Notice:
- not try jump to box.
- pickup weapon first.
On Sektor ver:

sektor_1.gif

- try jump to box usually. Sometimes long time.
- pickup ammo first.
- move unnaturally

sektor_2.gif

- try "pickup" not exists ammo. Because no any other path.
Even if ammo not respawn they walk here:

sektor_3.gif

Same as above but from different angle.

[sektor2111]

Even if ammo not respawn they walk here:

Can you see Vials ? Too bad, they have knowledge about them. You can see in M.G.S server what exactly bugs you so bad... They are not moving anywhere out of a reason - try to read FindBestInventoryPath and understand what it does...
I'll change this version with a more loaded one when reachspecs are clean and less than 2000. In Deck16 there is no reason for more, Bot is not Predator for iterating through PathNodes.

- try jump to box usually. Sometimes long time.

That's classic setup (not my setup) - my edit it's not for Classic Level - like I said, in HardCore and Turbo they do not have problems and I don't see funky moves or spots locking them.
Hit me with a Decent loaded map not with Paths for the sake of pathing.

[Buggie]

For understand what happens on gif I add screens with paths.
Beta3:

scr_1604415572.png

Sektor ver:

scr_1604415539.png

As you see on simple network bots move simple. Because no other way. Even if bots try randomize path, there nothing to randomize.
From vial exists only one path. Near pulse. But pickup pulse not an option yet, because jump is more cost from move to ammo.
So bot move to ammo firstly.
Also same thing if bot need move from lower left corner of picture to top.
Exists only one path - via pulse ammo.
Go next. Bot near vial and want to move to central ramp.
What possible path? Again only via pulse ammo. Even if not exists. no any direct way.

What About Beta3? Bot can go almost anywhere not pickup pammo. And if he pickup ammo. Then he want this not because no other way.

Summary: if navigation paths too simple bots too predictable. I do not need know what bot want to do. path only one. I can just make shock combo near pammo and bot died.
Does not matter what bot want - goes to me, pick up ammo, pickup pulse, avoid me or run to central ramp. Path only one for actions.

Of course if bots see enemy, they move less predictable and do not follow path as train on rails. but any case, this happen. So if path only one is weakness and predictability which can be used by human. if you hear vial sound stay at top right on picture, you can just make combo near pammo and bot died.
----
Yep, sektor forget mention about such things as: he built paths for martians XC™ stuff (or custom bots, whatever) with very special setup and so on.

Beta3 built for usual default bots without any special settings. So both version can not be compared at all.
this versions built for different condition and purpose.

[sektor2111]

Show me the code where Bot is "randomize" paths. This is new to me.
If I take in account what you say, it can be OFF-Topic in a way. You have more paths not "improved paths". Improved it's when a path heads to the death or an end-point and you fix it, not adding more and more and unused specs.
You should read first what Polge said about nodes nearby items before. Yeah, sure Polge was predicting XC... in years 1999 . What you do is desperate pathing.
My Edit:
= Not for every difficulty Level;
= Less processing;
= Less junks in map.
Your Edit = for me if it has 1800-2000 reachSpecs else you lost me.
Ah, and yes, ONLY custom Bots because I got tired of those stock morons years ago and their issues. For them you need to fix map, for custom Bots editing map is not always mandatory, they can do more than baboons.

[Buggie]

Just for clarify some things.

1. IDK how exactly bot works. So I write "Even if". This is mean "if they use this, this not work there". Of course if they not use - this not make any sense.

2. "improved paths" means "improved paths". Not "less count PathNodes". Not "more simple navigation network". It is your association with this words. Do not mess all at once.

3. Polge write very simple things just because it is 1999 year and it is common guide. Some things hardly formalize. For example let see this maximas:
- A pedestrian must always cross the road at a pedestrian crossing.
- You cannot go to a red traffic light.
Obviously in real life you can cross road anywhere if you in good mind and met some criteria. For example no car near in 1 km. So you can cross road even on red light.
But this condition very hardly can be formalized, so in road rules used maxima.
Same here. Polge say "smallest network as possible". it is maxima. Because CPU in 1999 is weak. And Unreal too hard for most PC from 1999. So it just common sense here.
You really think he can say "well built network with not produce high load, but make good result"? Too fuzzy determination. Need be clear in guides.
Guide written in 1999 and for 1999 year conditions. Not for blindly follow in 2020 year. It is not Holy Book.
So in conclusion: use your mind for decide why rules exists and for what. Not blindly follow it, just because someone say or because it work in past (conditions can change).

4. This forum called "unreal tournament community", not "personal sektor blog". All my builds not for you, it is for community. If its fit you - great. If not - then not.
I think I clearly explain my position on this. I prefer do not return to it anymore anywhere.

[sektor2111]

If we talk about 1999 and weak CPU-s - show me your amazing UBER CPU in MH-GardenOfDeath with this way of your "pathing" . Good luck with that... Yes, it's 2020, lol.
You simply don't get that we have CONST values, hard-coded things which are causing locks and errors when all goes over those "arrays" disregarding how strong is your CPU, and this is UT99.
I'm not against a Great Pathing Network, I'm against when it comes at NOT RESPECTING UnPath.h rules. Yes, my edited map has poor paths - I WANTED that, it's my practice, but yours is crazy - 7000+ reachSpecs in Deck16 ,

Code: Select all

MaxNumFound: I found 7154 reachspecs, from 0 to 7153.

both of them are working in end, but in servers people are running more things, you don't need to send all CPU Engine power for paths processing. Then no I did not say that this is my blog, so let's get rid of miss-understanding excuses because it's not a mature thing at all. Show me something clean and simple, and let's keep being constructive inside engine instead of going out engine and then bug-reporting our own junks.
As a side effect, I would like to know which programmers are writing apps for Android like you do paths in UT, because I will simple IGNORE them all. These CPU intensive things are the last thing for a device battery powered. This is Not An Optimization in any way and it's not demonstrating anything.
Polge said what he said because he knew what he did, you can stop dreaming at CPU stories because things are not working that way. This was already said in forum X times by smarter people than you and me together + others.

Spoiler

Code: Select all

int APawn::breadthPathFrom(AActor *start, AActor *&bestPath, int bSinglePath, int moveFlags)
{
	guard(APawn::breadthPathFrom);
	//FIXME - perhaps track and bound number of edges to search or max depth?
	ANavigationPoint* currentnode = (ANavigationPoint *)start;
	ANavigationPoint* nextnode;
	ANavigationPoint* BinTree = currentnode;

	int iRadius = (int)CollisionRadius;
	int iHeight = (int)CollisionHeight;
	int p = 0;
	int n = 0;
	int realSplit = 1;
	FReachSpec *spec;
	while ( currentnode )
	{
		if ( currentnode->bEndPoint )
		{
			//debugf("best path is %s", currentnode->GetName());
			((ANavigationPoint *)start)->previousPath = NULL;
			bestPath = currentnode;
			return 1;
		}
		if ( (!currentnode->bPlayerOnly || bIsPlayer) || (currentnode == start) )
		{
			int i = 0;
			while ( i<16 )
			{
				if (currentnode->upstreamPaths[i] == -1)
					i = 16;
				else
				{
					spec = &GetLevel()->ReachSpecs(currentnode->upstreamPaths[i]);
					//debugf("check path from %s to %s with %d, %d",spec->Start->GetName(), spec->End->GetName(), spec->CollisionRadius, spec->CollisionHeight);
					if (spec->supports(iRadius, iHeight, moveFlags))
					{
						ANavigationPoint* startnode = (ANavigationPoint* )spec->Start;
						int nextweight = spec->distance + startnode->cost;
						int newVisit = nextweight + currentnode->visitedWeight + startnode->bEndPoint * startnode->bestPathWeight; 
						//debugf("Path from %s to %s costs %d total %d versus %d", spec->Start->GetName(), spec->End->GetName(), nextweight, newVisit, startnode->visitedWeight);
						if ( startnode->visitedWeight > newVisit )
						{
							if ( startnode->prevOrdered ) //remove from old position
							{
								startnode->prevOrdered->nextOrdered = startnode->nextOrdered;
								if (startnode->nextOrdered)
									startnode->nextOrdered->prevOrdered = startnode->prevOrdered;
								if ( BinTree == startnode )
								{
									if ( startnode->prevOrdered->visitedWeight > newVisit )
										BinTree = startnode->prevOrdered;
								}
								else if ( (startnode->visitedWeight > BinTree->visitedWeight)
									&& (newVisit < BinTree->visitedWeight) )
									realSplit--;	
							}
							else if ( newVisit > BinTree->visitedWeight )
								realSplit++;
							else
								realSplit--;
							//debugf("find spot for %s with BinTree = %s",startnode->GetName(), BinTree->GetName());
							startnode->previousPath = currentnode;
							startnode->visitedWeight = newVisit;
							if (  BinTree->visitedWeight < startnode->visitedWeight )
								nextnode = BinTree;
							else
								nextnode = currentnode;
							//debugf(" start at %s with %d to place %s with %d",nextnode->GetName(),nextnode->visitedWeight, startnode->GetName(), startnode->visitedWeight);
							int numList = 0; //TEMP FIXME
							while ( nextnode->nextOrdered && (nextnode->nextOrdered->visitedWeight < startnode->visitedWeight) )
							{
								numList++;
								if ( numList > 500 )
								{
									debugf( TEXT("Breadth path list overflow from %s"), start->GetName() );
									return 0;
								}
								nextnode = nextnode->nextOrdered;
							}
							if (nextnode->nextOrdered != startnode)
							{
								if (nextnode->nextOrdered)
									nextnode->nextOrdered->prevOrdered = startnode;
								startnode->nextOrdered = nextnode->nextOrdered;
								nextnode->nextOrdered = startnode;
								startnode->prevOrdered = nextnode;
							}
							//if (startnode->nextOrdered)
							//	debugf(" place %s after %s before %s ", startnode->GetName(), nextnode->GetName(), startnode->nextOrdered->GetName());
							//else
							//	debugf(" place %s after %s at end ", startnode->GetName(), nextnode->GetName());
						}
					}
					i++;
				}
			}
			realSplit++;
			int move = (int)(0.5 * realSplit);
			while ( p < move )
			{
				p++;
				if (BinTree->nextOrdered)
					BinTree = BinTree->nextOrdered;
			}
		}
		n++;
		if ( bSinglePath && ( n > 4) )
			return 0;
		if ( n > 1000 )
		{
			debugf(NAME_DevPath, TEXT("1000 navigation nodes searched from %s!"), start->GetName() );
			return 0;
		}
		//debugf("Done with %s",currentnode->GetName());
		currentnode = currentnode->nextOrdered;
		/*
		nextnode = currentnode;
		while (nextnode)
		{
			debugf("Next %s",nextnode->GetName());
			nextnode = nextnode->nextOrdered;
		}
		*/
	}
	//debugf("No path found");
	return 0;
	
	unguard;
}

n > 1000 is the same for ANY CPU in 1999 and 2021 like other LIMITS happily hard-coded - yeah, CPU-s , there is nothing beyond 1000 even on a NASA machine. When map is insanely loaded 1000 Paths boundary is closer and closer. What for ? After 1000 it doesn't show anything. Is this the only limitation in UE1 ? NO, it's not... Past recalls: I'm nominating MH-LandsOfNapali, MH-LostInTimeV3 - another muppets shows for DevPath - see what logs are saying - Bot Support for stronger machines = Myths - thanks that this map is small, lol. Then, that Vortex map in HomeBase room was another demonstration of a nothing usable. Probably it's why people knowing stuff won't say more due to this sort of reaction: My edit is great, yours is poor, when actually you are far from doing things as should - inside arrays, and of course I'm not pleasing anybody for using my edits - that's my point of view - I think you have noticed Higor's reaction at that Vortex room which I mentioned... Else if here paths are toggled using some mutator, probably engine will hit another boundary with that "500" limitation - even Barbie knows it, I posted that before... exactly. So be happy trying harder to get over limits or closer to them. I don't understand this "art".
Okay, there is no problem, at your next map editing topic I won't bother to read, post or download anything - this way is not for me or my devs, so you don't have to be upset or angry at me, okay, we are on different grounds here, me one I'm inside Engine, you are trying to escape, good luck with that.

@Barbie - You might want to drop an eye at that 227. Perhaps you'll find some entertaining stuff for use... ?

EDIT: If it comes to images showing Evil stuff by sektor, let's see the images from the side with amazing optimization for Stronger CPUs a la 2020.
Improved paths, huh ? Such things are always looping, it's a lucky shot if these are not used due to other shortest routes, lol, or else only in a combat situation they leave these:

Loop0.PNG

Loop1.PNG

PreviousPath found in two nearby nodes it's not exactly a need, not mention similar NextNavigationpoint and PreviousPath which takes place in spots dedicated for "natural movement".
Why are these good ? For showing "How Not To Do".

But interestingly, they began to arrange mass battle in unexpected places.

Yeah, one of them loops in spot and not leaving, next one comes around, and next one. OMG, whatta fight . After 10 games you will know where to drop a combo shock killing 3 at once. It's why Bot should do some pruning to RouteCache 2, 3, or whatever - plain Bot won't do this.
Another time in similar places from various other maps I'll show you when they run in circle ignoring each-other and ignoring all your CPU fantasies, these are logic reactions of these devs and it's why Polge told what he told - paths are mapped through multiple directions and there are chances to have duplicated data in certain spots - see images, I'm not talking crap, it's not only about CPU, it's a logical processing method going wrong due to overcrowding spots. I hope tomorrow you won't try to put two wheels in the same place of your car...

[f7r]

This thread is more interesting than the Netflix series I'm currently watching. You have opened my eyes to many UT things. For example, I didn’t know that Classic-Hardcore-Turbo affects bot behavior. I have always played Hardcore, 100% speed and 35% aircontrol and recently with XC_Engine.

[sektor2111]

Let me tell you something relevant.
You can start a CTF session with map CTF-Coret, no Bots. Use Classic Level. Try to jump directly in Flak Area.
Now Start the same map alone in HardCore. See how goes jumping.
Okay, what's the deal ? The deal is that in Deck16 in years 2015 sektor wanted Bots jumping on a Box for weapon - Exactly Mr. Buggie - I WANTED that on purpose if you look how I set those combos, yeah, I used combos. What was first try ? Using Combos a la CTF-Coret. The fact is in HardCore and Turbo, they don't need impact Jumping or translocator because it looks dumb. They can jump directly and then... using stuff doesn't look good. In other hand, games played as arena out of ImpactHammer and Translocator will discard paths through these "special" jumpy routes. In that time I took a decision: NO CLASSIC, there isn't any challenge in limited jumping and dodging in ages where DoubleJump is the common Piece. And then... if Deck16 version from sektor shows Bots always on a box with weapon it's because I wanted that for training myself toward whatever jumps required in maps - a MH sample: MH-Freeky... whatever suffix. There jumping is mandatory or else they won't finish map. X times Buggie said about boxes where sektor's bots are always jumping. So what ? Move unnaturally ? Sure, those loops in his edits due to overcrowding spots are "natural" and others without incoming paths - never used - I could add 3 nodes here and there and connecting them post-Pathing, causing reachable buggers for making them to gain weapons by not jumping on any Box, but I did not wanted that, as long as for me it looks pretty good - okay, not perfect, but definitely Bot class has way more troubles than a custom one.
Story with CPU was one problem in the past, there is another one which these users with desperate pathing don't seem to understand - logic of mapping internal paths chain not seen normally in Editor - that's why Polge said what he said - and paths searching number is limited by a hard-coded condition - pretty stupid for an Engine which was claiming: The most Powerful CPU used = The better for UT -> false, hard-coded limitations written directly, 32 bits access = LOGIC LIMITATIONS, they were hypocrites at this point launching fake stories - strong machine loads files faster, has faster graphics but accessing the same memory size and the same navigation bugs, the same for everyone - it's Engine. I wrote a mutator pointing paths for human player as for Bots for understanding their navigation, Deck16 it's not a bright thing at this point. Why ground paths at UDamage ? Because JumpSpot is discarding other Pawns than Bots which are about to jump for it making map restricted at Bot class when Navigation stuff as purpose is addressing Pawn generally - a living creature in Unreal. I don't like this original practice, "Only-Bot", as I don't like automated pathing from Editor making user to test X times map and split hair, when things are cleaned up with a few moves in Editor before doing a final set of tests in Game. I'm not expecting to see more mappers understanding my points as long as some of them are living in their bubble and not even trying to learn simple things. Unreal Pathing is not that hard, it has a bit of logic and I think it's even funny when you can guide pawn as you like.
Instead of adding some blind WarpZones for connections or Teleporters for spoofing connections, and limiting creatures at using paths I'd rather prefer less actors and paths for everyone - PathNodes more or less manually connected and bad paths with lousy angles deleted, my Network doesn't have any loop, all that internal chain (I think it was added later in UT not in Unreal) I deleted, they did not finish that work, I really believe that's a beta released as a final stage of these devs including "//fixme" - Aiming who ? They never updated anything later.

If I'm looking at those codes, I think we can have paths togglers in XXI century without to deploy that spam message Breadth Paths list overflow and blocking pawn - using a Small Number of paths and deleting internal chain and/or calling some C++ helper which Pawn has - that is what I need to check but I need to write some tool first. This is how I see action, a static navigation network offers enough prediction and an overloaded one has default bugs as well. To be honest I got tired of the same bugs all these years and then I might look a bit abrasive in my posts, looking at the same goofing over and over. If we talk about CPUs from 2020 then let's truly step in 2020 with more dynamic stuff and not the same as it was in 1999.

[Buggie]

This thread is more interesting than the Netflix series I'm currently watching. You have opened my eyes to many UT things. For example, I didn’t know that Classic-Hardcore-Turbo affects bot behavior. I have always played Hardcore, 100% speed and 35% aircontrol and recently with XC_Engine.

I even notice such thing: if slowmo > 1.0 then bot fall to acid when try to jump from down to central ramp. If slomo <= 1.0 then all fine. Maybe round error.

[Buggie]

Beta4:

DM-Deck16][bots4.7z

- more ways to armor.
- more ways to medboxes.
- some improvement.

[Que]

Beta4: DM-Deck16][bots4.7z
- more ways to armor.
- more ways to medboxes.
- some improvement.

NewNet compatible @Buggie?
any Teleporter hacks in this version?

[Buggie]

Started from Beta3 all beta NewNet compatible.

Teleporter hack not used. Used WarpZoneInfo hack.

[sektor2111]

In previous version 598_NN I forgot to re-link Navigation chain after adding ground nodes for redeemer, they work but...
Mmm... nope - I don't want this way... if teleporter is disconnected, ground nodes are heading at redeemer even a pupae if it wants redeemer... and Bot won't attempt any movement at Teleporter - this one gets turned into a blind node like NewNet does. I gotta test more places before throwing it in a public server.

Game summary:
HardCore + 8 Adept Bots (not MBots) = Flawless game, normal movement, jumping and all that kick-ass stuff which means I won't listen any complaining.

Edit: When it's about releasing it will go in a different thread letting Buggie to work as he wants his edits.
Edit2: We can have a topic not with Bot Pathing requests but maybe... Bot Pathing for NewNet (stock teleporters tunes) - in other hand NewNet might stop messing up with ALL Teleporters because that's not how to make mods by doing blind replacements, that's stupid, seriously... it ruins a lot of Woot-Type maps, it's pretty jerky.

[Ubir4]

MORE MAPS WITH THIS CONFIGURATIONS...