ai sensors/directions?( for random maps)

[ConsciouS-0nE]

Au right my turn, not even gonna beg just gonna give what yah guys wanted.
(Its not even an official request)

By the way, what's the point of having directional awareness if you can't control the bot's orientation?

Nah with certain things on pre generated maps.its possible to order directions from other entities.


///
Dunno if this would make sense but yeah.


Okay heres the thing;
(*Agh omg hes starting it again!*, shh listen)

Imagine if your blind, apparently gladiabots ais are.

Lets say your in a empty room just with three things (W-X-Y)

On your left theres a CHAIR(W)
On your front theres a TABLE(X)
On your right theres a TV(Y)

You stand on the point called (Z)


Wich is the only movment you can do is go to POINT-POINT.(A→B)


Soooo, if your move forward you will bump to the TABLE(X)

Butttt what if someone told you that there are two more things.


So now the scenario is,
You know that there are things on your left and right.

Itll go like this,

Move forward to the table(X)
BUT someone said theres a chair(W) on my left
So Go to the chair.

So your distination got from,
Z→W
Instead from
Z→X


But thats a PRE-Generated map,

So if a sniper in your left and a mg in your right and your moving toward an enemy,
In certain conditions you can go to the sniper.


Conclusion:
The PLAYER can know if, lets say "the bot is surrounded"
Or something is on the left or right wich can be used to command with, BUT the bot is not aware of it.
orientation is already there, but with human awareness


But what if random maps?
-The player doest know which is which or where
-soo the orientation got removed
-or basically the player got blind and the bot is already blinded and the thing is the sniper will never be ALWAYS in the left because its RANDOM


The thing is, if the bot can know where is where, it go on its own.

*caps on are not shouting, please tell me if theres something you don't understand.

[harthag]

By the way, what's the point of having directional awareness if you can't control the bot's orientation?

I think ConsciouS-0nE did a good job of answering that, but I'm going to give my own example, as an alternative point of view, and alternative descriptive method:

Bot orientation is actually not particularly important in these cases, what's important is the bot 'knowing' what else is on the map, and where. And the exact same information is gained, regardless of the bots own orientation.

Here's the example (I'll use the most commonly requested "retreat route"):
A bot is facing the map's 'north', so north is 'front'. There are 3 enemy bots, an assault at front medium range, a sniper at front+left long range, and a machinegun at back+left medium range. There is an ally bot at front+right long range, and another ally at back+right short range. If retreat becomes needed, where does the bot go? Personally, I'd choose "move toward" anything "east" of the bot, using anything else the bot can "see" in that direction. There are no allys there (in this example) but there might be a resource, a base, a fixed point on a map or map wall (when map references become available), etc.

firing arc.png (9.82 KiB) Viewed 1848 times

Now, rotate the bot's orientation 90 degrees, so let's say the bot is now facing west (why? doesn't matter, but let's say it had been firing at an enemy that retreated out of range) but nothing else about the situation changes (the bots orientation, which the human player can't control, is the only factor here). Now which way does the bot retreat?

firing arc 2.png (9.84 KiB) Viewed 1848 times

Still East. Why? Because that's still the best route. How does the bot know that's the best route? Because it can 'see', the bot itself can make an informed decision, regarding it's surroundings. The AI would have to be good enough to accomplish it, and yes, it would have to be an advanced AI, but I have no doubt there are already many players playing that could, and would, accomplish it.

Again, that was the "retreat route" example, use to answer the specific question about bot orientation, and I've mentioned battle tactics in earlier posts (flanking, formations, etc.), but here's a simple example for resource gathering:

firing arc 3.png (29.19 KiB) Viewed 1848 times

Currently, the bot can't tell the difference between any of those situations, they are all true when checking for resource at short range and base at long range. But with directional awareness, the bot's AI could tell that situation "C" is the best time to grab a resource and run for the base. Which direction the bot is facing at the time makes no difference, the bot can still tell that a resource 'on the way' to the base is a better option than a resource that is not along the same general path.

These types of things are what ConsciouS-0nE and I mean by the 'bot' being able to 'see', even when the human player doesn't know the map layout due to random maps. So the 'bot' has to make (Artifically) Intelligent decisions, not the human player.

[pier4r]

Nice writeups harthag And yes knowing where the enemy is (left, back, right, front, mid, short, long range) would be a huge step. Even bigger than the alpha 7.8 filters.

[ConsciouS-0nE]

this better be a boy or an abortion


Btw
Good one,

But buttttt, gfx is planning to make a small fix for this,
Its just that the thing is too big and by said thats gonna be alot of work because of chained request.

But iff you guys want to finalize this thing, its just okay for me, dunno for gfx tho(i can already see that migrane)
Maybe less prioritized than 'long-term'
Cuz i just showed this thing that came up in my head which is pratical btw, and you guys made it really better.


I was just like,
"Hey guys look scrap metal,"
Then you guys was like,
"Yeaaahhhhhhhj feed meee imma make this a super doomsday machine!"
Then i was like,
"Yep will leave that there, hands up going home."

[GFX47]

OK, it could be useful for cases like the resource / base example you showed.
I can only imagine how complex an AI would have to be if you want to check every possible combination.
Besides, try to picture the fleeing situation where you start with your main enemy at your front left and suddenly becomes your rear right because you started to flee...

Maybe the directions should be absolute (North, South, East, West) instead of relative (front, rear, left, right) to avoid being impacted by your every action? I'm not even sure it simplifies anything XD

[pier4r]

OK, it could be useful for cases like the resource / base example you showed.
I can only imagine how complex an AI would have to be if you want to check every possible combination.
Besides, try to picture the fleeing situation where you start with your main enemy at your front left and suddenly becomes your rear right because you started to flee...

A fleeing bot, IIRC, does not turn so it will stay the same.

If it would turn, then yes. (so the "non turning" approach now it is a feature)

[harthag]

try to picture the fleeing situation where you start with your main enemy at your front left and suddenly becomes your rear right because you started to flee...

If (situation parameters) are met, Bottag1 self then bottag2 destination item X, if self is bottagged1 and if X is not close range move toward X, if X is close range, remove bot tags 1 and 2.

Yes, defining those 'situation parameters' is where it *can* get extremely complex, and I've admitted that complexity issue before, but the execution of actions based on the 'situation' is relatively simple. And being able to define situations is closer to AI than anything the bots can do currently.

A fleeing bot, IIRC, does not turn so it will stay the same.

If it would turn, then yes. (so the "non turning" approach now it is a feature)

If movement directions are added, that makes it even easier to react to these 'situations', because then you can just "move right" or "move left" instead of tagging and following tags. But I actually think those, and bot rotation, are actually separate features from the directional awareness.

[pier4r]

this better be a boy or an abortion

????

[ConsciouS-0nE]

????

Its not invisible?
Some classic dark Valentine joke.


////

Here's the example (I'll use the most commonly requested "retreat route"):
A bot is facing the map's 'north', so north is 'front'. There are 3 enemy bots, an assault at front medium range, a sniper at front+left long range, and a machinegun at back+left medium range. There is an ally bot at front+right long range, and another ally at back+right short range. If retreat becomes needed, where does the bot go? Personally, I'd choose "move toward" anything "east" of the bot, using anything else the bot can "see" in that direction. There are no allys there (in this example) but there might be a resource, a base, a fixed point on a map or map wall (when map references become available), etc.

firing arc.png

All i imagine this as


Thats not how the game retreats work,
Remember the bot can do 2 actions
Which is
Move forward, backwards,
No rotation as they have the same speed right?

I only see automatic 180 rotation when,
-secure resource/move to ally base at the back
-moving to ally that is in the back.
*there is a fix point



As simple as,
"Flee from enemy sniper class"
"If enemy exist [invert] out of range"
"If self under attack"

The bot will flee to the snipers opposite direction as long as its still under attack by the enemy from the med range.


But but that is a pre generated map
And i already know that situation, the bot doesnt
So ill program the counter in advance.

What if random?, dead.
/////

OK, it could be useful for cases like the resource / base example you showed.
I can only imagine how complex an AI would have to be if you want to check every possible combination.
Besides, try to picture the fleeing situation where you start with your main enemy at your front left and suddenly becomes your rear right because you started to flee...

Maybe the directions should be absolute (North, South, East, West) instead of relative (front, rear, left, right) to avoid being impacted by your every action? I'm not even sure it simplifies anything XD

Looking forward for your own modifications in this concept.

[harthag]

I'd choose "move toward" anything "east" of the bot, using anything else the bot can "see" in that direction.... a resource, a base, a fixed point on a map or map wall (when map references become available), etc.

I guess I should clarify "I'd choose..." should read "I'd program the bot to choose..."

Thats not how the game retreats work,
Remember the bot can do 2 actions
Which is
Move forward, backwards,
No rotation as they have the same speed right?

Yes, and because that's not how "flee from" works, currently, that's exactly why "move toward" is what I'd program the bot to do. Because there is no bot rotation speed limitation currently, "move toward" is the best option available in the game currently. If rotation speed is added later, then I might have to adjust the programming for new tactics, since a full turn might take too long and the bot could get killed too fast while turning. Also, if additional movement options are ever added, such as "retreat toward"(move in reverse, but with the 'target' item behind the bot instead of in front of it), or "right-strafe toward"(crab-walk to the right), then I might program the bot to select one of those instead.

I only see automatic 180 rotation when...*there is a fix point

Correct again. And that's why my example specifically says that I'd have the bot select a 'fixed point' that it could 'see', and that was in the desired direction, so that it had something to 'move toward'.

As simple as,
"Flee from enemy sniper class"
"If enemy exist [invert] out of range"
"If self under attack"
The bot will flee to the snipers opposite direction as long as its still under attack by the enemy from the med range.

Make it an assault instead of a sniper, add a fourth enemy, change the distance, etc., and "flee from sniper" could become unclear or impossible options, but once map position points are implemented that option never will be unavailable. If a simpler solution, such as flee from sniper, is better for a particular situation, use it, yes. But the point is that directional awareness makes it possible for the bot to determine *any* situation that the programmer can come up with. What the bot does about the situation is also up to the programmer. But without awareness, we're back to the zigzag retreat because the bot can't tell that it's the sniper it should be moving away from, so we can't create a program to tell the bot to avoid it.

But but that is a pre generated map
And i already know that situation, the bot doesnt
So ill program the counter in advance.

What if random?, dead.

No, my example was assuming a random map, not a pre-generated one. This was meant as an example of a situation that could arise at some point after the match starts, it is not meant to show the map before the battle begins.

Maybe the directions should be absolute (North, South, East, West)

[/quote]
Since fixed points on the map, and map walls, as identifiable 'objects' or 'targets' is already on the roadmap, I've been working under the assumption that those will already be available if awareness is ever implemented.

[pier4r]

If movement directions are added, that makes it even easier to react to these 'situations', because then you can just "move right" or "move left" instead of tagging and following tags. But I actually think those, and bot rotation, are actually separate features from the directional awareness.

Why tags? When you retreat or push the bot goes straight from the object you are retreating/moving to. So I don't see the need of tags.

[harthag]

If movement directions are added, that makes it even easier to react to these 'situations', because then you can just "move right" or "move left" instead of tagging and following tags. But I actually think those, and bot rotation, are actually separate features from the directional awareness.

Why tags?

Because, as GFX mentioned, without them the bot's rotation would change the orientation and that would change the 'situation' the bot perceives, so it would get part of the way through the rotation and then see a different situation, change it's mind about what to do, most likely rotating back the other way to it's previous objective, and encounter the situation that caused it to flee, and repeat, caught in a partial rotation back and forth.

Besides, try to picture the fleeing situation where you start with your main enemy at your front left and suddenly becomes your rear right because you started to flee...

The "tag self" part is to tell the bot it's in trouble. The "tag destination" is to tell the bot where it needs to be to get out of trouble. so it doesn't stop until it gets there and untags itself to show it's no longer in trouble.

*If* the best route doesn't require any bot rotation, then tags would not be needed. Or *if* the AI is complex enough to take the same situation in to account from all possible bot rotation angles, then tags would also not be needed. Since I'm not a good enough programmer to believe I'm capable of either of those, I'd use tags.

[pier4r]

.

I still don't follow: viewtopic.php?f=2&t=441&start=40#p5157

A bot does not rotate while fleeing.

[harthag]

A bot does not rotate while fleeing.

No, a bot does not rotate while following the existing "flee from" command, *if* it is already pointed at the 'target' that it's fleeing from.

firing arc.png (9.82 KiB) Viewed 1782 times

But in this situation it is facing 'north' so if it does not rotate, then it will have to move in reverse which will take it 'south' instead of the desired direction of 'east'. It must either rotate to face 'west', and then "flee from" something at the west, or it must rotate to the 'east' and "move toward" something in the 'east'. Either way, it will be rotating approximately 90 degrees before moving, and that rotation could change the bots perspective and decision. The tags prevent confusion from the rotation.

ConsciouS-0nE pointed out that in this very specific example, "flee from sniper" would be a viable escape command, and it is, but even that would cause a rotation of about 45 degrees, again opening the possibility of confusing the bot during the rotation, and failing to flee at all.

[pier4r]

First: thanks for the patience to explain to me.

Second: The example is not fitting. In my opinion when one is discussing a possible new feature one has to compare what can happen now and what can happen with the feature.

With the current commands the example that you report can cause a zig zag fleeing from the bot. I retreat from north, but then I have at south west an enemy medium range, and I retreat from that and so on. Zig /Zag.

Instead one has to analyze the example with the possible new commands. So having directional views and ranges I would say, given that rotation is not possible for a bot unless pointing to some point (base/ally/whatever):

- Conditions to detect the actual enemy positions.
- Those condition determines that if the self flee from north, there will be a closing enemy from south west
- The conditions determine that south there is an ally more or less in the same range of the enemy.
- Therefore the decision is not to retreat, but to shoot at the enemy south west, that has chances to be chosen by the ally too since it is near. So it is a tactical advantage.
- then the perspective change because the bot rotated.
- But this time, with the same conditions as before, still retreat is no so good (otherwise I would flee from the guy SW and then ending up north), but I'm already near to my ally so I just attack.

To me, even without bot rotation, having directions and range filters (I like the 8 directions in the image), would be a tactical leap because I can decide way better what to do, given the simple movements of the bots (moving straight to or away from objects).

If there would be bot rotation, then yes, the topic would be way more complicated. But since rotation is absent it is a non problem.

Moreover, while relative directions are cool, would be way easier (I guess also in terms of programming) to have fixed directions as someone said. North, Sud, East, West and combinations. So rotation or not, the problem does not apply because the directions remain equal.

[harthag]

First: thanks for the patience to explain to me.

You're welcome. Sorry I was not more clear originally.

when one is discussing a possible new feature one has to compare what can happen now and what can happen with the feature.

I'm not sure I understand what you mean, so I guess it's my turn to be confused.

Instead one has to analyze the example with the possible new commands. So having directional views and ranges I would say, given that rotation is not possible for a bot unless pointing to some point (base/ally/whatever):

- Conditions to detect the actual enemy positions.
- Those condition determines that if the self flee from north, there will be a closing enemy from south west
- The conditions determine that south there is an ally more or less in the same range of the enemy.
- Therefore the decision is not to retreat, but to shoot at the enemy south west, that has chances to be chosen by the ally too since it is near. So it is a tactical advantage.
- then the perspective change because the bot rotated.
- But this time, with the same conditions as before, still retreat is no so good (otherwise I would flee from the guy SW and then ending up north), but I'm already near to my ally so I just attack.

To me, even without bot rotation, having directions and range filters (I like the 8 directions in the image), would be a tactical leap because I can decide way better what to do, given the simple movements of the bots (moving straight to or away from objects).

Yes, tactics other than 'flee' or 'retreat' would certainly be possible with awareness in place, but the OP of the topic used retreat as a specific example, so my example was based on the assumption that the bot had considered other tactics and decided retreat was the best option, and then continued to decide which direction was best for that retreat. It's that leap in tactics that makes awareness so appealing to me.

while relative directions are cool, would be way easier (I guess also in terms of programming) to have fixed directions as someone said. North, Sud, East, West and combinations. So rotation or not, the problem does not apply because the directions remain equal.

It was GFX that mentioned it.

Maybe the directions should be absolute (North, South, East, West) instead of relative (front, rear, left, right) to avoid being impacted by your every action? I'm not even sure it simplifies anything XD

At the time, I didn't understand what he meant, but this post from you clarified it for me, and I actually like that better than relative to the bot.

[pier4r]

I'm not sure I understand what you mean, so I guess it's my turn to be confused.

My argument is like:

I have the AI controls alpha 7.8 and the possibility to include the new AI controls that we are discussing, say in alpha 8, where alpha 8 is alpha 7.8 plus the new direction filters and that's it (it is not the case of course, I'm using alpha 8 to make the example easier).

So to see if the new filters make sense, I compare a situation with the controls of alpha 7.8 and the controls in alpha 8.

With alpha 7.8 I have limited retreat possibilities and I can end up zig zagging. With alpha 8 I can do way more.

Now I understand that the original example mentions retreat, but if we discuss about a feature, we should use it completely and say "ok, maybe it is not good to retreat, but with those filters at least I can choose to hold my ground in a productive way".

There are a ton of situations where those filters can be useful.

Is the sniper a bit off from the enemy team? For example the sniper is North west long range while most of the enemy team for some reason is Nort East medium and long range. Go taking the sniper away!

Is that resource long range, the closest enemy is long range, but the resource is south and the enemy is north? Go take the resource!

Instead now it is very difficult to determine if a resource long range is in the near of the enemy or not.

[harthag]

Ok, I understand now, but you've also identified the flaw with attempting to make the discussion "complete"

if we discuss about a feature, we should use it completely...There are a ton of situations where those filters can be useful.

There are simply far to many possible situations, and far too many possible reactions. When you consider the possible combinations of situations and reactions, the number is exponentially greater. We have no hope of addressing them all.

I've picked 2, and given examples of them, one possible retreat situation and one possible resource situation. I've also mentioned, briefly, battle tactics as a possibilities: Flanking, formations, etc. but to try and discuss all of them is unrealistic.

In short:

Currently, the player does the analysis in advance.
With directional awareness, the bots can do the analysis in the bot's own 'real time'.

Currently, the bot's AI is based on the players advance knowledge of the battlefield, and their guesses at what the enemy will attempt and how it will make the battle unfold.
With directional awareness, the bots can detect the 'current' battlefield situation, and their AI can decide the best course of action based on actual information, instead of the player's previous guess.

[pier4r]

I've picked 2, and given examples of them, one possible retreat situation and one possible resource situation. I've also mentioned, briefly, battle tactics as a possibilities: Flanking, formations, etc. but to try and discuss all of them is unrealistic.

Sure. My point was more "if we find one example in which the new filters are not so helpful, this does not rule out all the other situations where the new filters can help a lot".

With directional awareness, the bots can detect the 'current' battlefield situation, and their AI can decide the best course of action based on actual information, instead of the player's previous guess.

This.