BOS trickery isn't particularly special. Damage modifiers are useful, but a unit that features them is still going to have a higher damage to health ratio than a unit without them.
But yes, I can see where an AI would have some trouble with certain weapons from fibre, such as the turret cannon. But they aren't insurmountable for a malleable AI. Adding a new unit would require a rework of the AI for a configured AI.
Hmm... turret cannon. I wonder if there's a way to make the AI's read a much larger build radius for certain units? DT Bombers, or Turret Cannons being prime examples.
Mind you, nothing is as fun as the Vulcan Tooth cannon.
ThreatValue Multipliers
Moderators: hoijui, Moderators
Re: ThreatValue Multipliers
I disagree triaxx for the following reasons, some of which you fo all people should be familiar with:
stargate units
transportable weapon modules on tanks
cob/bos based shields
OTA desync cob scripts for spoofing fake buildings
any kind of fake building at all
bridge units
teleport beacons
And we havent even left the original TA engine, some of these units have been running for years, and there're plenty more. Zwzsg should have a youtube video of them in spring and a few hundred megabytes of animated GIFs.
Heck even mine layers and transports are fiddly because they rarely fit into the TA game play schema despite even being in the game!
stargate units
transportable weapon modules on tanks
cob/bos based shields
OTA desync cob scripts for spoofing fake buildings
any kind of fake building at all
bridge units
teleport beacons
And we havent even left the original TA engine, some of these units have been running for years, and there're plenty more. Zwzsg should have a youtube video of them in spring and a few hundred megabytes of animated GIFs.
Heck even mine layers and transports are fiddly because they rarely fit into the TA game play schema despite even being in the game!
Re: ThreatValue Multipliers
What I mean, is that the AI has no way to guess from the FBI if the bos will activate a damage multiplier and lock all weapons most of time (turning an apparently powerful unit into a weak one) or the contrary, nor what wicked conditions trigger it.Triaxx2 wrote:BOS trickery isn't particularly special. Damage modifiers are useful, but a unit that features them is still going to have a higher damage to health ratio than a unit without them.
Or maybe your AI is so advanced it decodes and analyses cob?
Re: ThreatValue Multipliers
Like the artillery that only fires when not moving? No, I don't know how to reach in and decode that. It's not impossible, because programs can do it, such as Descriptor. But it's not viable because of performance issues.
I do catch your point though.
I do catch your point though.
Re: ThreatValue Multipliers
The problem is not decoding COB into BOS, anway it would be much easier for the AI code to read the compiled script code. The problem is understanding the high level purpose of a program. Kinda like the difference between understanding words one by one, and understanding a story.
The only way around that would be to run loads of test game and see how the unit perform. Save that:
- Some unit have specific purpose, and only peform well used properly. Like Argh's assault shell: it has such a high autoheal rate that it's awesome for hit and run, but poor for prolonged assault. Ok, I'm sure you already have covered that, that your AI already read the autoheal from the FBI to use it appropriatly, no doubt about that.
- Some units purpose is not to kill. During the thousands games I played, never a solar scored any kill. They must underpowered good-for-nothing units then! Even mexxes are infinitely better! Similarly, how do you measure the efficiency of radar, shields, etc??
- For SWTA I coded a little script that makes infantry get armour bonus when around jedi. It's not related to any FBI tag and isn't at all a standard behavior. How would your AI understand that? Would it just understand that from reading the BOS (lol), or store the efficiency of each possible unit combination separately to learn what's the best combo (lol)?
The only way around that would be to run loads of test game and see how the unit perform. Save that:
- Some unit have specific purpose, and only peform well used properly. Like Argh's assault shell: it has such a high autoheal rate that it's awesome for hit and run, but poor for prolonged assault. Ok, I'm sure you already have covered that, that your AI already read the autoheal from the FBI to use it appropriatly, no doubt about that.
- Some units purpose is not to kill. During the thousands games I played, never a solar scored any kill. They must underpowered good-for-nothing units then! Even mexxes are infinitely better! Similarly, how do you measure the efficiency of radar, shields, etc??
- For SWTA I coded a little script that makes infantry get armour bonus when around jedi. It's not related to any FBI tag and isn't at all a standard behavior. How would your AI understand that? Would it just understand that from reading the BOS (lol), or store the efficiency of each possible unit combination separately to learn what's the best combo (lol)?
Re: ThreatValue Multipliers
QFT. Which is what I've been saying about the general "understanding" of complexity within RTS game designs, and complex game designs in general, since the beginning.The problem is understanding the high level purpose of a program. Kinda like the difference between understanding words one by one, and understanding a story.
Re: ThreatValue Multipliers
Resources like the Solar are treated differently than combat units. Combat units cause threat. Resource units only have value.
I'm not sure about Jedi. The AI would simply build them because they increase threat level. If it had a hard code that they were leaders, thus having special abilities, it would probably alter the ratio to 1 to X units, with X being the number of units that fit in the radius.
As for it being easier to read as a Cob, that's fine, one would just have to know which coding to lay down to extract the information. Unless every unit is doing it, it seems more effort than it's worth. It'd be a fun thing to add after the rest of the AI works though.
I'm not sure about Jedi. The AI would simply build them because they increase threat level. If it had a hard code that they were leaders, thus having special abilities, it would probably alter the ratio to 1 to X units, with X being the number of units that fit in the radius.
As for it being easier to read as a Cob, that's fine, one would just have to know which coding to lay down to extract the information. Unless every unit is doing it, it seems more effort than it's worth. It'd be a fun thing to add after the rest of the AI works though.
Re: ThreatValue Multipliers
So, how does the code treat an armored mine that can go from armed-but-not-armored to armored-but-not-armed via Activate()?Resources like the Solar are treated differently than combat units. Combat units cause threat. Resource units only have value.
Heh. PURE's next version is going to seriously screw with your design philosophy. Think outside the OTA box, man.
Re: ThreatValue Multipliers
Doesn't change. It still generates threat. Generally, it would treat it the same as the Annihilator, or the pop-up guns.
Re: ThreatValue Multipliers
Threat is a subjective context that you're treating like a reliable measurable factor.
Sometimes threats can be menacing to one person and irrelevant to another. By your design your AI would take the uber mega gigantica mech as a huge threat and attack it, whereas I would ignore it because it would take hours for my base to enter weapons range of the thing.
Or perhaps the big armada of krogoths is a threat? Or maybe I can ignore it because the armada belongs to two sets of enemies that're currently using them to destroy each other resulting in the two cancelling out.
Simply assigning a value of threat to a unit will not work because threat itself is highly subjective, as is the very definition of threat itself.
Your model is incomplete and cannot possibly encompass every single situation, and it would be folly to attempt to claim so. At some point your AI will need to be told what to do either via you the programmer in a specialized routine or by the players and modders in directive information.
While in an ideal world you can reason your way out of this, your not the one that needs to do the reasoning, and you need to be able to do it automatically on the end users machine in C++ not on a forum with a collection of neurons.
Sometimes threats can be menacing to one person and irrelevant to another. By your design your AI would take the uber mega gigantica mech as a huge threat and attack it, whereas I would ignore it because it would take hours for my base to enter weapons range of the thing.
Or perhaps the big armada of krogoths is a threat? Or maybe I can ignore it because the armada belongs to two sets of enemies that're currently using them to destroy each other resulting in the two cancelling out.
Simply assigning a value of threat to a unit will not work because threat itself is highly subjective, as is the very definition of threat itself.
Your model is incomplete and cannot possibly encompass every single situation, and it would be folly to attempt to claim so. At some point your AI will need to be told what to do either via you the programmer in a specialized routine or by the players and modders in directive information.
While in an ideal world you can reason your way out of this, your not the one that needs to do the reasoning, and you need to be able to do it automatically on the end users machine in C++ not on a forum with a collection of neurons.
Re: ThreatValue Multipliers
As Spring-based games diverge, the feasibility of a monolithic AI rapidly reduces, imo.zwzsg wrote:The problem is not decoding COB into BOS, anway it would be much easier for the AI code to read the compiled script code. The problem is understanding the high level purpose of a program. Kinda like the difference between understanding words one by one, and understanding a story.
The only way around that would be to run loads of test game and see how the unit perform. Save that:
- Some unit have specific purpose, and only peform well used properly. Like Argh's assault shell: it has such a high autoheal rate that it's awesome for hit and run, but poor for prolonged assault. Ok, I'm sure you already have covered that, that your AI already read the autoheal from the FBI to use it appropriatly, no doubt about that.
- Some units purpose is not to kill. During the thousands games I played, never a solar scored any kill. They must underpowered good-for-nothing units then! Even mexxes are infinitely better! Similarly, how do you measure the efficiency of radar, shields, etc??
- For SWTA I coded a little script that makes infantry get armour bonus when around jedi. It's not related to any FBI tag and isn't at all a standard behavior. How would your AI understand that? Would it just understand that from reading the BOS (lol), or store the efficiency of each possible unit combination separately to learn what's the best combo (lol)?
Re: ThreatValue Multipliers
I couldn't have put it better myself.
Re: ThreatValue Multipliers
I see your point, but I'm building an AI for an RTS. Anything not an RTS is moot to the point.
For the megalith unit, there are three factors. Offensive threat, speed/weapon damage. It might have high damage, but the low speed means it's not a threat offensively. Defensive threat, health/weapon damage. High health, powerful weapons? Very high defensive threat, needs softening to lower the threat. Value is Producer? Cost? If it is a producer, cost doesn't matter. If not, it's the average of the three factors of cost. Low offensive threat, high defensive threat, high cost, attack with artillery.
The Krogoth Armada's will likely be treated the same. The AI'll direct it's attack forces towards the lower threat, equally high value production that exists towards the actual bases.
For the megalith unit, there are three factors. Offensive threat, speed/weapon damage. It might have high damage, but the low speed means it's not a threat offensively. Defensive threat, health/weapon damage. High health, powerful weapons? Very high defensive threat, needs softening to lower the threat. Value is Producer? Cost? If it is a producer, cost doesn't matter. If not, it's the average of the three factors of cost. Low offensive threat, high defensive threat, high cost, attack with artillery.
The Krogoth Armada's will likely be treated the same. The AI'll direct it's attack forces towards the lower threat, equally high value production that exists towards the actual bases.
Re: ThreatValue Multipliers
Did I mention that on the furthest reach of the map is a third faction with a teleport machine that could transport the uber megalith mech into the middle of your base at any time?
There are plenty of hardcore RTS game mechanics that could really foobar an AI.
There are plenty of hardcore RTS game mechanics that could really foobar an AI.
Re: ThreatValue Multipliers
I very like that last sentence. You made it sound like some trivial code that will be relaxing to implement.Triaxx2 wrote:As for it being easier to read as a Cob, that's fine, one would just have to know which coding to lay down to extract the information. Unless every unit is doing it, it seems more effort than it's worth. It'd be a fun thing to add after the rest of the AI works though.
Re: ThreatValue Multipliers
I know, it's easy to be cheerful when you know it's a long way off. Of course, I'd have to figure out how to do it first.
But it has to reach the teleporter first, right? Then the stupid thing is still going to be in range until it gets there.
But it has to reach the teleporter first, right? Then the stupid thing is still going to be in range until it gets there.