Building a better RTS AI part 3. APM [edit]
-
A lot of ideas in this post, just like those posted in other places but on the same theme, are half backed. The only thing that will work is what I wrote in the edit at the very end. In the last one or two years I wrote a lot of stuff, things about mirrors and other useless and maybe bad stuff, it was my understanding at the time. What I wrote does have relevant information too. Have fun reading. APM In RTS esports the term APM means actions per minute, the term is being used to measure the number of actions a player is executing while playing. The type of orders that are counted as “actions” are selecting, Issuing attack orders, issuing move orders, issuing build orders. Very good Starcraft players used to have 200 APM or more when Starcraft was a trendy game. When a player has a high APM his gameplay method is being labeled with the term “micromanagement”. High APM is not just a skill, it’s a goal. In a esport RTS you always have to be doing something. You have to either assemble a rush and try to keep your enemy busy defending himself making him unable to expand or you should be using your APM elsewhere like explore the map, hunt creeps, build expansions etc. In other words you should be always doing something. Freedom of will The thing at stake is to make AI micromanage units. If you have a base made of a command center (we’re talking Starcraft now), six SCV’s, three buildings for training units and six marines, That’s 16 entities total that may have their status changed every second. Every second you can leave your units doing what they did a second before or you can make them do something new. 16 decisions you’re taking every second because omission counts as a decision too. If you treat the marines like a group that’s a single decision instead of six, same thing could be applied to the SCV’s. Now you’re left with 6 decisions per time step. Building the prediction tree When I said decisions I meant not a single decision but rather deciding about a unit or building. When a unit of yours is chased by the enemy to get away you can send it in up to 8 different directions, that means you have 8 options to consider ( OLA Option Level A) for each of the 8 OLA the chaser unit has 8 ways to counter (OLAC) we have 64 OLAC total until now. To each OLAC 8 ways to counter by chased, that’s 64 * 8 OLB, etc. The problem here is that the tree of possibilities grows fast on the horizontal. To solve this you may use possibilities in high resolution for lower level branches ( OLA, OLB 8 leaves per branch
-
A lot of ideas in this post, just like those posted in other places but on the same theme, are half backed. The only thing that will work is what I wrote in the edit at the very end. In the last one or two years I wrote a lot of stuff, things about mirrors and other useless and maybe bad stuff, it was my understanding at the time. What I wrote does have relevant information too. Have fun reading. APM In RTS esports the term APM means actions per minute, the term is being used to measure the number of actions a player is executing while playing. The type of orders that are counted as “actions” are selecting, Issuing attack orders, issuing move orders, issuing build orders. Very good Starcraft players used to have 200 APM or more when Starcraft was a trendy game. When a player has a high APM his gameplay method is being labeled with the term “micromanagement”. High APM is not just a skill, it’s a goal. In a esport RTS you always have to be doing something. You have to either assemble a rush and try to keep your enemy busy defending himself making him unable to expand or you should be using your APM elsewhere like explore the map, hunt creeps, build expansions etc. In other words you should be always doing something. Freedom of will The thing at stake is to make AI micromanage units. If you have a base made of a command center (we’re talking Starcraft now), six SCV’s, three buildings for training units and six marines, That’s 16 entities total that may have their status changed every second. Every second you can leave your units doing what they did a second before or you can make them do something new. 16 decisions you’re taking every second because omission counts as a decision too. If you treat the marines like a group that’s a single decision instead of six, same thing could be applied to the SCV’s. Now you’re left with 6 decisions per time step. Building the prediction tree When I said decisions I meant not a single decision but rather deciding about a unit or building. When a unit of yours is chased by the enemy to get away you can send it in up to 8 different directions, that means you have 8 options to consider ( OLA Option Level A) for each of the 8 OLA the chaser unit has 8 ways to counter (OLAC) we have 64 OLAC total until now. To each OLAC 8 ways to counter by chased, that’s 64 * 8 OLB, etc. The problem here is that the tree of possibilities grows fast on the horizontal. To solve this you may use possibilities in high resolution for lower level branches ( OLA, OLB 8 leaves per branch
There's usually only "one" direction to counter being chased: the opposite direction. (That would be a type of "pattern": avoiding) I hear of some "great RTS" ... except when their troops "flee" in the direction of the enemy.
"Before entering on an understanding, I have meditated for a long time, and have foreseen what might happen. It is not genius which reveals to me suddenly, secretly, what I have to say or to do in a circumstance unexpected by other people; it is reflection, it is meditation." - Napoleon I
-
>There’s usually only “one” direction to counter being chased Depends where the opposite direction is leading to. It could lead to a pocket made from map features or buildings, or a combination of both. Or it could lead to another group of enemy units waiting. If you end up in a place like that the only way to escape is running towards the chaser. Another problem is that you want to be running in the clear Usually you’re moving a group of units not just one, that means your units need lots of space to pass all at once if the chase is taking place in a base. If you don’t have enough space a queue is created and the enemy will catch up with you.
For that matter, there are more than "8" directions. The part included in "avoiding", includes "threat assessment"; which then yields the most obvious line of retreat. "Dispersed" troops do not move in an orderly manner, and based on the type of dispersion, are "out of command" for a particular amount of time. And if the second is too close, they then too get dispersed when the first line breaks. The idea that "they'll have lots of space" is counter to the whole concept maneuvering and how to pass defiles.
"Before entering on an understanding, I have meditated for a long time, and have foreseen what might happen. It is not genius which reveals to me suddenly, secretly, what I have to say or to do in a circumstance unexpected by other people; it is reflection, it is meditation." - Napoleon I
-
There's usually only "one" direction to counter being chased: the opposite direction. (That would be a type of "pattern": avoiding) I hear of some "great RTS" ... except when their troops "flee" in the direction of the enemy.
"Before entering on an understanding, I have meditated for a long time, and have foreseen what might happen. It is not genius which reveals to me suddenly, secretly, what I have to say or to do in a circumstance unexpected by other people; it is reflection, it is meditation." - Napoleon I
>There’s usually only “one” direction to counter being chased Depends where the opposite direction is leading to. It could lead to a pocket made from map features or buildings, or a combination of both. Or it could lead to another group of enemy units waiting. If you end up in a place like that the only way to escape is running towards the chaser. Another problem is that you want to be running in the clear Usually you’re moving a group of units not just one, that means your units need lots of space to pass all at once if the chase is taking place in a base. If you don’t have enough space a queue is created and the enemy will catch up with you.
-
For that matter, there are more than "8" directions. The part included in "avoiding", includes "threat assessment"; which then yields the most obvious line of retreat. "Dispersed" troops do not move in an orderly manner, and based on the type of dispersion, are "out of command" for a particular amount of time. And if the second is too close, they then too get dispersed when the first line breaks. The idea that "they'll have lots of space" is counter to the whole concept maneuvering and how to pass defiles.
"Before entering on an understanding, I have meditated for a long time, and have foreseen what might happen. It is not genius which reveals to me suddenly, secretly, what I have to say or to do in a circumstance unexpected by other people; it is reflection, it is meditation." - Napoleon I
If you take the movement on two or three consecutive tiles and make an average you may get a direction that is none of the 8. I don’t think the number of directions is a stake with great value here. >The part included in avoiding includes Figuring out the behavior of the someone being chased is not as difficult as figuring out the behavior of someone trying to chase something into a corner. On the other hand the chased might need the mind of the chaser to understand the danger spots.
-
A lot of ideas in this post, just like those posted in other places but on the same theme, are half backed. The only thing that will work is what I wrote in the edit at the very end. In the last one or two years I wrote a lot of stuff, things about mirrors and other useless and maybe bad stuff, it was my understanding at the time. What I wrote does have relevant information too. Have fun reading. APM In RTS esports the term APM means actions per minute, the term is being used to measure the number of actions a player is executing while playing. The type of orders that are counted as “actions” are selecting, Issuing attack orders, issuing move orders, issuing build orders. Very good Starcraft players used to have 200 APM or more when Starcraft was a trendy game. When a player has a high APM his gameplay method is being labeled with the term “micromanagement”. High APM is not just a skill, it’s a goal. In a esport RTS you always have to be doing something. You have to either assemble a rush and try to keep your enemy busy defending himself making him unable to expand or you should be using your APM elsewhere like explore the map, hunt creeps, build expansions etc. In other words you should be always doing something. Freedom of will The thing at stake is to make AI micromanage units. If you have a base made of a command center (we’re talking Starcraft now), six SCV’s, three buildings for training units and six marines, That’s 16 entities total that may have their status changed every second. Every second you can leave your units doing what they did a second before or you can make them do something new. 16 decisions you’re taking every second because omission counts as a decision too. If you treat the marines like a group that’s a single decision instead of six, same thing could be applied to the SCV’s. Now you’re left with 6 decisions per time step. Building the prediction tree When I said decisions I meant not a single decision but rather deciding about a unit or building. When a unit of yours is chased by the enemy to get away you can send it in up to 8 different directions, that means you have 8 options to consider ( OLA Option Level A) for each of the 8 OLA the chaser unit has 8 ways to counter (OLAC) we have 64 OLAC total until now. To each OLAC 8 ways to counter by chased, that’s 64 * 8 OLB, etc. The problem here is that the tree of possibilities grows fast on the horizontal. To solve this you may use possibilities in high resolution for lower level branches ( OLA, OLB 8 leaves per branch
this is just to draw your attention, the edit is at the end of the original post