Plan for Shayne Guiliano
by Shayne Guiliano · 09/25/2002 (4:13 pm) · 0 comments
Well, things have been going well over the past few months.
I've started learning C++ and have continued working with Steve on Odin and the Tranquility design. That stuff is going slow since we've both decided to teach ourselves C++, so we can have the power to implement ourselves. The first thing we'll do is try to get the tranquility working.
I've spent a lot of my energy the past two months working on Coaching Youth Baseball. I finished the first draft of the design doc (about 40 pages now) and invented a cool method for designing, communicating and debugging a state machine no matter the complexity, which gives me confidence that the state machine will work out very well. I'm gonna write up a paper for gamasutra and try to get it into AI programming wisdom as well. I don't think there's enough good design methods that are public knowledge so hopefully it might help someone out there who's struggling with complex state machine design.
Chris P, my programmer for Coaching Youth Baseball, has begun the core. He estimates 2-4 months before completion of the basics like throwing the ball and stuff like that, which is good since I think it will take about that long before I've finished the state machine and design doc.
I've been mentally swimming in the algorithms that define the rules that will make the agents act like they would in real youth baseball, and have a good grasp on most of them. I have reached a point where there doesn't seem to be any gaping holes in the design logic, which is something I couldn't say for the first 8 months of designing it. The agents will not only improve their skills based on the practice structure and game experience, but also learn the strategy of the coach via a couple cool learning algorithms I've worked out. e.g. with men on 1st and third, who charges on a bunt?
An agent will develop from the age of six all the way to 12 just like an agent in real life will. I can say this with surity because I will be doing research with two other very knowledgable baseball guys in Florida, my father and my best friend, to determine the talent spectrum for all age groups and experience. We will gather stats on the accuracy of players arms and bats, the number of frames it takes a player to inact an action from frame-by-frame analysis, how quickly they learn their skills, how drills affect their development, etc., etc. The variables and learning curves derived from research will really be what pulls the entire game together in the end.
A six year old player will have a very low probability of hitting a 12 year old, and there is a difference between the way different players learn different parts of the strike zone, catching zone, and pitching zone.
The players also have different personalities, they go into streaks and slumps, get affected by high pressure situations, and have moods. They respond to discipline based on their personality traits, mood, pressure, etc. as well. Not only that, but every skill for every player develops uniquely because of the way that I have designed the learning engine, so that a coach must get to know and understand how each player learns in order to get the most out of them. This means that a player learns inside fastballs differently than he does outside fastballs, which is what happens in real life.
Anyways, there's lots more to say about the development of the learning AI, but I don't want to spoil it nor bore you anymore. I'll wait till it's clicking along at millions of calcs per sec. Then the agents will come to life and show you what they can do.
The only issue we haven't started tackling is art resources, though Chris has designed a program that will allow us to use the research footage for animating the models, which will help immensely.
Overall, I'm very excited about the project despite the obvious mountain of work remaining. I've realized over the development of the design that we are breaking new ground with this one. I hope I am capable of making it work well despite there being no precedent for it, but I feel more confident about it every day.
Anyways, it's exciting to be a part of something that might be truly unique and fun at the same time. I also think it has great potential for success, especially if we can tie up the Little League license. Little league has already agreed to a meeting when I have a workable demo. Since there's one million volunteers just in Little League, not even counting all the volunteers from other major youth leagues, it seems like there should be a good market to sell to. This isn't even counting the kids, real-time RTS fans, and non-volunteer baseball fans.
If anyone wants to know more about the game, please contact me. We are looking for some artists to get in on it over the next few months.
I've started learning C++ and have continued working with Steve on Odin and the Tranquility design. That stuff is going slow since we've both decided to teach ourselves C++, so we can have the power to implement ourselves. The first thing we'll do is try to get the tranquility working.
I've spent a lot of my energy the past two months working on Coaching Youth Baseball. I finished the first draft of the design doc (about 40 pages now) and invented a cool method for designing, communicating and debugging a state machine no matter the complexity, which gives me confidence that the state machine will work out very well. I'm gonna write up a paper for gamasutra and try to get it into AI programming wisdom as well. I don't think there's enough good design methods that are public knowledge so hopefully it might help someone out there who's struggling with complex state machine design.
Chris P, my programmer for Coaching Youth Baseball, has begun the core. He estimates 2-4 months before completion of the basics like throwing the ball and stuff like that, which is good since I think it will take about that long before I've finished the state machine and design doc.
I've been mentally swimming in the algorithms that define the rules that will make the agents act like they would in real youth baseball, and have a good grasp on most of them. I have reached a point where there doesn't seem to be any gaping holes in the design logic, which is something I couldn't say for the first 8 months of designing it. The agents will not only improve their skills based on the practice structure and game experience, but also learn the strategy of the coach via a couple cool learning algorithms I've worked out. e.g. with men on 1st and third, who charges on a bunt?
An agent will develop from the age of six all the way to 12 just like an agent in real life will. I can say this with surity because I will be doing research with two other very knowledgable baseball guys in Florida, my father and my best friend, to determine the talent spectrum for all age groups and experience. We will gather stats on the accuracy of players arms and bats, the number of frames it takes a player to inact an action from frame-by-frame analysis, how quickly they learn their skills, how drills affect their development, etc., etc. The variables and learning curves derived from research will really be what pulls the entire game together in the end.
A six year old player will have a very low probability of hitting a 12 year old, and there is a difference between the way different players learn different parts of the strike zone, catching zone, and pitching zone.
The players also have different personalities, they go into streaks and slumps, get affected by high pressure situations, and have moods. They respond to discipline based on their personality traits, mood, pressure, etc. as well. Not only that, but every skill for every player develops uniquely because of the way that I have designed the learning engine, so that a coach must get to know and understand how each player learns in order to get the most out of them. This means that a player learns inside fastballs differently than he does outside fastballs, which is what happens in real life.
Anyways, there's lots more to say about the development of the learning AI, but I don't want to spoil it nor bore you anymore. I'll wait till it's clicking along at millions of calcs per sec. Then the agents will come to life and show you what they can do.
The only issue we haven't started tackling is art resources, though Chris has designed a program that will allow us to use the research footage for animating the models, which will help immensely.
Overall, I'm very excited about the project despite the obvious mountain of work remaining. I've realized over the development of the design that we are breaking new ground with this one. I hope I am capable of making it work well despite there being no precedent for it, but I feel more confident about it every day.
Anyways, it's exciting to be a part of something that might be truly unique and fun at the same time. I also think it has great potential for success, especially if we can tie up the Little League license. Little league has already agreed to a meeting when I have a workable demo. Since there's one million volunteers just in Little League, not even counting all the volunteers from other major youth leagues, it seems like there should be a good market to sell to. This isn't even counting the kids, real-time RTS fans, and non-volunteer baseball fans.
If anyone wants to know more about the game, please contact me. We are looking for some artists to get in on it over the next few months.
About the author