Hiro's Run
by Chuck Waters · in Game Design and Creative Issues · 09/11/2004 (5:39 am) · 14 replies
Hiro's Run
We at Death By Ninja Games are currently developing a third person shooter based loosely on both Shadowrun and Neil Stephensons Snow Crash. No need to post on the copyright infringing, it's an academically charged project. This is our first run at the Torque Gaming Engine, and so far we are absolutely impressed by it. Torques ability to handle alot of generic things lets us focus on gameplay and our complicated AI.
AI is the biggest part of Hiro's Run. We will be utilizing waypoint node meshes for path finding, perceptrons for accurate AI vision and sight. However, we believe our use of Nueral networks is what make it special. Our AI will have the innate aility to learn from you. If you try to camp and snuff out the enemy, the bots with start manuevering around and try to flank you, if you run out and start shooting without a care, they will lay back and camp you. However you play, you will be able to control the pacing of the game, but the AI will stay at your level, no need to increase stats or make them faster as a difficulty setting.
We have an Asset team working on most everything we need, but if anyone would like to contriute models, graphics or music, you can email me at corylamontagne@yahoo.com
The game is based in Seattle, 2063, in a world controlled by evil corporations. The player will have missions in the real world, and also on the net world for anyone who has read Snow Crash will understand what I mean. These two worlds are completely separate, and will be rendered differently as well has have their own separate physics systems.
The only thing we dont have a hold on is how to change the way things are rendered, as in Cel shaded, regular lighting, etc. Any nudge towards the right tut or site would be greatly appreciated.
Thanks,
The DBNE team
We at Death By Ninja Games are currently developing a third person shooter based loosely on both Shadowrun and Neil Stephensons Snow Crash. No need to post on the copyright infringing, it's an academically charged project. This is our first run at the Torque Gaming Engine, and so far we are absolutely impressed by it. Torques ability to handle alot of generic things lets us focus on gameplay and our complicated AI.
AI is the biggest part of Hiro's Run. We will be utilizing waypoint node meshes for path finding, perceptrons for accurate AI vision and sight. However, we believe our use of Nueral networks is what make it special. Our AI will have the innate aility to learn from you. If you try to camp and snuff out the enemy, the bots with start manuevering around and try to flank you, if you run out and start shooting without a care, they will lay back and camp you. However you play, you will be able to control the pacing of the game, but the AI will stay at your level, no need to increase stats or make them faster as a difficulty setting.
We have an Asset team working on most everything we need, but if anyone would like to contriute models, graphics or music, you can email me at corylamontagne@yahoo.com
The game is based in Seattle, 2063, in a world controlled by evil corporations. The player will have missions in the real world, and also on the net world for anyone who has read Snow Crash will understand what I mean. These two worlds are completely separate, and will be rendered differently as well has have their own separate physics systems.
The only thing we dont have a hold on is how to change the way things are rendered, as in Cel shaded, regular lighting, etc. Any nudge towards the right tut or site would be greatly appreciated.
Thanks,
The DBNE team
About the author
#3
09/20/2004 (11:31 am)
Were using a small simple neural net. were focusing mainly on the Gameplay. Our AI lead is handling it mostly so i dont know the specifics.
#4
I would suggest that your AI lead read up on neural nets before spending the time to implement and later realizing their potential pitfalls. Some suggestions for readings include the articles in the books AI Game Programming 1 and 2, and the ones at this site: ai-depot.com/Main.html
Good luck!
09/20/2004 (12:10 pm)
"perceptrons for accurate AI vision and sight": last I checked (with a quick google search) perceptrons are simply 2 layer Neural Nets, they dont really have anything to do with "accurate AI vision" Also, while neural nets are very cool and useful for some things, they are not an absolute solution to game AI.I would suggest that your AI lead read up on neural nets before spending the time to implement and later realizing their potential pitfalls. Some suggestions for readings include the articles in the books AI Game Programming 1 and 2, and the ones at this site: ai-depot.com/Main.html
Good luck!
#5
But by using perceptrons, the computer will slowly learn to keep it's eyes open when looking for the player
09/21/2004 (12:53 am)
perceptrons for accurate AI vision and sightBut by using perceptrons, the computer will slowly learn to keep it's eyes open when looking for the player
#6
09/22/2004 (10:07 am)
Well, youre need to google perceptrons informs me that you arent exactly sure of what perceptrons are. If youd like I can let my AI lead email you about how we are using them. Thanks for the input and the quick research.
#7
Nice retort there in the previous post btw :)
09/22/2004 (11:18 pm)
How many bots do you expect to have in the game at once? Will they share NNs or each have their own?Nice retort there in the previous post btw :)
#8
09/23/2004 (1:30 am)
What does the perceptron do in layman's terms? How would you describe it to a gamer? 
#9
09/23/2004 (2:58 am)
I dont think it's such a great retort. I would stand behind what Josh said...perceptrons are just 2 layer neural nets...they are not, by themselves, gointg to solve any obvoius real-world problem with game AI for a shooter. You claim that Neural nets will produce realistic behaviour where the bots "learn" from you and will "out flank you". It is not clear that Neural Nets will deliver on this promise.
#10
Give that a read. Short (very) simple overview of how you could use a NN for game AI.
The important point of using NNs (or perceptrons) is the bots learn and behave differently (in an ideal setup). Easier said than done...
09/23/2004 (3:03 am)
www.gamedev.net/reference/articles/article1988.aspGive that a read. Short (very) simple overview of how you could use a NN for game AI.
The important point of using NNs (or perceptrons) is the bots learn and behave differently (in an ideal setup). Easier said than done...
#11
The neural network is to get the AI bots thinking and not just follwing commands telling it to chase you wherever you are. Pefrceptrons will help it think so it wont chase you unless it can actively percieve you in the game world. Anyway, i didnt start this thread to argue about who thinks what AI code is good and what works and what doesnt. Im just posting my idea. Ill post a link when its done. The game and most likely the source will be open for viewing, so yall can check it out and see how we did it. Its only three weeks away. maybe four.
Thanks for all the input.
Oh, and we will be having a multitude of bots, i believe to save on resources and because we only have two months to design a fully functional game, were sticking to everyone working of one, or perhas squads working off one with a few squads. Its being decided by my AI guy.
Anyother comments or questions are more than welcome. Thanks again.
09/23/2004 (7:37 am)
Ok...well....I didnt say we plan on implementing neural nets and have it solve all our problems....we are using them in conjunction with other things, perceptrons, neural nets, hierarchical state machines etc. our AI isnt running solely on neural nets. Dont you worry, we know what were doin.The neural network is to get the AI bots thinking and not just follwing commands telling it to chase you wherever you are. Pefrceptrons will help it think so it wont chase you unless it can actively percieve you in the game world. Anyway, i didnt start this thread to argue about who thinks what AI code is good and what works and what doesnt. Im just posting my idea. Ill post a link when its done. The game and most likely the source will be open for viewing, so yall can check it out and see how we did it. Its only three weeks away. maybe four.
Thanks for all the input.
Oh, and we will be having a multitude of bots, i believe to save on resources and because we only have two months to design a fully functional game, were sticking to everyone working of one, or perhas squads working off one with a few squads. Its being decided by my AI guy.
Anyother comments or questions are more than welcome. Thanks again.
#12
In the last post, you said the source may be open for viewing, if it is, would you be kind enough to give us a peek :] I'm interested in seeing how one would go about doing this.
05/11/2005 (1:27 pm)
MMmMM snow crash. In the last post, you said the source may be open for viewing, if it is, would you be kind enough to give us a peek :] I'm interested in seeing how one would go about doing this.
#13
The idea itself sounds extremely interesting, and yes, NN's can help assist with a wide variety of problems, but as others have pointed out they aren't a "sole source" solution for AI--but it sounds like your AI dev knows that!
It would be great to keep us updated as a community via .plans for your project, so please consider giving us updates from time to time!
05/11/2005 (1:36 pm)
One thing to keep in mind (and sorry for the minor hijack)--the TGE EULA specifically prohibits making any Torque code (C++/ASM) open source/available to the public--so if you do plan on making your work product available, you must be very careful to completely isolate it from the engine source code.The idea itself sounds extremely interesting, and yes, NN's can help assist with a wide variety of problems, but as others have pointed out they aren't a "sole source" solution for AI--but it sounds like your AI dev knows that!
It would be great to keep us updated as a community via .plans for your project, so please consider giving us updates from time to time!
#14
The game can be downloaded. I'll have to check the link to make sure it is still up. and basically all our code is viewable because it's all in the script files.
05/11/2005 (2:44 pm)
Well all of the implementation we did was mainly through script files. We rarely modified the engine code. If you are interested in seeing a certain part I may be able to dig up the scripts. Most of the engine code modifications we made were possible due to the tutorials from GG that you get access to after you buy the engine. The community helps out a lot by giving you some direction if you tried everything and just can't figure it out. The game can be downloaded. I'll have to check the link to make sure it is still up. and basically all our code is viewable because it's all in the script files.
Torque Owner Will Harrison
Re custom special effects: maybe look into TSE and the Lighting Pack. I havent delved deeply into these subjects so I dont know exactly.
I know with OpenGL, a celshading outline is fairly easy to implement because there is an option to define line thickness of a wireframe... but first you need to detect the edges of the model by scanning for backface polys. Whereas, in DirectX, to create the outline, you have to extrude the vertices on their normals, effectively using geometry to create the outline... there are other ways to do it too, but there is no built in line thickness function. Dont know of any tut sites off hand...