Posts for moth

Posted: 4/10/2018 3:12 PM
Quote
moth
He/Him
Experienced Forum User
Joined: 4/8/2018
Posts: 3
A display like this, where green is an enemy pulse, red is the player pulse and blue is the player. The whole thing is 100 x 100 cell array and it will be passed into the neural network as input nodes. Obtaining position of enemies from memory is not straightforward as there is none, but we figured out a solution which will be posted on github once it is polished.
Posted: 4/10/2018 5:19 AM
Quote
moth
He/Him
Experienced Forum User
Joined: 4/8/2018
Posts: 3
Repos have been synced and the code has been corrected, it should have been playerY instead of playerX, thanks for pointing it out. I have not worked with neural networks but after reading the code for MarI/O , I figured its NEAT Neural Network code can be used make a similar AI program for Space Invaders with some changes. Right now we are trying to load the the positions of entities into the cells array of size 100x100 and displaying it. Then we will pass it as an input into the neural network. Also, what's the other approach apart from mapping memory, are you talking about taking screenshots every 5-6 frames and processing them for inputs?
Posted: 4/8/2018 5:13 AM
Post subject: Space Invaders
Quote
moth
He/Him
Experienced Forum User
Joined: 4/8/2018
Posts: 3
Hey guys I am new to TAS and having been working on BizHawk for about two weeks now. So we are trying to make a Neural Network play Space Invaders(NES), something like MarI/O by SethBling, we have got the memory location of almost all of the entities except for the aliens and bunkers. If someone with experience in combing through RAM can help us that would be great. Here is our GitHub link where we have documented the memory map: (For Projectiles): https://github.com/BhayanakMoth2/AI-nvader/wiki (For Player and UFO position): https://github.com/CryoScythe/AI-nvader/wiki Many thanks if you are gonna chip some time and efforts.