NEW BLOG POST
You weren't doing anything on a Friday afternoon were you?
Here's 4 thousand words about the new and improved Chronicles behavior planning and how I leveraged reinforcement learning techniques used by Chess and Go algorithms to make my baddies smart and efficient. Enjoy!
https://blog.brianna.town/the-bigger-better-chronicles-npc-behavior-system
the opening paragraphs of this post have been slightly modified for correctness because the original actions post mentioned predated the first working implementation of behavior planning by several months.
This bothered nobody but me, please enjoy.
@britown I see, thanks!
Another question:
How do you handle hidden information?
Let's say the player places a trap that should be invisible to the enemy.
Now if the enemy simulates the outcome of moving into the traps tile it'd notice it takes damage and avoid the tile (possible getting stuck in a hallway).
Do you simulate as if the trap wasn't there?
@britown thanks, that was a great read - using MCTS to coordinate multiple enemy AIs is quite a novel approach!
A couple of questions. Do you do anything to simulate the player's move after each step? And how do you 'score' moves like 'block the exit'?
@bbbscarter I'm glad it was coherent! NPCs treat the player as doing nothing on their turn which is great when they go first and less reliable if they go after, but the player has the automatic into the breach advantage of seeing everyone's planned turn before deciding what to do.
The simulation does take into account end of turn tick down and recovery for the look ahead though, they'll see the damage of standing in fire or understand that recovering stamina for a turn allows a good ability
gamedevjeff
Brianna Townsend
Aviv
Inglo Games
Kampffrosch
Simon Carter
Taylor Hadden