Vote count:
0
Currently, I am trying to create a simple 2-D predator prey model where there are two different agent types, the predator and the prey. In this model, we assume that the only actions that predator and the prey is that they can only move around at certain directions and if the predator is close to a prey, then the prey dies. A predator cannot kill other predators. The velocity, acceleration, location of all the prey are all known. Also, the agents are in a n-by-n grid and there are x predators and y prey. The user can control these parameters.
The behavior of the prey is defined by a differential equation and I am trying to develop a reinforcement learning algorithm for the predator agents. The goal of these agents is to try to maximize the amount of prey that they eat. Obviously, the state space for this environment is very big and I am trying to determine the best features and states that can represent the environment for the prey. Does anybody have any suggestions?
Originally, I was thinking about having the features only be represent a 10-by-10 gird that sounds a predator and they information will be held in a 100 size vector. If a prey is in the coordinate of this grid, then that corresponding vector point in that grid will contain information about the prey's speed, acceleration and position. Otherwise, that information is listed as 0.
Aucun commentaire:
Enregistrer un commentaire