Adversarial Pursuit

environment gif

This environment is part of the magent environments. Please read that page first for general information.

Name Value
Actions Discrete
Agents 75
Parallel API True
Manual Control No
Action Shape (9),(13)
Action Values Discrete(9),(13)
Observation Shape (9,9,15), (10,10,19)
Observation Values [0,2]
Import pettingzoo.magent import adversarial_pursuit_v1
Agents agents= [predator_[0-24], prey_[0-49]]

Agent Environment Cycle

environment aec diagram

Adversarial Pursuit

The red agents must navigate the obstacles and attack the blue agents. The blue agents should try to avoid being attacked. Since the red agents are slower (but larger) than the blue agents, they must work together to trap the blue agents, so they can attack them continually (note that they blue agent’s won’t die if attacked, so they can be used as an infinite source of reward).

Predator action options: [do_nothing, move_4, attack_8]

Predator’s reward is given as:

Prey action options: [do_nothing, move_8]

Prey’s reward is given as:

Observation space: [empty, obstacle, predators, prey, one_hot_action, last_reward]

Map size: 45x45

Arguments

adversarial_pursuit_v1.env(attack_penalty=-0.2, max_frames=500)

attack_penalty: Adds the following value to the reward whenever an attacking action is taken

max_frames: number of frames (a step for each agent) until game terminates