environment gif

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

Name Value
Actions Discrete
Agents 495
Parallel API Yes
Manual Control No
Action Shape (33)
Action Values Discrete(33)
Observation Shape (15,15,5)
Observation Values [0,2]
Import pettingzoo.magent import gather_v3
Agents agents= [ omnivore_[0-494] ]

Agent Environment Cycle

environment aec diagram


In gather, the agents gain reward by eating food. Food needs to be broken down by 5 “attacks” before it is absorbed. Since there is finite food on the map, there is competitive pressure between agents over the food. You expect to see that agents coordinate by not attacking each other until food is scarce. When food is scarce, agents may attack each other to try to monopolize the food. Agents can kill each other with a single attack.

Action Space

Key: move_N means N separate actions, one to move to each of the N nearest squares on the grid.

Action options: [do_nothing, move_28, attack_4]


Reward is given as:

Observation space

The observation space is a 15x15 map with the below channels (in order):

name number of channels
obstacle/off the map 1
omnivore_presence 1
omnivore_hp 1
omnivore_minimap(minimap_mode=True) 1
food_presense 1
food_hp 1
food_minimap(minimap_mode=True) 1
one_hot_action(extra_features=True) 33
last_reward(extra_features=True) 1
agent_position(minimap_mode=True) 2


gather_v3.env(minimap_mode=False, step_reward=-0.01, attack_penalty=-0.1, dead_penalty=-1, attack_food_reward=0.5, max_cycles=500, extra_features=False)

minimap_mode: Turns on global minimap observations. These observations include your and your opponents piece densities binned over the 2d grid of the observation space. Also includes your agent_position, the absolute position on the map (rescaled from 0 to 1).

step_reward: reward added unconditionally

dead_penalty: reward added when killed

attack_penalty: reward added for attacking

attack_food_reward: Reward added for attacking a food

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

extra_features: Adds additional features to observation (see table). Default False