Lightweight version of MAPPO to help you quickly migrate to your local environment.
- Video (in Chinese)
This is a translated English version. Please click here for the orginal Chinese readme.
The original MAPPO code was too complex in terms of environment encapsulation, so this project directly extracts and encapsulates the environment. This makes it easier to transfer the MAPPO code to your own project.
Simply download the code, create a Conda environment, and then run the code, adding packages as needed. Specific packages will be added later.
- The environment part is an empty implementation, and the implementation of the environment part in the light_mappo/envs/env_core.py file is: [Code] (https://github.com/tinyzqh/light_mappo/blob/main/envs/env_core.py)
import numpy as np
class EnvCore(object):
"""
# Environment Agent
"""
def __init__(self):
self.agent_num = 2 # set the number of agents(aircrafts), here set to two
self.obs_dim = 14 # set the observation dimension of agents
self.action_dim = 5 # set the action dimension of agents, here set to a five-dimensional
def reset(self):
"""
# When self.agent_num is set to 2 agents, the return value is a list, and each list contains observation data of shape = (self.obs_dim,)
"""
sub_agent_obs = []
for i in range(self.agent_num):
sub_obs = np.random.random(size=(14, ))
sub_agent_obs.append(sub_obs)
return sub_agent_obs
def step(self, actions):
"""
# When self.agent_num is set to 2 agents, the input of actions is a two-dimensional list, and each list contains action data of shape = (self.action_dim,).
# By default, the input is a list containing two elements, because the action dimension is 5, so each element has a shape of (5,)
"""
sub_agent_obs = []
sub_agent_reward = []
sub_agent_done = []
sub_agent_info = []
for i in range(self.agent_num):
sub_agent_obs.append(np.random.random(size=(14,)))
sub_agent_reward.append([np.random.rand()])
sub_agent_done.append(False)
sub_agent_info.append({})
return [sub_agent_obs, sub_agent_reward, sub_agent_done, sub_agent_info]Just write this part of the code, and you can seamlessly connect with MAPPO. After env_core.py, two files, env_discrete.py and env_continuous.py, were separately extracted to encapsulate the action space and discrete action space. In elif self.continuous_action: in algorithms/utils/act.py, this judgment logic is also used to handle continuous action spaces. The # TODO here in runner/shared/env_runner.py is also used to handle continuous action spaces.
In the train.py file, choose to comment out continuous environment or discrete environment to switch the demo environment.
- on-policy - 💌 Learn the author implementation of MAPPO.
MIT © tinyzqh