Hi, I'm trying to use this library to solve some modified multi-UAV problems, but I'm

Thank you to <a class="user-mention notranslate" data-hovercard-type="user" data-hover

Working with my own customized env about marllib HOT 3 OPEN

HawkQ commented on June 10, 2024

Working with my own customized env

from marllib.

Comments (3)

Aequatio-Space commented on June 10, 2024

You can use an example from marllib/envs/base_env to check how a custom environment is wrapped for RLlib. I write my own environment according to such example.

MARLlib/marllib/envs/base_env/mpe.py

Lines 94 to 153 in 368c617

 class RLlibMPE(MultiAgentEnv): 

 def __init__(self, env_config): 

 map = env_config["map_name"] 

 env_config.pop("map_name", None) 

 env = REGISTRY[map](**env_config) 

 # keep obs and action dim same across agents 

 # pad_action_space_v0 will auto mask the padding actions 

 env = ss.pad_observations_v0(env) 

 env = ss.pad_action_space_v0(env) 

 self.env = ParallelPettingZooEnv(env) 

 self.action_space = self.env.action_space 

 self.observation_space = GymDict({"obs": Box( 

 low=-100.0, 

 high=100.0, 

 shape=(self.env.observation_space.shape[0],), 

 dtype=self.env.observation_space.dtype)}) 

 self.agents = self.env.agents 

 self.num_agents = len(self.agents) 

 env_config["map_name"] = map 

 self.env_config = env_config 

 def reset(self): 

 original_obs = self.env.reset() 

 obs = {} 

 for i in self.agents: 

 obs[i] = {"obs": original_obs[i]} 

 return obs 

 def step(self, action_dict): 

 o, r, d, info = self.env.step(action_dict) 

 rewards = {} 

 obs = {} 

 for key in action_dict.keys(): 

 rewards[key] = r[key] 

 obs[key] = { 

 "obs": o[key] 

 } 

 dones = {"__all__": d["__all__"]} 

 return obs, rewards, dones, info 

 def close(self): 

 self.env.close() 

 def render(self, mode=None): 

 self.env.render() 

 time.sleep(0.05) 

 return True 

 def get_env_info(self): 

 env_info = { 

 "space_obs": self.observation_space, 

 "space_act": self.action_space, 

 "num_agents": self.num_agents, 

 "episode_limit": 25, 

 "policy_mapping_info": policy_mapping_dict 

 } 

 return env_info

from marllib.

ChenJiangxi commented on June 10, 2024

I'm having the same problem, and I think the key to customizing your environment is in this statement:
self.env = ParallelPettingZooEnv(env)
But I don't really know how to access my own simulation environment

from marllib.

Theohhhu commented on June 10, 2024

Thank you to @Aequatio-Space for offering guidance. MARLlib utilizes a standard agent-environment interface based on RLlib. A practical approach to integrating your own environment is to begin by identifying a similar environment (e.g., with the same data structure or interaction style) and then follow the integration approach used for that environment.

from marllib.

Recommend Projects

Working with my own customized env about marllib HOT 3 OPEN

Comments (3)

Related Issues (20)

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent

	class RLlibMPE(MultiAgentEnv):

	def __init__(self, env_config):
	map = env_config["map_name"]
	env_config.pop("map_name", None)
	env = REGISTRY[map](**env_config)

	# keep obs and action dim same across agents
	# pad_action_space_v0 will auto mask the padding actions
	env = ss.pad_observations_v0(env)
	env = ss.pad_action_space_v0(env)

	self.env = ParallelPettingZooEnv(env)
	self.action_space = self.env.action_space
	self.observation_space = GymDict({"obs": Box(
	low=-100.0,
	high=100.0,
	shape=(self.env.observation_space.shape[0],),
	dtype=self.env.observation_space.dtype)})
	self.agents = self.env.agents
	self.num_agents = len(self.agents)
	env_config["map_name"] = map
	self.env_config = env_config

	def reset(self):
	original_obs = self.env.reset()
	obs = {}
	for i in self.agents:
	obs[i] = {"obs": original_obs[i]}
	return obs

	def step(self, action_dict):
	o, r, d, info = self.env.step(action_dict)
	rewards = {}
	obs = {}
	for key in action_dict.keys():
	rewards[key] = r[key]
	obs[key] = {
	"obs": o[key]
	}
	dones = {"__all__": d["__all__"]}
	return obs, rewards, dones, info

	def close(self):
	self.env.close()

	def render(self, mode=None):
	self.env.render()
	time.sleep(0.05)
	return True

	def get_env_info(self):
	env_info = {
	"space_obs": self.observation_space,
	"space_act": self.action_space,
	"num_agents": self.num_agents,
	"episode_limit": 25,
	"policy_mapping_info": policy_mapping_dict
	}
	return env_info