[RLlib] Unity3D integration (n Unity3D clients vs learning server). (#8590)

rllib/examples/env/multi_agent.py

@@ -4,12 +4,16 @@ from ray.rllib.env.multi_agent_env import MultiAgentEnv
 from ray.rllib.tests.test_rollout_worker import MockEnv, MockEnv2
 
 
-def make_multiagent(env_name):
+def make_multiagent(env_name_or_creator):
     class MultiEnv(MultiAgentEnv):
         def __init__(self, config):
-            self.agents = [
-                gym.make(env_name) for _ in range(config["num_agents"])
-            ]
+            num = config.pop("num_agents", 1)
+            if isinstance(env_name_or_creator, str):
+                self.agents = [
+                    gym.make(env_name_or_creator) for _ in range(num)
+                ]
+            else:
+                self.agents = [env_name_or_creator(config) for _ in range(num)]
             self.dones = set()
             self.observation_space = self.agents[0].observation_space
             self.action_space = self.agents[0].action_space

rllib/examples/env/random_env.py

@@ -1,7 +1,9 @@
 import gym
-from gym.spaces import Tuple
+from gym.spaces import Discrete, Tuple
 import numpy as np
 
+from ray.rllib.examples.env.multi_agent import make_multiagent
+
 
 class RandomEnv(gym.Env):
     """A randomly acting environment.
@@ -14,9 +16,9 @@ class RandomEnv(gym.Env):
 
     def __init__(self, config):
         # Action space.
-        self.action_space = config["action_space"]
+        self.action_space = config.get("action_space", Discrete(2))
         # Observation space from which to sample.
-        self.observation_space = config["observation_space"]
+        self.observation_space = config.get("observation_space", Discrete(2))
         # Reward space from which to sample.
         self.reward_space = config.get(
             "reward_space",
@@ -43,3 +45,7 @@ class RandomEnv(gym.Env):
             bool(np.random.choice(
                 [True, False], p=[self.p_done, 1.0 - self.p_done]
             )), {}
+
+
+# Multi-agent version of the RandomEnv.
+RandomMultiAgentEnv = make_multiagent(lambda c: RandomEnv(c))