Partial conversion to latset mlagents

Author: awjuliani

obstacle_tower_env.py

@@ -9,7 +9,7 @@
 import time
 from collections import deque
 from gym import error, spaces
-from mlagents.envs.environment import UnityEnvironment
+from mlagents_envs.environment import UnityEnvironment
 
 
 class UnityGymException(error.Error):
@@ -24,7 +24,7 @@ class UnityGymException(error.Error):
 
 
 class ObstacleTowerEnv(gym.Env):
- ALLOWED_VERSIONS = ['3.1']
+ ALLOWED_VERSIONS = ['4.0?team=0']
 
  def __init__(self, environment_filename=None, docker_training=False, worker_id=0, retro=True,
  timeout_wait=30, realtime_mode=False, config=None, greyscale=False):
@@ -41,11 +41,12 @@ def __init__(self, environment_filename=None, docker_training=False, worker_id=0
  """
  self._env = UnityEnvironment(environment_filename,
  worker_id,
- docker_training=docker_training,
  timeout_wait=timeout_wait)
 
- split_name = self._env.academy_name.split('-v')
- if len(split_name) == 2 and split_name[0] == "ObstacleTower":
+ self._env.reset()
+ behavior_name = list(self._env.behavior_specs)[0]
+ split_name = behavior_name.split('-v')
+ if len(split_name) == 2 and split_name[0] == "ObstacleTowerAgent":
  self.name, self.version = split_name
  else:
  raise UnityGymException(
@@ -80,48 +81,41 @@ def __init__(self, environment_filename=None, docker_training=False, worker_id=0
  flatten_branched = self.retro
  uint8_visual = self.retro
 
- # Check brain configuration
- if len(self._env.brains) != 1:
+ # Check behavior configuration
+ if len(self._env.behavior_specs) != 1:
  raise UnityGymException(
- "There can only be one brain in a UnityEnvironment "
+ "There can only be one agent in this environment "
  "if it is wrapped in a gym.")
- self.brain_name = self._env.external_brain_names[0]
- brain = self._env.brains[self.brain_name]
+ self.behavior_name = behavior_name
+ behavior_spec = self._env.behavior_specs[behavior_name]
 
- if brain.number_visual_observations == 0:
- raise UnityGymException("Environment provides no visual observations.")
+ if len(behavior_spec) < 2:
+ raise UnityGymException("Environment provides too few observations.")
 
  self.uint8_visual = uint8_visual
 
- if brain.number_visual_observations > 1:
- logger.warning("The environment contains more than one visual observation. "
- "Please note that only the first will be provided in the observation.")
-
  # Check for number of agents in scene.
- initial_info = self._env.reset(train_mode=not self.realtime_mode)[self.brain_name]
- self._check_agents(len(initial_info.agents))
+ initial_info, _ = self._env.get_steps(behavior_name)
+ self._check_agents(len(initial_info))
 
  # Set observation and action spaces
- if len(brain.vector_action_space_size) == 1:
- self._action_space = spaces.Discrete(brain.vector_action_space_size[0])
+ if len(behavior_spec.action_shape) == 1:
+ self._action_space = spaces.Discrete(behavior_spec.action_shape[0])
  else:
  if flatten_branched:
- self._flattener = ActionFlattener(brain.vector_action_space_size)
+ self._flattener = ActionFlattener(behavior_spec.action_shape)
  self._action_space = self._flattener.action_space
  else:
- self._action_space = spaces.MultiDiscrete(brain.vector_action_space_size)
-
- high = np.array([np.inf] * brain.vector_observation_space_size)
- self.action_meanings = brain.vector_action_descriptions
+ self._action_space = spaces.MultiDiscrete(behavior_spec.action_shape)
 
  if self._greyscale:
  depth = 1
  else:
  depth = 3
  image_space_max = 1.0
  image_space_dtype = np.float32
- camera_height = brain.camera_resolutions[0]["height"]
- camera_width = brain.camera_resolutions[0]["width"]
+ camera_height = behavior_spec.observation_shapes[0][0]
+ camera_width = behavior_spec.observation_shapes[0][1]
  if self.retro:
  image_space_max = 255
  image_space_dtype = np.uint8
@@ -163,7 +157,7 @@ def reset(self, config=None):
 
  self.reset_params = self._env.reset_parameters
  info = self._env.reset(config=reset_params,
- train_mode=not self.realtime_mode)[self.brain_name]
+ train_mode=not self.realtime_mode)[self.behavior_name]
  n_agents = len(info.agents)
  self._check_agents(n_agents)
  self.game_over = False
@@ -191,7 +185,7 @@ def step(self, action):
  # Translate action into list
  action = self._flattener.lookup_action(action)
 
- info = self._env.step(action)[self.brain_name]
+ info = self._env.step(action)[self.behavior_name]
  n_agents = len(info.agents)
  self._check_agents(n_agents)
  self._current_state = info
@@ -246,9 +240,6 @@ def close(self):
  """
  self._env.close()
 
- def get_action_meanings(self):
- return self.action_meanings
-
  def seed(self, seed=None):
  """Sets a fixed seed for this env's random number generator(s).
  The valid range for seeds is [0, 99999). By default a random seed