Updated code to run with newest version of OpenAI gym (>=0.10)

Author: ryan-lowe

README.md

@@ -10,7 +10,7 @@ Used in the paper [Multi-Agent Actor-Critic for Mixed Cooperative-Competitive En
 - To interactively view moving to landmark scenario (see others in ./scenarios/):
 `bin/interactive.py --scenario simple.py`
 
-- Known dependencies: OpenAI gym, numpy
+- Known dependencies: OpenAI gym version >=0.10, numpy
 
 - To use the environments, look at the code for importing them in `make_env.py`.
 

multiagent/environment.py

@@ -43,14 +43,14 @@ def __init__(self, world, reset_callback=None, reward_callback=None,
  if self.discrete_action_space:
  u_action_space = spaces.Discrete(world.dim_p * 2 + 1)
  else:
- u_action_space = spaces.Box(low=-agent.u_range, high=+agent.u_range, shape=(world.dim_p,))
+ u_action_space = spaces.Box(low=-agent.u_range, high=+agent.u_range, shape=(world.dim_p,), dtype=np.float32)
  if agent.movable:
  total_action_space.append(u_action_space)
  # communication action space
  if self.discrete_action_space:
  c_action_space = spaces.Discrete(world.dim_c)
  else:
- c_action_space = spaces.Box(low=0.0, high=1.0, shape=(world.dim_c,))
+ c_action_space = spaces.Box(low=0.0, high=1.0, shape=(world.dim_c,), dtype=np.float32)
  if not agent.silent:
  total_action_space.append(c_action_space)
  # total action space
@@ -65,7 +65,7 @@ def __init__(self, world, reset_callback=None, reward_callback=None,
  self.action_space.append(total_action_space[0])
  # observation space
  obs_dim = len(observation_callback(agent, self.world))
- self.observation_space.append(spaces.Box(low=-np.inf, high=+np.inf, shape=(obs_dim,)))
+ self.observation_space.append(spaces.Box(low=-np.inf, high=+np.inf, shape=(obs_dim,), dtype=np.float32))
  agent.action.c = np.zeros(self.world.dim_c)
 
  # rendering
@@ -76,7 +76,7 @@ def __init__(self, world, reset_callback=None, reward_callback=None,
  self.viewers = [None] * self.n
  self._reset_render()
 
- def _step(self, action_n):
+ def step(self, action_n):
  obs_n = []
  reward_n = []
  done_n = []
@@ -102,7 +102,7 @@ def _step(self, action_n):
 
  return obs_n, reward_n, done_n, info_n
 
- def _reset(self):
+ def reset(self):
  # reset world
  self.reset_callback(self.world)
  # reset renderer
@@ -196,7 +196,7 @@ def _reset_render(self):
  self.render_geoms_xform = None
 
  # render environment
- def _render(self, mode='human', close=True):
+ def render(self, mode='human'):
  if mode == 'human':
  alphabet = 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'
  message = ''
@@ -211,14 +211,6 @@ def _render(self, mode='human', close=True):
  message += (other.name + ' to ' + agent.name + ': ' + word + ' ')
  print(message)
 
- if close:
- # close any existic renderers
- for i,viewer in enumerate(self.viewers):
- if viewer is not None:
- viewer.close()
- self.viewers[i] = None
- return []
-
  for i in range(len(self.viewers)):
  # create viewers (if necessary)
  if self.viewers[i] is None:
@@ -313,7 +305,7 @@ def action_space(self):
  def observation_space(self):
  return self.env_batch[0].observation_space
 
- def _step(self, action_n, time):
+ def step(self, action_n, time):
  obs_n = []
  reward_n = []
  done_n = []
@@ -328,14 +320,14 @@ def _step(self, action_n, time):
  done_n += done
  return obs_n, reward_n, done_n, info_n
 
- def _reset(self):
+ def reset(self):
  obs_n = []
  for env in self.env_batch:
  obs_n += env.reset()
  return obs_n
 
  # render environment
- def _render(self, mode='human', close=True):
+ def render(self, mode='human', close=True):
  results_n = []
  for env in self.env_batch:
  results_n += env.render(mode, close)