def __init__(self, env_fns):
DummyVecEnv.__init__(self, env_fns)
num_agents = getattr_unwrapped(self.envs[0], "num_agents")
VecMultiEnv.__init__(self, self.num_envs, num_agents,
self.observation_space, self.action_space)
self.buf_rews = np.zeros((self.num_envs, self.num_agents),
dtype=np.float32)
def __init__(self, env_fns, start_method=None):
if start_method is None:
start_method = "forkserver" # thread safe by default
SubprocVecEnv.__init__(self, env_fns, start_method=start_method)
env = env_fns[0]()
num_agents = getattr_unwrapped(env, "num_agents")
env.close()
VecMultiEnv.__init__(self, self.num_envs, num_agents,
self.observation_space, self.action_space)
def __init__(self, env):
if hasattr(env, "_max_episode_steps"):
# We step multiple times, then reset to a previous state.
# Timestep limit doesn't make much sense at this level.
# (Instead, apply it outside of the controller.)
raise TypeError("Environment must not have a timestep limit.")
self.env = env
self.sim = getattr_unwrapped(env, "sim")
state = MujocoState.from_mjdata(self.sim.data).flatten()
self._state_size = len(state)
self._action_size = reduce(lambda x, y: x * y, env.action_space.shape)
def __init__(
self,
env,
filename,
our_idx=None,
allow_early_resets=False,
reset_keywords=(),
info_keywords=(),
):
num_agents = getattr_unwrapped(env, "num_agents")
extra_rks = tuple("r{:d}".format(i) for i in range(num_agents))
super().__init__(
env,
filename,
allow_early_resets=allow_early_resets,
reset_keywords=reset_keywords,
info_keywords=extra_rks + info_keywords,
)
self.our_idx = our_idx
self.info_keywords = info_keywords
def __init__(self, env):
Wrapper.__init__(self, env)
MultiAgentEnv.__init__(self, getattr_unwrapped(env, "num_agents"))