def step(self, model_states: StatesModel,
env_states: StatesEnv) -> StatesEnv:
"""
Set the environment to the target states by applying the specified \
actions an arbitrary number of time steps.
The state transitions will be calculated in parallel.
Args:
model_states: :class:`StatesModel` representing the data to be used \
to act on the environment.
env_states: :class:`StatesEnv` representing the data to be set in \
the environment.
Returns:
:class:`StatesEnv` containing the information that describes the \
new state of the Environment.
"""
split_env_states = [
env.step.remote(model_states=ms, env_states=es)
for env, ms, es in zip(
self.envs,
model_states.split_states(self.n_workers),
env_states.split_states(self.n_workers),
)
]
env_states = ray.get(split_env_states)
new_env_states: StatesEnv = StatesEnv.merge_states(env_states)
return new_env_states
def step(self, model_states: StatesModel,
env_states: StatesEnv) -> StatesEnv:
"""
Forward a batch of actions to the wrapped environments.
Args:
model_states: States representing the data to be used to act on the environment.
env_states: States representing the data to be set in the environment.
Returns:
Batch of observations, rewards, and done flags.
"""
split_states = self._make_transitions(model_states=model_states,
env_states=env_states)
states: StatesEnv = StatesEnv.merge_states(split_states)
return states
