|
|
|
|
|
|
|
n_agents = len(self._env_state[group_name][0]) |
|
|
|
self._env_actions[group_name] = self._env_specs[ |
|
|
|
group_name |
|
|
|
].create_empty_action(n_agents) |
|
|
|
].action_spec.create_empty_action(n_agents) |
|
|
|
step_input = self._generate_step_input(self._env_actions) |
|
|
|
with hierarchical_timer("communicator.exchange"): |
|
|
|
outputs = self._communicator.exchange(step_input) |
|
|
|
|
|
|
|
if behavior_name not in self._env_state: |
|
|
|
return |
|
|
|
spec = self._env_specs[behavior_name] |
|
|
|
expected_type = np.float32 if spec.is_action_continuous() else np.int32 |
|
|
|
expected_shape = (len(self._env_state[behavior_name][0]), spec.action_size) |
|
|
|
expected_type = ( |
|
|
|
np.float32 if spec.action_spec.is_action_continuous() else np.int32 |
|
|
|
) |
|
|
|
expected_shape = ( |
|
|
|
len(self._env_state[behavior_name][0]), |
|
|
|
spec.action_spec.action_size, |
|
|
|
) |
|
|
|
if action.shape != expected_shape: |
|
|
|
raise UnityActionException( |
|
|
|
f"The behavior {behavior_name} needs an input of dimension " |
|
|
|
|
|
|
|
if behavior_name not in self._env_state: |
|
|
|
return |
|
|
|
spec = self._env_specs[behavior_name] |
|
|
|
expected_shape = (spec.action_size,) |
|
|
|
expected_shape = (spec.action_spec.action_size,) |
|
|
|
if action.shape != expected_shape: |
|
|
|
raise UnityActionException( |
|
|
|
f"The Agent {agent_id} with BehaviorName {behavior_name} needs " |
|
|
|
|
|
|
|
expected_type = np.float32 if spec.is_action_continuous() else np.int32 |
|
|
|
expected_type = ( |
|
|
|
np.float32 if spec.action_spec.is_action_continuous() else np.int32 |
|
|
|
) |
|
|
|
self._env_actions[behavior_name] = spec.create_empty_action( |
|
|
|
self._env_actions[behavior_name] = spec.action_spec.create_empty_action( |
|
|
|
len(self._env_state[behavior_name][0]) |
|
|
|
) |
|
|
|
try: |
|
|
|
|
Andrew Cohen