class EnvWrapper(object):
'''
Wrapper around UnityEnvironment that resets each arena if the episode is done
It will only work correctly if using a single arena on each environment
'''
def __init__(self, *args, **kwargs):
'''
Check UnityEnvironment parameters
'''
self._env = UnityEnvironment(*args, **kwargs)
self._arenas_configurations = None
def __getattr__(self, attr):
if attr in self.__dict__:
return getattr(self, attr)
return getattr(self._env, attr)
def reset(self, arenas_configurations=None, train_mode=True):
""" Shuffle arenas and reset configuration """
if arenas_configurations is not None:
self._arenas_configurations = arenas_configurations
self._arenas_configurations.shuffle_arenas()
return self._env.reset(self._arenas_configurations, train_mode)
def step(self, *args, **kwargs):
ret = self._env.step(*args, **kwargs)
if ret['Learner'].local_done[0]:
new_ret = self.reset()
ret['Learner'].visual_observations = new_ret['Learner'].visual_observations
return ret
def debug_confirm_arena_config(env_path, arena_config):
env_path = (env_path.strip().replace('.app', '').replace(
'.exe', '').replace('.x86_64', '').replace('.x86', ''))
from animalai.envs.environment import UnityEnvironment
env = UnityEnvironment(n_arenas=16,
file_name=env_path,
worker_id=1,
seed=0,
docker_training=False,
play=True)
env.reset(arenas_configurations=arena_config)
try:
while True:
continue
except KeyboardInterrupt:
env.close()
def main(args):
docker_training = docker_target_name is not None
env = UnityEnvironment(
n_arenas=args.n_arenas,
file_name=env_path,
worker_id=worker_id,
seed=seed,
docker_training=docker_training,
play=False,
resolution=resolution
)
arena_config_in = ArenaConfig('configs/3-Obstacles.yaml')
env.reset(arenas_configurations=arena_config_in)
start_time = time.time()
for i in range(args.frames):
res = env.step(np.random.randint(0, 3, size=2 * args.n_arenas))
elapsed_time = time.time() - start_time
fps = float(args.frames) / elapsed_time
print("n_arenas={0}, fps={1:.3f}".format(args.n_arenas, fps))
env.close()
line, = ax.plot(
[], []) # the direction that the agent want to go given the visual input
sca = ax.scatter([], [], s=5, c="yellow") # the path to the target(food)
# loop over all the config files above
for arenaConfig in arenaConfigs:
print(arenaConfig)
arena_config_in = ArenaConfig(arenaConfig)
# run multiple tests derived from each config file
for sample_n in range(sample_size_per_task):
print("ArenaConfig; {} Sample: {}".format(arenaConfig, sample_n))
# initialize(reset) the agent and the env
agent.reset(arena_config_in.arenas[0].t)
brainInfo = env.reset(arenas_configurations=arena_config_in)
while True:
# information given by the env
obs = brainInfo['Learner'].visual_observations[0][
0, :, :, :], brainInfo['Learner'].vector_observations[0]
reward = brainInfo['Learner'].rewards[0]
done = brainInfo['Learner'].local_done[0]
info = {"brain_info": brainInfo}
# let the agent generate an action based on the information
action = agent.step(obs, reward, done, info)
# Visualization{visual, direction, path}
image.set_data(obs[0])
n_arenas = 4
if env_path is not None:
env_path = (env_path.strip().replace('.app', '').replace(
'.exe', '').replace('.x86_64', '').replace('.x86', ''))
docker_training = docker_target_name is not None
env = UnityEnvironment(n_arenas=n_arenas,
file_name=env_path,
worker_id=worker_id,
seed=seed,
docker_training=docker_training,
play=False)
arena_config_in = ArenaConfig('configs/lightsOff.yaml')
env.reset(arenas_configurations=arena_config_in)
fig, axes = plt.subplots(2, 2)
imshows = []
for i in range(2):
for j in range(2):
axes[i, j].set_title('Arena ' + str(i * 2 + j))
axes[i, j].axis('off')
imshows.append(axes[i, j].imshow(np.zeros((84, 84, 3))))
def initialize_animation():
for i in range(4):
imshows[i].set_data(np.zeros((84, 84, 3)))
def run_step_imshow(step):