class IndoorEnvironment(environment.Environment):
ACTION_LIST = [
[1,0,0],
[0,1,0],
[0,0,1]
]
@staticmethod
def get_action_size(env_name):
return len(IndoorEnvironment.ACTION_LIST)
@staticmethod
def get_objective_size(env_name):
simargs = sim_config.get(env_name)
return simargs.get('objective_size', 0)
def __init__(self, env_name, env_args, thread_index):
environment.Environment.__init__(self)
self.last_state = None
self.last_action = 0
self.last_reward = 0
simargs = sim_config.get(env_name)
simargs['id'] = 'sim%02d' % thread_index
simargs['logdir'] = os.path.join(IndoorEnvironment.get_log_dir(), simargs['id'])
# Merge in extra env args
if env_args is not None:
simargs.update(env_args)
self._sim = RoomSimulator(simargs)
self._sim_obs_space = self._sim.get_observation_space(simargs['outputs'])
self.reset()
def reset(self):
result = self._sim.reset()
self._episode_info = result.get('episode_info')
self._last_full_state = result.get('observation')
obs = self._last_full_state['observation']['sensors']['color']['data']
objective = self._last_full_state.get('measurements')
state = { 'image': self._preprocess_frame(obs), 'objective': objective }
self.last_state = state
self.last_action = 0
self.last_reward = 0
def stop(self):
if self._sim is not None:
self._sim.close_game()
def _preprocess_frame(self, image):
if len(image.shape) == 2: # assume gray
image = np.dstack([image, image, image])
else: # assume rgba
image = image[:, :, :-1]
image = image.astype(np.float32)
image = image / 255.0
return image
def process(self, action):
real_action = IndoorEnvironment.ACTION_LIST[action]
full_state = self._sim.step(real_action)
self._last_full_state = full_state # Last observed state
obs = full_state['observation']['sensors']['color']['data']
reward = full_state['rewards']
terminal = full_state['terminals']
objective = full_state.get('measurements')
if not terminal:
state = { 'image': self._preprocess_frame(obs), 'objective': objective }
else:
state = self.last_state
pixel_change = self._calc_pixel_change(state['image'], self.last_state['image'])
self.last_state = state
self.last_action = action
self.last_reward = reward
return state, reward, terminal, pixel_change
def is_all_scheduled_episodes_done(self):
return self._sim.is_all_scheduled_episodes_done()
class IndoorEnvironment(environment.Environment):
ACTION_LIST = [[1, 0, 0], [0, 1, 0], [0, 0, 1]]
@staticmethod
def get_action_size(env_name):
return len(IndoorEnvironment.ACTION_LIST)
@staticmethod
def get_objective_size(env_name):
simargs = sim_config.get(env_name)
return simargs.get('objective_size', 0)
def __init__(self, env_name, env_args, termination_time, thread_index):
environment.Environment.__init__(self)
try:
self.last_state = None
self.last_action = 0
self.last_reward = 0
self.prev_state = None
self.prev_action = 0
self.prev_reward = 0
simargs = sim_config.get(env_name)
simargs['id'] = 'sim%02d' % thread_index
simargs['logdir'] = os.path.join(IndoorEnvironment.get_log_dir(),
simargs['id'])
# Merge in extra env args
if env_args is not None:
simargs.update(env_args)
simargs["measure_fun"].termination_time = termination_time
self.termination_time = termination_time
# try:
self._sim = RoomSimulator(simargs)
self._sim_obs_space = self._sim.get_observation_space(
simargs['outputs'])
self.reset()
except Exception as e:
print("Error in indoor_env init():", str(e)) #, flush=True)
raise Exception
def reset(self):
result = self._sim.reset()
self._episode_info = result.get('episode_info')
self._last_full_state = result.get('observation')
img = self._last_full_state['observation']['sensors']['color']['data']
objective = self._last_full_state.get(
'measurements') # with measure function!
state = {'image': self._preprocess_frame(img), 'objective': objective}
object_type = self._last_full_state["observation"]["sensors"].get(
"objectType", None)
if object_type is not None:
object_type = object_type["data"][:, :, 2]
state.update(
{'objectType': self._preprocess_frame(object_type, "segm")})
# print(object_type.shape)
self.last_state = state
self.last_action = 0
self.last_reward = 0
self.prev_state = None
self.prev_action = 0
self.prev_reward = 0
def stop(self):
if self._sim is not None:
self._sim.close_game()
def _preprocess_frame(self, image, mode="segm"):
if len(image.shape) == 2: # assume object_type or depth
image = image.reshape((image.shape[1], image.shape[0]))
if "segm" in mode:
image[image == 255] = 0
return image.astype(np.int32)
#image = np.dstack([image, image, image])
else: # assume rgba
image = image[:, :, :-1]
image = image.reshape((image.shape[1], image.shape[0], image.shape[2]))
#print(image.shape)
#Reshape is essential, when non-square image from simulator!
image = image.astype(np.float32)
image = image / 255.0
return image
def process(self, action, flag=1):
real_action = IndoorEnvironment.ACTION_LIST[action]
full_state = self._sim.step(real_action, flag=flag)
#print("Step made")
self._last_full_state = full_state # Last observed state
obs = full_state['observation']['sensors']['color']['data']
reward = full_state[
'rewards'] / self.termination_time # reward clipping
terminal = full_state['terminals']
objective = full_state.get('measurements')
object_type = self._last_full_state["observation"]["sensors"].get(
"objectType", None)
if not terminal:
state = {
'image': self._preprocess_frame(obs),
'objective': objective
}
if object_type is not None:
object_type = object_type["data"][:, :, 2]
state.update({
'objectType':
self._preprocess_frame(object_type, "segm")
})
else:
state = self.last_state
pixel_change = None
if object_type is None:
pixel_change = self._calc_pixel_change(state['image'],
self.last_state['image'])
self.prev_state = self.last_state
self.prev_action = self.last_action
self.prev_reward = self.last_reward
self.last_state = state
self.last_action = action
self.last_reward = reward
return state, reward, terminal, pixel_change
def is_all_scheduled_episodes_done(self):
return self._sim.is_all_scheduled_episodes_done()
class IndoorEnv(gym.Env):
metadata = {'render.modes': ['human', 'rgb_array']}
def __init__(self):
self._last_state = None
self._sim = None
self.viewer = None
def configure(self, sim_args):
self._sim = RoomSimulator(sim_args)
#signal.signal(signal.SIGINT, self.signal_handler)
self._sim_obs_space = self._sim.get_observation_space(sim_args['outputs'])
#self.action_space = spaces.Discrete(self._sim.num_buttons)
self.action_space = spaces.MultiBinary(self._sim.num_buttons)
self.screen_height = self._sim_obs_space['color'].shape[1]
self.screen_width = self._sim_obs_space['color'].shape[0]
self.observation_space = spaces.Box(low=0, high=255,
shape=(self.screen_height, self.screen_width, 3))
# TODO: have more complex observation space with additional modalities and measurements
# obs_space = self._sim.get_observation_space
#self.observation_space = spaces.Dict({"images": ..., "depth": ...})
@property
def simulator(self):
return self._sim.sim
def _seed(self, seed=0xA3C):
"""Sets the seed for this env's random number generator(s).
Note:
Some environments use multiple pseudorandom number generators.
We want to capture all such seeds used in order to ensure that
there aren't accidental correlations between multiple generators.
Returns:
list<bigint>: Returns the list of seeds used in this env's random
number generators. The first value in the list should be the
"main" seed, or the value which a reproducer should pass to
'seed'. Often, the main seed equals the provided 'seed', but
this won't be true if seed=None, for example.
"""
# TODO: generate another seed for use in simulator?
# What happens to this seed?
self.np_random, seed = seeding.np_random(seed)
return [seed]
def _reset(self):
"""Resets the state of the environment and returns an initial observation.
Returns: observation (object): the initial observation of the
space.
"""
res = self._sim.reset()
return res.get('observation')
def _step(self, action):
"""Run one timestep of the environment's dynamics. When end of
episode is reached, you are responsible for calling `reset()`
to reset this environment's state.
Accepts an action and returns a tuple (observation, reward, done, info).
Args:
action (object): an action provided by the environment
Returns:
observation (object): agent's observation of the current environment
reward (float) : amount of reward returned after previous action
done (boolean): whether the episode has ended, in which case further step() calls will return undefined results
info (dict): contains auxiliary diagnostic information (helpful for debugging, and sometimes learning)
"""
## a = [0]*self._sim.num_buttons
## a[action] = 1
state = self._sim.step(action)
self._last_state = state # Last observed state
observation = {k:v for k,v in state.items() if k not in ['rewards','terminals']}
info = state['info']
return observation, state['rewards'], state['terminals'], info
def _render(self, mode='human', close=False):
"""Renders the environment.
The set of supported modes varies per environment. (And some
environments do not support rendering at all.) By convention,
if mode is:
- human: render to the current display or terminal and
return nothing. Usually for human consumption.
- rgb_array: Return an numpy.ndarray with shape (x, y, 3),
representing RGB values for an x-by-y pixel image, suitable
for turning into a video.
- ansi: Return a string (str) or StringIO.StringIO containing a
terminal-style text representation. The text can include newlines
and ANSI escape sequences (e.g. for colors).
Note:
Make sure that your class's metadata 'render.modes' key includes
the list of supported modes. It's recommended to call super()
in implementations to use the functionality of this method.
Args:
mode (str): the mode to render with
close (bool): close all open renderings
"""
if close:
if self.viewer is not None:
self.viewer.close()
self.viewer = None # If we don't None out this reference pyglet becomes unhappy
return
if self._last_state is not None:
img = self._last_state['observation']['sensors']['color']['data']
if len(img.shape) == 2: # assume gray
img = np.dstack([img, img, img])
else: # assume rgba
img = img[:, :, :-1]
img = img.reshape((img.shape[1], img.shape[0], img.shape[2]))
if mode == 'human':
from gym.envs.classic_control import rendering
if self.viewer is None:
if self.viewer is None:
self.viewer = rendering.SimpleImageViewer()
self.viewer.imshow(img)
elif mode == 'rgb_array':
return img
def _close(self):
if self._sim is not None:
self._sim.close_game()
class IndoorEnvironment(environment.Environment):
ACTION_LIST = [[1, 0, 0], [0, 1, 0], [0, 0, 1]]
@staticmethod
def get_action_size(env_name):
return len(IndoorEnvironment.ACTION_LIST)
@staticmethod
def get_objective_size(env_name):
simargs = sim_config.get(env_name)
return simargs.get('objective_size', 0)
def __init__(self, env_name, env_args, thread_index):
environment.Environment.__init__(self)
self.i_episode = 0
self.last_state = None
self.last_action = 0
self.last_reward = 0
simargs = sim_config.get(env_name)
simargs['id'] = 'sim%02d' % thread_index
simargs['logdir'] = os.path.join(IndoorEnvironment.get_log_dir(),
simargs['id'])
self.viewer = rendering.SimpleImageViewer()
# Merge in extra env args
if env_args is not None:
simargs.update(env_args)
print(simargs)
self._sim = RoomSimulator(simargs)
self._sim_obs_space = self._sim.get_observation_space(
simargs['outputs'])
self.reset()
def render(self, img):
img = img[:, :, :-1]
img = img.reshape((img.shape[1], img.shape[0], img.shape[2]))
img = cv2.resize(img, (512, 512), cv2.INTER_CUBIC)
self.viewer.imshow(img)
time.sleep(.1)
def reset(self):
result = self._sim.reset()
self._episode_info = result.get('episode_info')
self._last_full_state = result.get('observation')
obs = self._last_full_state['observation']['sensors']['color']['data']
# self.render(obs)
objective = self._last_full_state.get('measurements')
state = {'image': self._preprocess_frame(obs), 'objective': objective}
self.last_state = state
self.last_action = 0
self.last_reward = 0
# self.i_episode = self.i_episode + 1
# print("Saving episode {}".format(self.i_episode))
# self.directory = "./{}".format(self.i_episode)
# os.mkdir(self.directory)
# with open(os.path.join(self.directory, "episode_info.txt"), "w") as outfile:
# json.dump(self._episode_info, outfile, indent=4, cls=NumpyEncoder)
# self.i = 0
def stop(self):
if self._sim is not None:
self._sim.close_game()
def _preprocess_frame(self, image):
if len(image.shape) == 2: # assume gray
image = np.dstack([image, image, image])
else: # assume rgba
image = image[:, :, :-1]
image = image.astype(np.float32)
image = image / 255.0
return image
def process(self, action):
real_action = IndoorEnvironment.ACTION_LIST[action]
full_state = self._sim.step(real_action)
self._last_full_state = full_state # Last observed state
obs = full_state['observation']['sensors']['color']['data']
# self.render(obs)
# depth = full_state['observation']['sensors']['depth']['data']
# Image.fromarray(obs.astype('uint8')).save(os.path.join(self.directory, 'color{}.png'.format(self.i)))
# Image.fromarray(depth, 'L').save(os.path.join(self.directory, 'depth{}.png'.format(self.i)))
# self.i+=1
reward = full_state['rewards']
terminal = full_state['terminals']
success = full_state['success']
objective = full_state.get('measurements')
if not terminal:
state = {
'image': self._preprocess_frame(obs),
'objective': objective
}
else:
state = self.last_state
pixel_change = self._calc_pixel_change(state['image'],
self.last_state['image'])
self.last_state = state
self.last_action = action
self.last_reward = reward
return state, reward, terminal, pixel_change, success
def is_all_scheduled_episodes_done(self):
return self._sim.is_all_scheduled_episodes_done()