def stepEnv(self, action=None):
# If no manual action was specified by the user
if action is None:
action = random.randint(0, self.env.action_space.n - 1)
action = int(action)
obs, reward, done, info = self.env.step(action)
self.current_actions.append(action)
self.current_images.append(self.lastObs['image'])
self.current_directions.append(self.lastObs['direction'])
self.showEnv(obs)
self.lastObs = obs
if done:
if reward > 0: # i.e. we did not lose
if self.shift < len(self.demos):
self.demos[self.shift] = self.current_demo, self.shift
else:
self.demos.append(
(self.current_mission,
blosc.pack_array(np.array(self.current_images)),
self.current_directions, self.current_actions))
utils.save_demos(self.demos, self.demos_path)
self.missionBox.append('Demonstrations are saved.')
utils.synthesize_demos(self.demos)
self.shift += 1
self.resetEnv()
else:
self.shiftEnv()
def stepEnv(self, action=None):
# If no manual action was specified by the user
if action is None:
action = random.randint(0, self.env.action_space.n - 1)
action = int(action)
obs, reward, done, info = self.env.step(action)
self.current_demo.append((self.lastObs, action, reward, done))
self.showEnv(obs)
self.lastObs = obs
if done:
if reward > 0: # i.e. we did not lose
if self.shift < len(self.demos):
self.demos[self.shift] = self.current_demo
else:
self.demos.append(self.current_demo)
utils.save_demos(self.demos, args.env, "human")
self.missionBox.append('Demonstrations are saved.')
utils.synthesize_demos(self.demos)
self.shift += 1
self.resetEnv()
else:
self.shiftEnv()
def __init__(self, env):
super().__init__()
self.initUI()
# By default, manual stepping only
self.fpsLimit = 0
self.env = env
self.lastObs = None
# Demonstrations
self.demos = utils.load_demos(args.env, "human")
utils.synthesize_demos(self.demos)
self.current_demo = []
self.shift = len(self.demos) if args.shift is None else args.shift
self.shiftEnv()
# Pointing and naming data
self.pointingData = []
def __init__(self, env):
super().__init__()
self.initUI()
# By default, manual stepping only
self.fpsLimit = 0
self.env = env
self.lastObs = None
# Demonstrations
self.demos_path = utils.get_demos_path(args.demos,
args.env,
origin="human",
valid=False)
self.demos = utils.load_demos(self.demos_path, raise_not_found=False)
utils.synthesize_demos(self.demos)
self.shift = len(self.demos) if args.shift is None else args.shift
self.shiftEnv()
# Pointing and naming data
self.pointingData = []
utils.seed(args.seed)
# Generate environment
env = gym.make(args.env)
env.seed(args.seed)
# Define agent
agent = utils.load_agent(args, env)
# Load demonstrations
demos = utils.load_demos(args.env, "agent")
utils.synthesize_demos(demos)
for i in range(1, args.episodes+1):
# Run the expert for one episode
done = False
obs = env.reset()
demo = []
while not(done):
action = agent.get_action(obs)
new_obs, reward, done, _ = env.step(action)
agent.analyze_feedback(reward, done)
demo.append((obs, action, reward, done))
obs = new_obs