cumulated_reward = 0 #Should going forward give more reward then L/R ?
print ("Episode = "+str(x))
observation = env.reset()
if qlearn.epsilon > 0.05:
qlearn.epsilon *= epsilon_discount
#render()
env.render()
state = ''.join(map(str, observation))
for i in range(1000):
# Pick an action based on the current state
action = qlearn.chooseAction(state)
#print ("Action Chosen"+str(action))
# Execute the action and get feedback
observation, reward, done, info = env.step(action)
cumulated_reward += reward
#print ("Reward="+str(reward))
if highest_reward < cumulated_reward:
highest_reward = cumulated_reward
nextState = ''.join(map(str, observation))
qlearn.learn(state, action, reward, nextState)
#env.monitor.flush(force=True)
if not(done):
# for each episode, we test the robot for nsteps
for t in itertools.count():
state_ = np.zeros(2)
#state_[0] = int(np.digitize(state[2],vertical_bin))# z first
state_[0] = int(np.digitize(state[0], horizontal_bins[0]))
state_[1] = int(np.digitize(state[1], horizontal_bins[1]))
#Clip the state
for j in range(2):
if state_[j] < 0:
state_[j] = 0
elif state_[j] > env.shape[j] - 1:
state_[j] = env.shape[j] - 1
# Pick an action based on the current state
action = qlearn.chooseAction(tuple(state_))
# Execute the action in the environment and get feedback
next_state, reward, done, info = env.step(action)
cumulated_reward += reward
next_state_ = np.zeros(2)
#next_state_[0] = int(np.digitize(next_state[2],vertical_bin)) # z first
next_state_[0] = int(np.digitize(next_state[0],
horizontal_bins[0]))
next_state_[1] = int(np.digitize(next_state[1],
horizontal_bins[1]))
for j in range(2):
if next_state_[j] < 0:
next_state_[j] = 0
for x in range(total_episodes):
done = False
cumulated_reward = 0 #Should going forward give more reward then L/R ?
observation = env.reset()
if qlearn.epsilon > 0.05:
qlearn.epsilon *= epsilon_discount
step_counter = 0
for i in range(1500):
step_counter += 1
binarized_observation = binarize_observation(observation)
action = qlearn.chooseAction(binarized_observation)
newObservation, reward, done, info = env.step(action)
binarized_new_observation = binarize_observation(newObservation)
cumulated_reward += reward
if highest_reward < cumulated_reward:
highest_reward = cumulated_reward
qlearn.learn(binarized_observation, action, reward,
binarized_new_observation)
if not (done):
observation = newObservation
else:
last_time_steps = numpy.append(last_time_steps, [int(i + 1)])
break
for t in range(total_steps):
step_counter += 1
[image, laser] = observation
hint_pos = getTargetPoints(image)
if hint_pos == 0:
hint_pos = getHintPoints(image)
height, width, depth = image.shape
image_action = get_image_action(width, hint_pos)
laser_action = get_expert_action(laser)
state = threshold_laser(laser, threshold_value)
rl_choice = qlearn.chooseAction(state)
if rl_choice == 1:
action = 20 - image_action
else:
action = laser_action
next_observation, reward, done, info = env.step(action)
cumulated_reward += reward
next_image, next_laser = next_observation
next_state = threshold_laser(next_laser, threshold_value)
qlearn.learn(state, rl_choice, reward, next_state)
if not done: