for episode in range(episode_number):
while (1):
a = pid.pid_algorithm(
s) # pass the state to the algorithm, get the actions
# Step through the simulation (1 step). Refer to Simulation Update in constants.py
s, r, done, info = env._step(a)
total_reward += r # Accumulate reward
# -------------------------------------
# Optional render
env._render()
# Draw the target
env.draw_marker(env.landing_coordinates[0],
env.landing_coordinates[1])
# Refresh render
env.refresh(render=False)
# When should the barge move? Water movement, dynamics etc can be simulated here.
if s[LEFT_GROUND_CONTACT] == 0 and s[RIGHT_GROUND_CONTACT] == 0:
env.move_barge_randomly(epsilon, left_or_right_barge_movement)
# Random Force on rocket to simulate wind.
env.apply_random_x_disturbance(
epsilon=0.005, left_or_right=left_or_right_barge_movement)
env.apply_random_y_disturbance(epsilon=0.005)
# Touch down or pass abs(THETA_LIMIT)
if done:
print('Episode:\t{}\tTotal Reward:\t{}'.format(
episode, total_reward))
total_reward = 0
env._reset()
last_best_noisy_game = -1000
max_game_average = -1000
noisy_game_no_longer_valid = False
#Create testing enviroment
settings = {
'Side Engines': True,
'Clouds': True,
'Vectorized Nozzle': True,
'Starting Y-Pos Constant': 1,
'Initial Force': 'random'
} # (6000, -10000)}
env = RocketLander(settings)
env.refresh(render=True)
env.render(mode="human")
env.reset()
print("-- Observations", env.observation_space)
print("-- actionspace", env.action_space)
#initialize training matrix with random states and actions
#dataX = np.random.random(( 5,num_env_variables+num_env_actions ))
dataX = tf.placeholder("float", [None, num_env_variables + num_env_actions])
#Only one output for the total score / reward
#dataY = np.random.random((5,1))
dataY = tf.placeholder("float", [None, 1])
#initialize training matrix with random states and actions