a = randaction
#print(" - selecting generated optimal policy ",a)
# for i in range (np.alen(a)):
# if a[i] < -1: a[i]=-0.99999999999
# if a[i] > 1: a[i] = 0.99999999999
if step % 50 == 0:
print("a =>", a)
env.render()
env.refresh(render=True)
qs_a = np.concatenate((qs, a), axis=0)
#get the target state and reward
s, r, done, info = env.step(a)
#record only the first x number of states
#if done and step<max_steps-3:
# r = -50
if step == 0:
gameSA[0] = qs_a
gameS[0] = qs
gameR[0] = np.array([r])
gameA[0] = np.array([r])
gameW[0] = np.array([0.000000005])
else:
gameSA = np.vstack((gameSA, qs_a))
gameS = np.vstack((gameS, qs))
gameR = np.vstack((gameR, np.array([r])))
initial_epsilon = 0.8
)
observation = env.reset()
left_or_right_barge_movement = np.random.randint(0, 2)
epsilon = 0.05
for episode in range(EPISODES):
while True:
# 1. Choose an action based on observation
action = PG.choose_action(observation)
# 2. Take action in the environment
observation_, reward, done, info = env.step(action)
# 3. Store transition for training
# if reward > -0.20:
PG.store_transition(observation, action, reward)
if RENDER_ENV:
# -------------------------------------
# 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.