i = 0
reward = None
while not done:
#print(env.t)
#env.render()
# For graph
add2loc_map(env)
#print(s)
s_ = env.get_state(env.t_start + i + 1)
v = s_[env.num_leading_cars * 3 + 0]
x = s_[env.num_leading_cars * 3 + 1]
a = s_[env.num_leading_cars * 3 + 2]
#print(s_)
#print(v)
#print(a)
s, reward, done, info = env.step(a, human=True)
#print(s)
#print(reward)
#print()
i = i + 1
print(reward)
rewards.append(reward)
print(i)
print("HUMAN")
print("Average Reward:", np.mean(rewards))
print("Median Reward:", np.median(rewards))
print("SE Reward:", np.std(rewards) / (len(rewards))**0.5)
#print(rewards)
plt.hist(rewards, bins='auto')
plt.show()
reward = None
t = 0
while True:
with torch.no_grad():
#env.render()
window.appendleft(torch.Tensor(state))
action_probs = agent(deque2state(env)).detach().numpy()
action = np.argmax(action_probs)
a = (env.a_max - env.a_min) * (
(action) / (agent.action_size - 1)) + env.a_min
#for i in range(agent.action_size):
#print((env.a_max - env.a_min)*((i)/(agent.action_size - 1)) + env.a_min)
#quit()
#print(a)
#input()
next_state, reward, done, _ = env.step(a)
# For Graph
add2loc_map(env)
#
state = next_state
t = t + 1
if done:
break
print(t)
print(reward)
rewards.append(reward)
print("Average Reward:", np.mean(rewards))
print("SE Reward:", np.std(rewards) / (len(rewards))**0.5)
quit()
#print(rewards)
env.normalize = False
#
done = 0
i = 0
reward = None
while not done:
#print(env.t)
#env.render()
# For graph
add2loc_map(env)
#print(s)
v, x, a = env.CACC(s,env.num_leading_cars)
#print(v)
#print(a)
s, reward, done, info = env.step(a,controller="CACC")
#print(reward)
#print()
i = i + 1
#print(reward)
results.append(env.results)
#print(i)
# For CACC
#create_loc_map(env)
results_CACC = results
print("DONE with CACC")
#################################
#action_probs = agent(deque2state(env)).detach().numpy()
#action = np.argmax(action_probs)
#a = (env.a_max - env.a_min)*((action)/(agent.action_size - 1)) + env.a_min
#for i in range(agent.action_size):
#print((env.a_max - env.a_min)*((i)/(agent.action_size - 1)) + env.a_min)
#quit()
#print(a)
#input()
#v, x, a = env.CACC(state,env.num_leading_cars)
a = env.get_state(env.t_start +
env.t)[3 * num_leading_vehicle + 2]
Replay_Buffer.appendleft([window, a])
#print(v)
next_state, reward, done, _ = env.step(a, human=True)
# For Graph
add2loc_map(env)
#
state = next_state
t = t + 1
if done:
break
#print(t)
#print(reward)
rewards.append(reward)
if (len(Replay_Buffer) > 2048):
mimic_optimize(env, agent, Replay_Buffer, 2048)
print("DONE")
acc = 0
env.normalize = False
#
done = 0
i = 0
reward = None
while not done:
#print(env.t)
#env.render()
# For graph
add2loc_map(env)
#print(s)
v, x, a = env.CACC(s, env.num_leading_cars)
#print(v)
#a = min(-2,max(a,2))
s, reward, done, info = env.step(a, controller="CACC")
#print(reward)
#print()
i = i + 1
print(reward)
rewards.append(reward)
print(i)
print("CACC")
print("Average Reward:", np.mean(rewards))
print("Median Reward:", np.median(rewards))
print("SE Reward:", np.std(rewards) / (len(rewards))**0.5)
#print(rewards)
plt.hist(rewards, bins='auto')
plt.show()