def calculate_score_for_leaderboard():
"""
Evaluate the performance of the network. This is the function to be used for
the final ranking on the course-wide leader-board, only with a different set
of seeds. Better not change it.
"""
# action variables
a = np.array([0.0, 0.0, 0.0])
# init environement
env = CarRacing()
env.render()
env.reset()
seeds = [
22597174, 68545857, 75568192, 91140053, 86018367, 49636746, 66759182,
91294619, 84274995, 31531469
]
total_reward = 0
for episode in range(10):
env.seed(seeds[episode])
observation = env.reset()
# init modules of the pipeline
LD_module = LaneDetection(gradient_threshold=25, spline_smoothness=20)
LatC_module = LateralController(gain_constant=1.8,
damping_constant=0.05)
LongC_module = LongitudinalController(KD=0.001)
reward_per_episode = 0
for t in range(600):
# perform step
s, r, done, speed, info = env.step(a)
# lane detection
lane1, lane2 = LD_module.lane_detection(s)
# waypoint and target_speed prediction
waypoints = waypoint_prediction(lane1, lane2)
target_speed = target_speed_prediction(waypoints,
max_speed=60,
exp_constant=6)
# control
a[0] = LatC_module.stanley(waypoints, speed)
a[1], a[2] = LongC_module.control(speed, target_speed)
# reward
reward_per_episode += r
env.render()
print('episode %d \t reward %f' % (episode, reward_per_episode))
total_reward += np.clip(reward_per_episode, 0, np.infty)
print('---------------------------')
print(' total score: %f' % (total_reward / 10))
print('---------------------------')
def multiple_runs(v, on):
env = CarRacing()
z_set = []
action_set = []
for run in range(MAX_RUNS):
zs = []
actions = []
state = env.reset()
env.render() # must have!
# done = False
counter = 0
for game_time in range(MAX_GAME_TIME):
# env.render()
action = generate_action()
obs = state_to_1_batch_tensor(state)
_, _, _, z = v(obs)
z = z.detach().numpy()
z = z.reshape(32)
# print(z.shape)
# if game_time == 5:
# plt.imshow(state)
# plt.show()
# state = _process_frame(state)
# plt.imshow(state)
# plt.show()
zs.append(z)
actions.append(action)
state, r, done, _ = env.step(action)
# print(r)
print(
'RUN:{},GT:{},DATA:{}'.format(
run, game_time, len(actions)
)
)
# if counter == REST_NUM:
#
# position = np.random.randint(len(env.track))
# env.car = Car(env.world, *env.track[position][1:4])
# counter = 0
# counter += 1
zs = np.array(zs, dtype=np.float16)
# print(zs.shape)
actions = np.array(actions, dtype=np.float16)
# print(actions.shape)
# np.save(dst + '/' + save_name, frame_and_action)
# np.savez_compressed(dst + '/' + save_name, action=actions, z=zs)
z_set.append(zs)
action_set.append(actions)
z_set = np.array(z_set)
# print(z_set.shape)
action_set = np.array(action_set)
# print(action_set.shape)
save_name = name_this + '_{}.npz'.format(on)
np.savez_compressed(dst + '/' + save_name, action=action_set, z=z_set)
def game_runner():
from pyglet.window import key
a = np.array([0.0, 0.0, 0.0])
def key_press(k, mod):
global restart
if k == 0xff0d: restart = True
if k == key.LEFT: a[0] = -1.0
if k == key.RIGHT: a[0] = +1.0
if k == key.UP: a[1] = +1.0
if k == key.DOWN: a[2] = +0.8
def key_release(k, mod):
if k == key.LEFT and a[0] == -1.0: a[0] = 0
if k == key.RIGHT and a[0] == +1.0: a[0] = 0
if k == key.UP: a[1] = 0
if k == key.DOWN: a[2] = 0
env = CarRacing()
env.render()
env.viewer.window.on_key_press = key_press
env.viewer.window.on_key_release = key_release
while True:
env.reset()
total_reward = 0.0
steps = 0
restart = False
while True:
s, r, done, info = env.step(a)
total_reward += r
if steps == 900:
print("\n")
print("_______________________________")
print("\n")
print("Human Intelligence Result:")
print("Total Steps: {}".format(steps))
print("Total Reward: {:.0f}".format(total_reward))
print("\n")
print("_______________________________")
print("\n")
break
steps += 1
env.render()
if restart: break
env.monitor.close()
def evaluate():
"""
"""
# action variables
a = np.array([0.0, 0.0, 0.0])
# init environement
env = CarRacing()
env.render()
env.reset()
for episode in range(5):
observation = env.reset()
# init modules of the pipeline
LD_module = LaneDetection()
LatC_module = LateralController()
LongC_module = LongitudinalController()
reward_per_episode = 0
for t in range(500):
# perform step
s, r, done, speed, info = env.step(a)
# lane detection
lane1, lane2 = LD_module.lane_detection(s)
# waypoint and target_speed prediction
waypoints = waypoint_prediction(lane1, lane2)
target_speed = target_speed_prediction(waypoints,
max_speed=60,
exp_constant=4.5)
# control
a[0] = LatC_module.stanley(waypoints, speed)
a[1], a[2] = LongC_module.control(speed, target_speed)
# reward
reward_per_episode += r
env.render()
print('episode %d \t reward %f' % (episode, reward_per_episode))
def run_caracing_by_hunman():
a = np.array([0.0, 0.0, 0.0])
def key_press(k, mod):
global restart
if k == 0xff0d: restart = True
if k == key.LEFT: a[0] = -1.0
if k == key.RIGHT: a[0] = +1.0
if k == key.UP: a[1] = +1.0
if k == key.DOWN:
a[2] = +0.8 # set 1.0 for wheels to block to zero rotation
def key_release(k, mod):
if k == key.LEFT and a[0] == -1.0: a[0] = 0
if k == key.RIGHT and a[0] == +1.0: a[0] = 0
if k == key.UP: a[1] = 0
if k == key.DOWN: a[2] = 0
env = CarRacing()
env.render()
env.viewer.window.on_key_press = key_press
env.viewer.window.on_key_release = key_release
while True:
env.reset()
total_reward = 0.0
steps = 0
restart = False
while True:
s, r, done, info = env.step(a)
total_reward += r
if steps % 200 == 0 or done:
print("\naction " + str(["{:+0.2f}".format(x) for x in a]))
print("step {} total_reward {:+0.2f}".format(
steps, total_reward))
steps += 1
env.render()
if done or restart: break
env.monitor.close()
# define variables
total_reward = 0.0
steps = 0
restart = False
# init modules of the pipeline
LD_module = LaneDetection()
# init extra plot
fig = plt.figure()
plt.ion()
plt.show()
while True:
# perform step
s, r, done, speed, info = env.step(a)
# lane detection
lane1, lane2 = LD_module.lane_detection(s)
# waypoint and target_speed prediction
waypoints = waypoint_prediction(lane1, lane2)
target_speed = target_speed_prediction(waypoints)
# reward
total_reward += r
# outputs during training
if steps % 2 == 0 or done:
print("\naction " + str(["{:+0.2f}".format(x) for x in a]))
print("step {} total_reward {:+0.2f}".format(steps, total_reward))
# if k==key.LEFT and a[0]==-1.0: a[0] = 0
# if k==key.RIGHT and a[0]==+1.0: a[0] = 0
# if k==key.UP: a[1] = 0
# if k==key.DOWN: a[2] = 0
env = CarRacing()
env.render()
# env.viewer.window.on_key_press = key_press
# env.viewer.window.on_key_release = key_release
while True:
env.reset()
total_reward = 0.0
steps = 0
restart = False
angle=[]
while True:
obs, reward, done, info = env.step(action)
total_reward += reward
obs=transform(obs).view(1,3,64,64)
recon_c, mu_c, var_c = model(obs)
mu, sigma = mu_c, var_c
#sigma = torch.exp(sigma/2.0)
epsilon = torch.randn_like(sigma)
z=mu+sigma*epsilon
z=z.cuda().view(obs.shape[0],-1).detach()
action_p=controller(z)
# action=action.view(-1).data.cpu().numpy().astype('float32')
#print(action_p)
action[0]=np.argmax(np.reshape(action_p[0].data.cpu().numpy().astype('float32'),-1))-1
action[1]=np.argmax(np.reshape(action_p[1].data.cpu().numpy().astype('float32'),-1))
action[2]=np.argmax(np.reshape(action_p[2].data.cpu().numpy().astype('float32'),-1))
action=np.array(action)
if k == key.UP: a[1] = +1.0
if k == key.DOWN:
a[2] = +0.8 # set 1.0 for wheels to block to zero rotation
def key_release(k, mod):
if k == key.LEFT and a[0] == -1.0: a[0] = 0
if k == key.RIGHT and a[0] == +1.0: a[0] = 0
if k == key.UP: a[1] = 0
if k == key.DOWN: a[2] = 0
env = CarRacing()
env.render()
env.viewer.window.on_key_press = key_press
env.viewer.window.on_key_release = key_release
while True:
env.reset()
total_reward = 0.0
steps = 0
restart = False
while True:
s, r, done, info = env.step(a)
total_reward += r
if steps % 200 == 0 or done:
print("\naction " + str(["{:+0.2f}".format(x) for x in a]))
print("step {} total_reward {:+0.2f}".format(
steps, total_reward))
steps += 1
env.render()
if done or restart: break
env.monitor.close()
# define variables
total_reward = 0.0
steps = 0
restart = False
# init modules of the pipeline
LD_module = LaneDetection()
# init extra plot
fig = plt.figure()
plt.ion()
plt.show()
while True:
# perform step
s, r, done, speed, _ = env.step(action=a)
# lane detection
splines = LD_module.lane_detection(s)
# reward
total_reward += r
# outputs during training
if steps % 2 == 0 or done:
print("\naction " + str(["{:+0.2f}".format(x) for x in a]))
print("step {} total_reward {:+0.2f}".format(steps, total_reward))
LD_module.plot_state_lane(s, steps, fig)
steps += 1
env.render()
if done or restart: break
from gym.envs.box2d.car_racing import CarRacing
from tqdm import tqdm
import numpy as np
if __name__ == "__main__":
env = CarRacing()
env.reset()
a = np.zeros(3)
for _ in tqdm(range(1000000)):
env.step(a)
# env.render()