def __init__(self,
num_of_cars,
lanes,
length,
speed_limit,
init_pos,
init_lane,
ego_speed_init,
dt,
r_seed,
x_range=10):
self.n_lanes = lanes
self.n_cars = num_of_cars
self.x_range = x_range
self.length = length
self.speed_limit = speed_limit / 3.6
self.T = 1.5
self.s0 = self.T * self.speed_limit
self.delta = 4
self.road_width = 4
self.lane_init = (init_lane -
1) * self.road_width + self.road_width * 0.5
self.dt = dt
random.seed(r_seed)
self.vehicle_list = []
# Ego Car
v_init = random.random() * (ego_speed_init / 3.6)
self.x_base = 0
self.vehicle_list.append(
car.Car(init_pos, self.lane_init, v_init, 0, speed_limit, self.dt))
self.dist_to_front = -1
self.non_ego_limit = self.speed_limit * 0.4
### RL Params
self.done = False
self.success = False
self.reward = 0
self.timestep = 0
self.lane = self.lane_init
self.lane_prev = self.lane_init
self.x_goal = 3
self.steering = 0
self.lateral_controller = lateral_agent.lateral_control(dt)
s = set()
while len(s) < num_of_cars:
s.add(np.random.choice(np.arange(0.3, 1, 0.05)) * self.length)
alist = list(s)
for n in range(0, self.n_cars):
y_random = random.randint(
0, self.n_lanes - 2) * self.road_width + self.road_width * 0.5
x_random = alist[n]
v_random = random.random() * self.non_ego_limit
self.vehicle_list.append(
car.Car(x_random, y_random, v_random, 0, speed_limit, self.dt))
def __init__(self,num_of_cars, lanes,length,speed_limit,init_pos,init_lane,ego_speed_init,dt,r_seed,x_range=10):
self.n_lanes = lanes
self.n_cars = num_of_cars
self.x_range = x_range
self.length = length
self.speed_limit = speed_limit/3.6
self.T = 2
self.s0 = self.T * self.speed_limit
self.delta = 4
self.road_width = 4
self.lane_init = (init_lane-1)*self.road_width + self.road_width*0.5
self.dt = dt
random.seed(r_seed)
self.vehicle_list = []
# Ego Car
v_init = random.random() * (ego_speed_init/3.6)
self.x_base = 0
self.vehicle_list.append(car.Car(init_pos,self.lane_init,v_init,0,speed_limit,self.dt))
### RL Params
self.done = False
self.success = False
self.reward = 0
self.timestep = 0
self.lane = self.lane_init
self.lane_prev = self.lane_init
self.x_goal = 3
self.lateral_controller = lateral_agent.lateral_control(dt)
s = set()
while len(s) < num_of_cars:
s.add(np.random.choice(np.arange(0.3,1,0.05))*self.length)
alist = list(s)
#print("Succesfully generated ",num_of_cars, "unique random numbers")
for n in range(0,self.n_cars):
y_random = random.randint(0,self.n_lanes-2)*self.road_width + self.road_width*0.5
#x_random = np.random.choice(np.arange(0.4,2,0.1))*self.length
x_random = alist[n]
v_random = random.random()*self.speed_limit*0.4
yaw_init = 0 #if lane change scenario
self.vehicle_list.append(car.Car(x_random,y_random,v_random,0,speed_limit,self.dt))
estep = 100000
#### Learning Parameters ####
max_train_episodes = 100000
pre_train_steps = 100000
random_sweep = 10
tau = 1
#### Environment ####
done = False
dt = 0.1
timestep = 0
lateral_controller = lateral_agent.lateral_control(dt)
env = world.World(num_of_cars, num_of_lanes, track_length, speed_limit,
ego_pos_init, ego_lane_init, ego_speed_init, dt, random_seed,
x_range)
goal_lane = (ego_lane_init - 1) * env.road_width + env.road_width * 0.5
goal_lane_prev = goal_lane
action = np.zeros(1) # acc/steer
tf.reset_default_graph()
mainQN = q_learning.qnetwork(input_dim, output_dim, hidden_units, layers,
learning_rate, clip_value)
targetQN = q_learning.qnetwork(input_dim, output_dim, hidden_units, layers,
learning_rate, clip_value)
init = tf.global_variables_initializer()
def __init__(self,
num_of_cars,
lanes,
length,
speed_limit,
init_pos,
init_lane,
ego_speed_init,
dt,
r_seed,
x_range=10):
self.n_lanes = lanes
self.n_cars = num_of_cars
self.x_range = x_range
self.length = length
self.speed_limit = speed_limit / 3.6
self.T = 1.5
self.s0 = self.T * self.speed_limit
self.delta = 4
self.road_width = 4
self.lane_init = (init_lane -
1) * self.road_width + self.road_width * 0.5
self.dt = dt
random.seed(r_seed)
self.vehicle_list = []
# Ego Car
v_init = random.random() * (ego_speed_init / 3.6)
self.x_base = 0
self.vehicle_list.append(
car.Car(init_pos, self.lane_init, v_init, 0, speed_limit, self.dt))
self.dist_to_front = -1
self.non_ego_limit = self.speed_limit * 0.4
#POMDP Params
self.x_view = 100
self.y_view = 5
self.observation_grid = np.zeros(
(self.x_view * 2 + 1, self.y_view * 2 + 1, 2))
### RL Params
self.done = False
self.success = False
self.reward = 0
self.timestep = 0
self.lane = self.lane_init
self.lane_prev = self.lane_init
self.x_goal = 3
self.steering = 0
self.lateral_controller = lateral_agent.lateral_control(dt)
s = set()
while len(s) < num_of_cars:
s.add(np.random.choice(np.arange(0.3, 1, 0.05)) * self.length)
alist = list(s)
prev = 0
for n in range(0, self.n_cars):
y_random = random.randint(
0, self.n_lanes - 1) * self.road_width + self.road_width * 0.5
if prev == y_random and y_random == 2:
y_random = 6
elif (prev == y_random and y_random == 6):
y_random = 2
x_random = alist[n]
v_random = random.random() * self.non_ego_limit
#print(v_random)
if y_random == 6:
self.vehicle_list.append(
car.Car(x_random + random.random() * 300, y_random,
-v_random, 0, -speed_limit, self.dt))
elif (y_random == 2):
self.vehicle_list.append(
car.Car(x_random, y_random, v_random, 0, speed_limit,
self.dt))
prev = y_random
def __init__(self,
num_of_cars,
lanes,
length,
speed_limit,
init_pos,
init_lane,
ego_speed_init,
dt,
r_seed,
x_range=10):
self.n_lanes = lanes
self.n_cars = num_of_cars
self.x_range = x_range
self.length = length
self.speed_limit = speed_limit / 3.6
self.T = 1.5
self.s0 = self.T * self.speed_limit
self.delta = 4
self.road_width = 4
self.lane_init = (init_lane -
1) * self.road_width + self.road_width * 0.5
self.dt = dt
random.seed(r_seed)
self.vehicle_list = []
# Ego Car
v_init = random.random() * (ego_speed_init / 3.6)
self.x_base = 0
self.vehicle_list.append(
car.Car(init_pos, self.lane_init, v_init, 0, speed_limit, self.dt))
self.dist_to_front = -1
self.non_ego_limit = self.speed_limit * 0.4
#POMDP Params
self.x_view = 150
self.y_view = 5
self.resolution = 10 # 1m = resolution points
self.observation_grid = np.zeros(
(self.x_view * 2 + 1, self.y_view * 2, 2))
# Car Dims
self.car_x_dim = 4 * self.resolution # 1x = 1m
self.car_y_dim = 2 * self.resolution # 2y = 1m
### RL Params
self.done = False
self.success = False
self.reward = 0
self.timestep = 0
self.lane = self.lane_init
self.lane_prev = self.lane_init
self.x_goal = 3
### Occlusion stuff
x_dim = (self.x_view * 2 + 1) * self.resolution
y_dim = (self.y_view * 2 + 1) * self.resolution
self.obs_body = np.zeros((x_dim, y_dim))
self.grid = np.zeros((x_dim, y_dim))
self.lateral_controller = lateral_agent.lateral_control(dt)
### Debug
self.x_data = []
self.y_data = []
s = set()
while len(s) < num_of_cars:
s.add(np.random.choice(np.arange(0.3, 1, 0.05)) * self.length)
alist = list(s)
for n in range(0, self.n_cars):
y_random = random.randint(
0, self.n_lanes - 2) * self.road_width + self.road_width * 0.5
x_random = alist[n]
v_random = random.random() * self.non_ego_limit
self.vehicle_list.append(
car.Car(x_random, y_random, v_random, 0, speed_limit, self.dt))