def add_car(self, x, y, vx, vy, v_des, p_des, direction, theta):
if y < 4.:
idx = self._lower_lane_next_idx
self._lower_lane_next_idx += 1
if self._lower_lane_next_idx > int(self.max_veh_num / 2.):
self._lower_lane_next_idx = 1
elif y > 4.:
idx = self._upper_lane_next_idx
self._upper_lane_next_idx += 1
if self._upper_lane_next_idx > self.max_veh_num:
self._upper_lane_next_idx = int(self.max_veh_num / 2.) + 1
car = Car(idx=idx,
length=self.car_length,
width=self.car_width,
color=random.choice(RED_COLORS),
max_accel=self.car_max_accel,
max_speed=self.car_max_speed,
max_rotation=0.,
expose_level=self.car_expose_level)
driver = YNYDriver(idx=idx,
car=car,
dt=self.dt,
x_driver=IDMDriver(idx=idx,
car=car,
sigma=self.driver_sigma,
s_des=self.s_des,
s_min=self.s_min,
axis=0,
min_overlap=self.min_overlap,
dt=self.dt),
y_driver=PDDriver(idx=idx,
car=car,
sigma=0.,
axis=1,
dt=self.dt))
car.set_position(np.array([x, y]))
car.set_velocity(np.array([vx, vy]))
car.set_rotation(theta)
driver.x_driver.set_v_des(v_des)
driver.x_driver.set_direction(direction)
driver.y_driver.set_p_des(p_des)
if np.random.rand() < self.yld:
driver.set_yld(True)
else:
driver.set_yld(False)
self._cars.append(car)
self._drivers.append(driver)
return car, driver
def __init__(self, v_des, y_des, x_sigma, y_sigma, **kwargs):
# x_driver = PDriver(sigma=x_sigma, v_des=v_des, a_max=1.0, axis=0, **kwargs)
x_driver = IDMDriver(sigma=x_sigma,
v_des=10.0,
s_des=1.0,
s_min=0.5,
axis=0,
min_overlap=0.,
**kwargs)
y_driver = PDDriver(sigma=y_sigma,
p_des=y_des,
a_max=1.0,
axis=1,
**kwargs)
super(EgoDriver, self).__init__(x_driver, y_driver, **kwargs)
def add_car(self, idx, x, y, vx, vy, v_des, p_des, direction, theta):
car = Car(idx=idx,
length=self.car_length,
width=self.car_width,
color=random.choice(RED_COLORS),
max_accel=self.car_max_accel,
max_speed=self.car_max_speed,
expose_level=self.car_expose_level)
driver = TwoTDriver(idx=idx,
car=car,
dt=self.dt,
x_driver=IDMDriver(idx=idx,
car=car,
sigma=self.driver_sigma,
s_min=self.s_min,
axis=0,
min_overlap=self.min_overlap,
dt=self.dt),
y_driver=PDDriver(idx=idx,
car=car,
sigma=0.,
axis=1,
dt=self.dt))
car.set_position(np.array([x, y]))
car.set_velocity(np.array([vx, vy]))
car.set_rotation(theta)
driver.x_driver.set_v_des(v_des)
driver.x_driver.set_direction(direction)
driver.y_driver.set_p_des(p_des)
driver.t1 = np.random.rand()
if np.random.rand() < self.yld:
driver.yld = True
driver.t2 = np.random.rand() * 1.5 - 0.5
else:
driver.yld = False
driver.t2 = None
self._cars.append(car)
self._drivers.append(driver)
return car, driver
def __init__(self, aggressive, x_sigma, y_sigma, car, **kwargs):
self.aggressive = aggressive
self.target_lane = None
self.car = car
if self.aggressive:
v_des = np.random.uniform(0.5, 1.0)
s_des = np.random.uniform(0.8, 1.0)
s_min = np.random.uniform(0.4, 0.6)
min_overlap = -self.car.width / 2.
self.min_front_x = self.car.length + s_min
self.min_back_x = self.car.length + s_min
self.min_advantage = self.car.length / 2.
else:
v_des = np.random.uniform(0.0, 0.5)
s_des = np.random.uniform(0.9, 1.1)
s_min = np.random.uniform(0.5, 0.7)
min_overlap = self.car.width / 2.
self.min_front_x = self.car.length + s_min
self.min_back_x = self.car.length + s_min
self.min_advantage = self.car.length
x_driver = IDMDriver(sigma=x_sigma,
v_des=v_des,
s_des=s_des,
s_min=s_min,
axis=0,
min_overlap=min_overlap,
car=car,
**kwargs)
y_driver = PDDriver(sigma=y_sigma,
p_des=0.,
a_max=1.0,
axis=1,
k_p=2.0,
k_d=5.0,
car=car,
**kwargs)
self.on_target = True
super(EnvDriver, self).__init__(x_driver, y_driver, car=car, **kwargs)
def __init__(self,
yld=0.5,
vs_actions=[0.,0.5,3.],
t_actions=[-1.5,0.,1.5],
desire_speed=3.,
speed_cost=0.01,
t_cost=0.01,
control_cost=0.01,
collision_cost=1.,
outroad_cost=1.,
goal_reward=1.,
road=Road([RoadSegment([(-100.,0.),(100.,0.),(100.,8.),(-100.,8.)]),
RoadSegment([(-2,-10.),(2,-10.),(2,0.),(-2,0.)])]),
n_cars=3,
num_updates=1,
dt=0.1,
**kwargs):
self.yld = yld
self.vs_actions = vs_actions
self.t_actions = t_actions
# we use target value instead of target change so system is Markovian
self.rl_actions = list(itertools.product(vs_actions,t_actions))
self.num_updates = num_updates
self.desire_speed = desire_speed
self.speed_cost = speed_cost
self.t_cost = t_cost
self.control_cost = control_cost
self.collision_cost = collision_cost
self.outroad_cost = outroad_cost
self.goal_reward = goal_reward
self._collision = False
self._outroad = False
self._goal = False
self._intentions = []
car_length=5.0
car_width=2.0
car_max_accel=10.0
car_max_speed=40.0
car_expose_level=4
cars = [Car(idx=cid, length=car_length, width=car_width, color=random.choice(BLUE_COLORS),
max_accel=car_max_accel, max_speed=car_max_speed,
expose_level=car_expose_level) for cid in range(n_cars)
]
driver_sigma = 0.0
s_min = 2.0
min_overlap = 1.0
drivers = []
drivers.append(EgoDriver(trajectory=EgoTrajectory(),idx=0,car=cars[0],dt=dt))
for did in range(1,n_cars):
driver = TwoTDriver(idx=did, car=cars[did], dt=dt,
x_driver=IDMDriver(idx=did, car=cars[did], sigma=driver_sigma, s_min=s_min, axis=0, min_overlap=min_overlap, dt=dt),
y_driver=PDDriver(idx=did, car=cars[did], sigma=driver_sigma, axis=1, dt=dt))
drivers.append(driver)
super(TIntersection, self).__init__(
road=road,
cars=cars,
drivers=drivers,
dt=dt,
**kwargs,)