def __init__(self):
env_input_size: int = 6
super().__init__(env_input_size)
self.post_fn_remote = self.post_milp
self.get_nn_fn = self.get_nn
self.plot_fn = self.plot
self.template_2d: np.ndarray = np.array([[1, 0, 0, 0, 0, 0],
[1, -1, 0, 0, 0, 0]])
input_boundaries, input_template = self.get_template(0)
self.input_boundaries: List = input_boundaries
self.input_template: np.ndarray = input_template
_, template = self.get_template(1)
self.analysis_template: np.ndarray = template
collision_distance = 0
distance = [Experiment.e(6, 0) - Experiment.e(6, 1)]
# self.use_bfs = True
# self.n_workers = 1
self.rounding_value = 2**10
self.use_rounding = False
self.time_horizon = 40000
self.unsafe_zone: List[Tuple] = [(distance,
np.array([collision_distance]))]
self.input_epsilon = 0
# self.nn_path = os.path.join(utils.get_save_dir(),"tune_PPO_stopping_car/PPO_StoppingCar_acc24_00000_0_cost_fn=0,epsilon_input=0_2021-01-21_02-30-49/checkpoint_39/checkpoint-39")
self.nn_path = os.path.join(
utils.get_save_dir(),
"tune_PPO_stopping_car/PPO_StoppingCar_acc24_00001_1_cost_fn=0,epsilon_input=0_2021-01-21_02-30-49/checkpoint_58/checkpoint-58"
)
def get_template(self, mode=0):
p = Experiment.e(self.env_input_size, 0)
v = Experiment.e(self.env_input_size, 1)
if mode == 0: # box directions with intervals
# input_boundaries = [0, 0, 10, 10]
input_boundaries = [9, -8, 0, 0.1]
# optimise in a direction
template = []
for dimension in range(self.env_input_size):
template.append(Experiment.e(self.env_input_size, dimension))
template.append(-Experiment.e(self.env_input_size, dimension))
template = np.array(template) # the 6 dimensions in 2 variables
return input_boundaries, template
if mode == 1: # directions to easily find fixed point
input_boundaries = None
template = np.array([v + p, -v - p, -p])
return input_boundaries, template
def __init__(self):
env_input_size: int = 2
super().__init__(env_input_size)
self.post_fn_remote = self.post_milp
self.get_nn_fn = self.get_nn
self.plot_fn = self.plot
self.template_2d: np.ndarray = np.array([[0, 1], [1, 0]])
input_boundaries, input_template = self.get_template(0)
self.input_boundaries: List = input_boundaries
self.input_template: np.ndarray = input_template
_, template = self.get_template(0)
self.analysis_template: np.ndarray = template
self.time_horizon = 500
self.rounding_value = 2**8
p = Experiment.e(self.env_input_size, 0)
v = Experiment.e(self.env_input_size, 1)
self.unsafe_zone: List[Tuple] = [([p, -v, v], np.array([0, 1, 0]))]
self.nn_path = os.path.join(
utils.get_save_dir(),
"tune_PPO_bouncing_ball/PPO_BouncingBall_c7326_00000_0_2021-01-16_05-43-36/checkpoint_36/checkpoint-36"
)
def get_template(self, mode=0):
x = Experiment.e(self.env_input_size, 0)
x_dot = Experiment.e(self.env_input_size, 1)
theta = Experiment.e(self.env_input_size, 2)
theta_dot = Experiment.e(self.env_input_size, 3)
if mode == 0: # box directions with intervals
# input_boundaries = [0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05]
input_boundaries = [0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05, 0.05]
# input_boundaries = [0.04373426, -0.04373426, -0.04980056, 0.04980056, 0.045, -0.045, -0.51, 0.51]
# optimise in a direction
template = []
for dimension in range(self.env_input_size):
template.append(Experiment.e(self.env_input_size, dimension))
template.append(-Experiment.e(self.env_input_size, dimension))
template = np.array(template) # the 6 dimensions in 2 variables
return input_boundaries, template
if mode == 1: # directions to easily find fixed point
input_boundaries = None
template = np.array([
theta, -theta, theta_dot, -theta_dot, theta + theta_dot,
-(theta + theta_dot), (theta - theta_dot), -(theta - theta_dot)
]) # x_dot, -x_dot,theta_dot - theta
return input_boundaries, template
if mode == 2:
input_boundaries = None
template = np.array([theta, -theta, theta_dot, -theta_dot])
return input_boundaries, template
if mode == 3:
input_boundaries = None
template = np.array([theta, theta_dot, -theta_dot])
return input_boundaries, template
if mode == 4:
input_boundaries = [0.09375, 0.625, 0.625, 0.0625, 0.1875]
# input_boundaries = [0.09375, 0.5, 0.5, 0.0625, 0.09375]
template = np.array([
theta, theta_dot, -theta_dot, theta + theta_dot,
(theta - theta_dot)
])
return input_boundaries, template
if mode == 5:
input_boundaries = [0.125, 0.0625, 0.1875]
template = np.array(
[theta, theta + theta_dot, (theta - theta_dot)])
return input_boundaries, template
def get_template(self, mode=0):
x_lead = Experiment.e(6, 0)
x_ego = Experiment.e(6, 1)
v_lead = Experiment.e(6, 2)
v_ego = Experiment.e(6, 3)
a_lead = Experiment.e(6, 4)
a_ego = Experiment.e(6, 5)
if mode == 0: # box directions with intervals
input_boundaries = [
50, -40, 10, -0, 28, -28, 36, -36, 0, -0, 0, -0, 0
]
# optimise in a direction
template = []
for dimension in range(6):
template.append(Experiment.e(6, dimension))
template.append(-Experiment.e(6, dimension))
template = np.array(template) # the 6 dimensions in 2 variables
# t1 = [0] * 6
# t1[0] = -1
# t1[1] = 1
# template = np.vstack([template, t1])
return input_boundaries, template
if mode == 1: # directions to easily find fixed point
input_boundaries = [20]
template = np.array([
a_lead, -a_lead, a_ego, -a_ego, -v_lead, v_lead,
-(v_lead - v_ego), (v_lead - v_ego), -(x_lead - x_ego),
(x_lead - x_ego)
])
return input_boundaries, template
if mode == 2:
input_boundaries = [
0, -100, 30, -31, 20, -30, 0, -35, 0, -0, -10, -10, 20
]
# optimise in a direction
template = []
for dimension in range(6):
t1 = [0] * 6
t1[dimension] = 1
t2 = [0] * 6
t2[dimension] = -1
template.append(t1)
template.append(t2)
# template = np.array([[0, 1], [1, 1], [1, 0], [1, -1], [0, -1], [-1, -1], [-1, 0], [-1, 1]]) # the 8 dimensions in 2 variables
template = np.array(template) # the 6 dimensions in 2 variables
t1 = [0] * 6
t1[0] = 1
t1[1] = -1
template = np.vstack([template, t1])
return input_boundaries, template
if mode == 3: # single point box directions +diagonal
input_boundaries = [
30, -30, 0, -0, 28, -28, 36, -36, 0, -0, 0, -0, 0
]
# optimise in a direction
template = []
for dimension in range(6):
t1 = [0] * 6
t1[dimension] = 1
t2 = [0] * 6
t2[dimension] = -1
template.append(t1)
template.append(t2)
# template = np.array([[0, 1], [1, 1], [1, 0], [1, -1], [0, -1], [-1, -1], [-1, 0], [-1, 1]]) # the 8 dimensions in 2 variables
template = np.array(template) # the 6 dimensions in 2 variables
t1 = [0] * 6
t1[0] = -1
t1[1] = 1
template = np.vstack([template, t1])
return input_boundaries, template
if mode == 4: # octagon, every pair of variables
input_boundaries = [20]
template = []
for dimension in range(6):
t1 = [0] * 6
t1[dimension] = 1
t2 = [0] * 6
t2[dimension] = -1
template.append(t1)
template.append(t2)
for other_dimension in range(dimension + 1, 6):
t1 = [0] * 6
t1[dimension] = 1
t1[other_dimension] = -1
t2 = [0] * 6
t2[dimension] = -1
t2[other_dimension] = 1
t3 = [0] * 6
t3[dimension] = 1
t3[other_dimension] = 1
t4 = [0] * 6
t4[dimension] = -1
t4[other_dimension] = -1
template.append(t1)
template.append(t2)
template.append(t3)
template.append(t4)
return input_boundaries, np.array(template)