def make_angular_integral_reward_components(
self) -> Tuple[rewards.RewardComponent, ...]:
ANGLE_DEG_ERROR_SCALING = 0.1
CMD_TRAVEL_MAX = 2 / 4 # a qarter of the max. absolute value of the delta-cmd;
ANGLE_INT_DEG_MAX = self.integral_limit
base_components = (
rewards.AngularAsymptoticErrorComponent(
name='rwd_Angle_error',
prop=self.prop_error,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=ANGLE_DEG_ERROR_SCALING,
weight=6),
rewards.LinearErrorComponent(name='rwd_cmd_travel_error',
prop=self.prop_delta_cmd,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=CMD_TRAVEL_MAX,
weight=4),
rewards.LinearErrorComponent(name='rwd_Angle_error_Integral',
prop=self.prop_error_integral,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=ANGLE_INT_DEG_MAX,
weight=10),
)
return base_components
def make_sideslip_angle_reward_components(
self) -> Tuple[rewards.RewardComponent, ...]:
SIDESLIP_ANGLE_DEG_ERROR_SCALING = 0.5
RUDDER_CMD_TRAVEL_MAX = 2 / 4 # a quarter of the max. absolute value of the delta-cmd; leverage the non-linearities
SIDESLIP_ANGLE_INT_DEG_MAX = self.integral_limit
base_components = (
rewards.AngularAsymptoticErrorComponent(
name='rwd_sideslipAngle_error',
prop=self.prop_error,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=SIDESLIP_ANGLE_DEG_ERROR_SCALING,
weight=4),
rewards.LinearErrorComponent(
name='rwd_cmd_travel_error',
prop=self.prop_delta_cmd,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=RUDDER_CMD_TRAVEL_MAX,
weight=1), #for the rudder, we increase this priority
rewards.LinearErrorComponent(name='rwd_sideslipAngle_error_Integral',
prop=self.prop_error_integral,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=SIDESLIP_ANGLE_INT_DEG_MAX,
weight=1),
)
return base_components
def make_glide_path_angle_reward_components(self):
GLIDE_ANGLE_DEG_ERROR_SCALING = 0.1
GLIDE_ANGLE_INT_DEG_MAX = self.integral_limit
base_components = (
rewards.AngularAsymptoticErrorComponent(
name='rwd_glideAngle_error',
prop=self.prop_error,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=GLIDE_ANGLE_DEG_ERROR_SCALING,
weight=4),
rewards.LinearErrorComponent(name='rwd_glideAngle_error_derivative',
prop=self.prop_error_derivative,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=2,
weight=1),
rewards.LinearErrorComponent(name='rwd_glideAngle_error_Integral',
prop=self.prop_error_integral,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=GLIDE_ANGLE_INT_DEG_MAX,
weight=9),
)
return base_components
def make_glide_angle_reward_components(self):
GLIDE_ANGLE_DEG_ERROR_SCALING = 0.1
GLIDE_ANGLE_INT_DEG_MAX = self.integral_limit
ELEVATOR_CMD_TRAVEL_MAX = 2 / 4 # a quarter of the max. absolute value of the delta-cmd; leverage the non-linearities
GLIDE_ANGULAR_VELOCITY_SCALING = 0.25
base_components = (
rewards.AngularAsymptoticErrorComponent(
name='rwd_glideAngle_error',
prop=self.prop_error,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=GLIDE_ANGLE_DEG_ERROR_SCALING,
weight=4),
rewards.LinearErrorComponent(name='rwd_glideAngle_error_derivative',
prop=self.prop_error_derivative,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=2,
weight=1),
rewards.LinearErrorComponent(name='rwd_glideAngle_error_Integral',
prop=self.prop_error_integral,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=GLIDE_ANGLE_INT_DEG_MAX,
weight=9),
rewards.LinearErrorComponent(name='rwd_elevator_cmd_travel_error',
prop=self.prop_delta_cmd,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=ELEVATOR_CMD_TRAVEL_MAX,
weight=2),
# rewards.LinearErrorComponent(name='rwd_angular_velocity_q_rad',
# prop=prp.q_radps,
# state_variables=self.obs_props,
# target=0.0,
# potential_difference_based=False,
# scaling_factor=GLIDE_ANGULAR_VELOCITY_SCALING,
# weight=1),
)
return base_components
def make_roll_angle_reward_components(
self) -> Tuple[rewards.RewardComponent, ...]:
ROLL_ANGLE_DEG_ERROR_SCALING = 0.5
ROLL_ANGLE_INT_DEG_MAX = self.integral_limit
AILERON_CMD_TRAVEL_MAX = 2 / 4 # a quarter of the max. absolute value of the delta-cmd; leverage the non-linearities
ROLL_ANGULAR_VELOCITY_SCALING = 0.25
base_components = (
rewards.AngularAsymptoticErrorComponent(
name='rwd_rollAngle_error',
prop=self.prop_error,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=ROLL_ANGLE_DEG_ERROR_SCALING,
weight=4),
rewards.LinearErrorComponent(name='rwd_rollAngle_error_Integral',
prop=self.prop_error_integral,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=ROLL_ANGLE_INT_DEG_MAX,
weight=9),
rewards.LinearErrorComponent(name='rwd_aileron_cmd_travel_error',
prop=self.prop_delta_cmd,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=AILERON_CMD_TRAVEL_MAX,
weight=2),
#check if the angular-velocity criterion is more helpful to avoid flittering.
#from a causality point of view, the command travel should be the criterion, as we want to avoid excessive movement.
#The control surface travel (derivative) must be presented to the ANN anyways.
rewards.LinearErrorComponent(
name='rwd_roll_angular_velocity',
prop=prp.p_radps,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=ROLL_ANGULAR_VELOCITY_SCALING,
weight=1),
)
return base_components
def make_elevator_actuation_reward_components(self):
ELEVATOR_CMD_TRAVEL_MAX = 2 / 4 # a quarter of the max. absolute value of the delta-cmd; leverage the non-linearities
base_components = (rewards.LinearErrorComponent(
name='rwd_elevator_cmd_travel_error',
prop=self.prop_delta_cmd,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=ELEVATOR_CMD_TRAVEL_MAX,
weight=2), )
return base_components
def make_roll_angle_integral_reward_components(
self) -> Tuple[rewards.RewardComponent, ...]:
ROLL_ANGLE_DEG_ERROR_SCALING = 0.5
AILERON_CMD_TRAVEL_MAX = 2 / 4 # a qarter of the max. absolute value of the delta-cmd;
ROLL_ANGLE_INT_DEG_MAX = self.integral_limit
ROLL_ANGLE_DER_DEG_MAX = 10
base_components = (
rewards.AngularAsymptoticErrorComponent(
name='rwd_rollAngle_error',
prop=self.prop_error,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=ROLL_ANGLE_DEG_ERROR_SCALING,
weight=4),
rewards.LinearErrorComponent(name='rwd_aileron_cmd_travel_error',
prop=self.prop_delta_cmd,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=AILERON_CMD_TRAVEL_MAX,
weight=2),
rewards.LinearErrorComponent(name='rwd_rollAngle_error_Integral',
prop=self.prop_error_integral,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=ROLL_ANGLE_INT_DEG_MAX,
weight=9),
rewards.QuadraticErrorComponent(
name='rwd_rollAngle_error_Derivative',
prop=self.prop_error_derivative,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=ROLL_ANGLE_DER_DEG_MAX**2,
weight=2),
)
return base_components
def make_roll_angle_error_punish_actuation_reward_components(
self) -> Tuple[rewards.RewardComponent, ...]:
ROLL_ANGLE_DEG_ERROR_SCALING = 0.5
AILERON_CMD_TRAVEL_MAX = 2 # the max. absolute value of the delta-cmd;
base_components = (
rewards.AngularAsymptoticErrorComponent(
name='rwd_rollAngle_error',
prop=self.prop_error,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=ROLL_ANGLE_DEG_ERROR_SCALING,
weight=8),
rewards.LinearErrorComponent(name='rwd_aileron_cmd_travel_error',
prop=self.prop_delta_cmd,
state_variables=self.obs_props,
target=0.0,
potential_difference_based=False,
scaling_factor=AILERON_CMD_TRAVEL_MAX,
weight=2),
)
return base_components