def trim_point_test():
flightLimits = FlightLimits()
ctrlLimits = CtrlLimits()
llc = LowLevelController(ctrlLimits)
ap = GcasAutopilot(llc.xequil, llc.uequil, flightLimits, ctrlLimits)
pass_fail = FlightLimitsPFA(flightLimits)
pass_fail.break_on_error = False
# ### Initial Conditions ###
# initialState = [Vt, alpha, beta, phi, theta, psi, 0, 0, 0, 0, 0, alt, power]
initialState = [
502.0, 0.03887505597600522, 0.0, 0.0, 0.03887505597600522, 0.0, 0.0,
0.0, 0.0, 0.0, 0.0, 1000.0, 9.05666543872074
]
# if not None will do animation. Try a filename ending in .gif or .mp4 (slow). Using '' will plot to the screen.
animFilename = 'gcas_stable.gif'
# Select Desired F-16 Plant
f16_plant = 'morelli' # 'stevens' or 'morelli'
tMax = 15 # simulation time
xcg_mult = 1.0 # center of gravity multiplier
val = 1.0 # other aerodynmic coefficient multipliers
cxt_mult = val
cyt_mult = val
czt_mult = val
clt_mult = val
cmt_mult = val
cnt_mult = val
multipliers = (xcg_mult, cxt_mult, cyt_mult, czt_mult, clt_mult, cmt_mult,
cnt_mult)
der_func = lambda t, y: controlledF16(
t, y, f16_plant, ap, llc, multipliers=multipliers)[0]
passed, times, states, modes, ps_list, Nz_list, u_list = RunF16Sim(\
initialState, tMax, der_func, f16_plant, ap, llc, pass_fail, multipliers=multipliers)
print("Simulation Conditions Passed: {}".format(passed))
if animFilename is not None:
plot3d_anim(times,
states,
modes,
ps_list,
Nz_list,
filename=animFilename)
def main():
'main function'
flightLimits = FlightLimits()
ctrlLimits = CtrlLimits()
llc = LowLevelController(ctrlLimits)
ap = GcasAutopilot(llc.xequil, llc.uequil, flightLimits, ctrlLimits)
pass_fail = FlightLimitsPFA(flightLimits)
pass_fail.break_on_error = False
### Initial Conditions ###
power = 9 # Power
# Default alpha & beta
alpha = deg2rad(2.1215) # Trim Angle of Attack (rad)
beta = 0 # Side slip angle (rad)
# Initial Attitude
alt = 3600
Vt = 540 # Pass at Vtg = 540; Fail at Vtg = 550;
phi = (pi / 2) * 0.5 # Roll angle from wings level (rad)
theta = (-pi / 2) * 0.8 # Pitch angle from nose level (rad)
psi = -pi / 4 # Yaw angle from North (rad)
# Build Initial Condition Vectors
# state = [VT, alpha, beta, phi, theta, psi, P, Q, R, pn, pe, h, pow]
initialState = [
Vt, alpha, beta, phi, theta, psi, 0, 0, 0, 0, 0, alt, power
]
# if not None will do animation. Try a filename ending in .gif or .mp4 (slow). Using '' will plot to the screen.
animFilename = 'gcas.gif'
# Select Desired F-16 Plant
f16_plant = 'morelli' # 'stevens' or 'morelli'
tMax = 15 # simulation time
xcg_mult = 1.0 # center of gravity multiplier
val = 1.0 # other aerodynmic coefficient multipliers
cxt_mult = val
cyt_mult = val
czt_mult = val
clt_mult = val
cmt_mult = val
cnt_mult = val
multipliers = (xcg_mult, cxt_mult, cyt_mult, czt_mult, clt_mult, cmt_mult,
cnt_mult)
der_func = lambda t, y: controlledF16(
t, y, f16_plant, ap, llc, multipliers=multipliers)[0]
passed, times, states, modes, ps_list, Nz_list, u_list = RunF16Sim(\
initialState, tMax, der_func, f16_plant, ap, llc, pass_fail, multipliers=multipliers)
print("Simulation Conditions Passed: {}".format(passed))
if animFilename is not None:
plot3d_anim(times,
states,
modes,
ps_list,
Nz_list,
filename=animFilename)
def __init__(self):
# Initial condition
self.power_low = 9
self.power_high = 9
# Default alpha & beta
self.alpha_low = deg2rad(21.215)
self.alpha_high = deg2rad(2.1215)
self.beta_low = 0
self.beta_high = 0
# Initial Attitude
self.alt_low = 3600
self.alt_high = 3600
self.Vt_low = 540
self.Vt_high = 540 # Pass at Vtg = 540; Fail at Vtg = 550;
self.phi_low = (pi / 2) * 0.5
self.phi_high = (pi / 2) * 0.5 # Roll angle from wings level (rad)
self.theta_low = (-pi / 2) * 0.8
self.theta_high = (-pi / 2) * 0.8 # Pitch angle from nose level (rad)
self.psi_low = -pi / 4
self.psi_high = -pi / 4 # Yaw angle from North (rad)
# state = [VT, alpha, beta, phi, theta, psi, P, Q, R, pn, pe, h, pow, Nz, Ps, Ny]
self.state_low = np.array([
self.Vt_low, self.alpha_low, self.beta_low, self.phi_low,
self.theta_low, self.psi_low, 0, 0, 0, 0, 0, self.alt_low,
self.power_low, 0, 0, 0
])
self.state_high = np.array([
self.Vt_high, self.alpha_high, self.beta_high, self.phi_high,
self.theta_high, self.psi_high, 0, 0, 0, 0, 0, self.alt_high,
self.power_high, 0, 0, 0
])
self.initial_space = spaces.Box(self.state_low,
self.state_high,
dtype=float)
# Safety Constrains
self.flightLimits = FlightLimits()
state_high = np.full(len(self.state_low), np.inf)
state_low = np.full(len(self.state_low), np.NINF)
state_low[0] = self.flightLimits.vMin
state_low[1] = self.flightLimits.alphaMinDeg
state_low[2] = -self.flightLimits.betaMaxDeg
state_low[11] = self.flightLimits.altitudeMin
state_high[0] = self.flightLimits.vMax
state_high[1] = self.flightLimits.alphaMaxDeg
state_high[2] = self.flightLimits.betaMaxDeg
state_high[11] = self.flightLimits.altitudeMax
self.observation_space = spaces.Box(
state_low, state_high, dtype=float) # Yaw angle from North (rad)
# control limits
self.ctrlLimits = CtrlLimits()
self.u_low = np.array([
self.ctrlLimits.ThrottleMin, self.ctrlLimits.ElevatorMinDeg,
self.ctrlLimits.AileronMinDeg, self.ctrlLimits.RudderMinDeg
])
self.u_high = np.array([
self.ctrlLimits.ThrottleMax, self.ctrlLimits.ElevatorMaxDeg,
self.ctrlLimits.AileronMaxDeg, self.ctrlLimits.RudderMaxDeg
])
# Stable states
self.xequil = np.array([502.0, 0.03887505597600522, 0.0, 0.0, 0.03887505597600522, 0.0, 0.0, 0.0, \
0.0, 0.0, 0.0, 1000.0, 9.05666543872074], dtype=float).transpose()
self.uequil = np.array(
[0.13946204864060271, -0.7495784725828754, 0.0, 0.0],
dtype=float).transpose()
self.stable_state = np.concatenate((self.xequil, [0.0, 0.0, 0.0]))
self.stable_action = np.concatenate((self.uequil, [0.0, 0.0, 0.0]))
# Select Desired F-16 Plant
self.f16_plant = 'morelli' # 'stevens' or 'morelli'
xcg_mult = 1.0 # center of gravity multiplier
val = 1.0 # other aerodynmic coefficient multipliers
cxt_mult = val
cyt_mult = val
czt_mult = val
clt_mult = val
cmt_mult = val
cnt_mult = val
self.multipliers = (xcg_mult, cxt_mult, cyt_mult, czt_mult, clt_mult,
cmt_mult, cnt_mult)
self.ap = GcasAutopilot(self.xequil, self.uequil, self.flightLimits,
self.ctrlLimits)
self.pass_fail = FlightLimitsPFA(self.flightLimits)
self.sim_step = 0.01
# Q: Penalty on State Error in LQR controller
# These were chosen to try to achieve a natural frequency of 3 rad/sec and a damping ratio (zeta) of 0.707
# see the matlab code for more analysis of the resultant controllers
# state = [VT, alpha, beta, phi, theta, psi, P, Q, R, pn, pe, h, pow, Nz, Ps, Ny]
q_alpha = 1000
q_q = 0
q_Nz = 1500
q_beta = 0
q_p = 0
q_r = 0
q_ps_i = 1200
q_Ny_r_i = 3000
q_list = [
0, q_alpha, q_beta, 0, 0, 0, q_p, q_q, q_r, 0, 0, 0, 0, q_Nz,
q_ps_i, q_Ny_r_i
]
self.Q = np.diag(q_list)
# R: Penalty on Control Effort in LRQ controller
r_elevator = 1
r_aileron = 1
r_rudder = 1
r_list = [0, r_elevator, r_aileron, r_rudder, 0, 0, 0]
self.R = np.diag(r_list)
self.viewer = None
class GymInterface:
def __init__(self):
# Initial condition
self.power_low = 9
self.power_high = 9
# Default alpha & beta
self.alpha_low = deg2rad(21.215)
self.alpha_high = deg2rad(2.1215)
self.beta_low = 0
self.beta_high = 0
# Initial Attitude
self.alt_low = 3600
self.alt_high = 3600
self.Vt_low = 540
self.Vt_high = 540 # Pass at Vtg = 540; Fail at Vtg = 550;
self.phi_low = (pi / 2) * 0.5
self.phi_high = (pi / 2) * 0.5 # Roll angle from wings level (rad)
self.theta_low = (-pi / 2) * 0.8
self.theta_high = (-pi / 2) * 0.8 # Pitch angle from nose level (rad)
self.psi_low = -pi / 4
self.psi_high = -pi / 4 # Yaw angle from North (rad)
# state = [VT, alpha, beta, phi, theta, psi, P, Q, R, pn, pe, h, pow, Nz, Ps, Ny]
self.state_low = np.array([
self.Vt_low, self.alpha_low, self.beta_low, self.phi_low,
self.theta_low, self.psi_low, 0, 0, 0, 0, 0, self.alt_low,
self.power_low, 0, 0, 0
])
self.state_high = np.array([
self.Vt_high, self.alpha_high, self.beta_high, self.phi_high,
self.theta_high, self.psi_high, 0, 0, 0, 0, 0, self.alt_high,
self.power_high, 0, 0, 0
])
self.initial_space = spaces.Box(self.state_low,
self.state_high,
dtype=float)
# Safety Constrains
self.flightLimits = FlightLimits()
state_high = np.full(len(self.state_low), np.inf)
state_low = np.full(len(self.state_low), np.NINF)
state_low[0] = self.flightLimits.vMin
state_low[1] = self.flightLimits.alphaMinDeg
state_low[2] = -self.flightLimits.betaMaxDeg
state_low[11] = self.flightLimits.altitudeMin
state_high[0] = self.flightLimits.vMax
state_high[1] = self.flightLimits.alphaMaxDeg
state_high[2] = self.flightLimits.betaMaxDeg
state_high[11] = self.flightLimits.altitudeMax
self.observation_space = spaces.Box(
state_low, state_high, dtype=float) # Yaw angle from North (rad)
# control limits
self.ctrlLimits = CtrlLimits()
self.u_low = np.array([
self.ctrlLimits.ThrottleMin, self.ctrlLimits.ElevatorMinDeg,
self.ctrlLimits.AileronMinDeg, self.ctrlLimits.RudderMinDeg
])
self.u_high = np.array([
self.ctrlLimits.ThrottleMax, self.ctrlLimits.ElevatorMaxDeg,
self.ctrlLimits.AileronMaxDeg, self.ctrlLimits.RudderMaxDeg
])
# Stable states
self.xequil = np.array([502.0, 0.03887505597600522, 0.0, 0.0, 0.03887505597600522, 0.0, 0.0, 0.0, \
0.0, 0.0, 0.0, 1000.0, 9.05666543872074], dtype=float).transpose()
self.uequil = np.array(
[0.13946204864060271, -0.7495784725828754, 0.0, 0.0],
dtype=float).transpose()
self.stable_state = np.concatenate((self.xequil, [0.0, 0.0, 0.0]))
self.stable_action = np.concatenate((self.uequil, [0.0, 0.0, 0.0]))
# Select Desired F-16 Plant
self.f16_plant = 'morelli' # 'stevens' or 'morelli'
xcg_mult = 1.0 # center of gravity multiplier
val = 1.0 # other aerodynmic coefficient multipliers
cxt_mult = val
cyt_mult = val
czt_mult = val
clt_mult = val
cmt_mult = val
cnt_mult = val
self.multipliers = (xcg_mult, cxt_mult, cyt_mult, czt_mult, clt_mult,
cmt_mult, cnt_mult)
self.ap = GcasAutopilot(self.xequil, self.uequil, self.flightLimits,
self.ctrlLimits)
self.pass_fail = FlightLimitsPFA(self.flightLimits)
self.sim_step = 0.01
# Q: Penalty on State Error in LQR controller
# These were chosen to try to achieve a natural frequency of 3 rad/sec and a damping ratio (zeta) of 0.707
# see the matlab code for more analysis of the resultant controllers
# state = [VT, alpha, beta, phi, theta, psi, P, Q, R, pn, pe, h, pow, Nz, Ps, Ny]
q_alpha = 1000
q_q = 0
q_Nz = 1500
q_beta = 0
q_p = 0
q_r = 0
q_ps_i = 1200
q_Ny_r_i = 3000
q_list = [
0, q_alpha, q_beta, 0, 0, 0, q_p, q_q, q_r, 0, 0, 0, 0, q_Nz,
q_ps_i, q_Ny_r_i
]
self.Q = np.diag(q_list)
# R: Penalty on Control Effort in LRQ controller
r_elevator = 1
r_aileron = 1
r_rudder = 1
r_list = [0, r_elevator, r_aileron, r_rudder, 0, 0, 0]
self.R = np.diag(r_list)
self.viewer = None
def step(self, ut):
u_ref = self.ap.get_u_ref(self.times[-1], self.states[-1])
u_deg = np.zeros((4, ))
u_deg[1:4] = ut
u_deg[0] = u_ref[3]
u_deg[0:4] += self.uequil
u_deg = np.clip(u_deg, self.u_low, self.u_high)
xd, u, Nz, ps, Ny_r = F16ManullyControl(self.times[-1], self.states[-1], self.f16_plant, self.ap, \
self.x_ctrl, u_deg, multipliers=self.multipliers)
self.pass_fail.advance(self.times[-1], self.states[-1], self.ap.state,
xd, u, Nz, ps, Ny_r)
self.Nz_list.append(Nz)
self.ps_list.append(ps)
self.u_list.append(u)
t = self.times[-1] + self.sim_step
state = self.states[-1] + self.sim_step * xd
self.times.append(t)
self.states.append(state)
updated = self.ap.advance_discrete_state(self.times[-1],
self.states[-1])
self.modes.append(self.ap.state)
if updated:
print("at time {}, state changes to {}".format(
self.times[-1], self.ap.state))
done = (not self.pass_fail.result()) and self.pass_fail.break_on_error
if not done:
reward = self.reward_func(self.states[-1], self.u_list[-1])
else:
reward = -1.0
return self.states[-1], reward, done, {}
def reward_func(self, state, action):
return -(np.sum(np.dot(self.Q, np.abs(state-self.stable_state)))\
+np.sum(np.dot(self.R, np.abs(action-self.stable_action))))*1e-7
@property
def x_ctrl(self):
# Calculate perturbation from trim state
x_delta = self.states[-1].copy()
x_delta[:len(self.xequil)] -= self.xequil
return np.array([x_delta[i] for i in [1, 7, 13, 2, 6, 8, 14, 15]],
dtype=float)
def reset(self, x0=None):
if x0 is None:
x0 = self.initial_space.sample()
assert type(x0) is np.ndarray
# run the numerical simulation
self.times = [0.0]
self.states = [x0]
self.modes = [self.ap.state]
self.Nz_list = []
self.ps_list = []
self.u_list = []
return x0