def __init__(self, environment, horizon, dt, path,
init_iterations=500,
step_iterations=1,
fcost=None,
constrained=True,
fastForward=False,
tolGrad = 1e-4,
tolFun = 1e-7,
save_interval=10,
printing=True,
saving=True,
init_optim = True,
finite_diff=True,
noise_gain=0.05,
add_noise=False,
ou_theta=0.15,
ou_sigma=0.2):
super(MPCAgent, self).__init__(environment.uDim)
self.traj_optimizer = iLQR(environment, horizon, dt,
path=path,
fcost=fcost,
constrained=constrained,
fastForward=fastForward,
save_interval=save_interval,
tolGrad = tolGrad,
tolFun = tolFun,
printing=printing,
finite_diff=finite_diff,
file_prefix='ilqr_',
init=False,
saving=saving,
reset_mu=True)
self.uMax = self.traj_optimizer.environment.uMax
self.init_iterations = init_iterations
self.step_iterations = step_iterations
self.x0 = environment.x
if init_optim==True:
self.init_optim(environment.x)
self.noise_gain = noise_gain
self.add_noise = add_noise
self.action_noise = OUnoise(self.uDim, dt, theta=ou_theta, sigma=ou_sigma)
def c_N(x):
x1, x2, x3, x4, x5, x6 = x
c = 100 * x1**2 + 100 * x2**2 + 100 * x3**2 + 10 * x4**2 + 10 * x5**2 + 10 * x6**2
return c
# initial state value
x0 = [0, np.pi, np.pi, 0, 0, 0]
t = 3. # simulation time
dt = 0.005 # time step-size
env = CartPoleDoubleSerial(c_k, x0, dt)
path = '../../../results/ilqr/cart_pole_double_serial2/'
algorithm = iLQR(env,
t,
dt,
constrained=True,
fcost=c_N,
path=path,
maxIters=1000) # instance of the iLQR algorithm
#algorithm.run_disk(x0)
algorithm.run_optim() # run trajectory optimization
# plot trajectories
algorithm.plot()
plt.show()
# create an animation
algorithm.animation()
c = 0.5 * (2 * x1**2 + 10. * x2**2 + 10. * x3**2 + 0.5 * x4**2 +
.1 * u1**2)
return c
# define the final cost at step N
def c_N(x):
x1, x2, x3, x4, x5, x6 = x
c = 0.5 * (100. * x1**2 + 100. * x2**2 + 100. * x3**2)
return c
# initial state value
x0 = [0, np.pi, np.pi, 0, 0, 0]
t = 8 # simulation time
dt = 0.01 # time step-size
env = CartPoleDoubleParallel(c_k, x0, dt)
path = '../../../results/ilqr/cart_pole_double_parallel/' # path, where results are saved
algorithm = iLQR(env, t, dt, constrained=True, fcost=c_N, path=path)
algorithm.run_optim() # run trajectory optimization
# plot trajectories
algorithm.plot()
plt.show()
# create an animation
algorithm.animation()
x4 - mod.pi)**2 + 10 * x5**2 + 10 * x6**2 + 10 * x7**2 + 10 * x8**2
return c
# initial state value
x0 = [0, np.pi, np.pi, np.pi, 0, 0, 0, 0]
t = 2.0 # simulation time
dt = 0.002 # time step-size
env = CartPoleTriple(c_k, x0, dt)
env.uMax = 20
path = '~/tripilqr/' # path, where results are saved
algorithm = iLQR(env,
t,
dt,
constrained=True,
fcost=c_N,
path=path,
maxIters=10,
fastForward=False) # instance of the iLQR algorithm#
#algorithm.run_disk(x0)
algorithm.run_optim() # run trajectory optimization
# plot trajectories
algorithm.plot()
# create an animation
#system
algorithm.animation()
# define the final cost at step N
def c_N(x):
x1, x2, x3, x4 = x
c = 100 * x1**2 + 100 * x2**2 + 10 * x3**2 + 10 * x4**2
return c
# initial state value
x0 = [0, np.pi, 0, 0]
t = 5 # simulation time
dt = 0.02 # time step-size
env = CartPole(c_k, x0, dt)
env.uMax = 4 * env.uMax
path = './results/ilqr/cart_pole5/' # path, where results are saved
algorithm = iLQR(env,
t,
dt,
path=path,
fcost=None,
constrained=True,
maxIters=500) # instance of the iLQR algorithm
#algorithm.run_disk(x0)
algorithm.run_optim() # run trajectory optimization
# plot trajectories
algorithm.plot()
plt.show()
# create an animation
algorithm.animation()
x1, x2 = x
c = 100 * (1 - np.cos(x1)) + 0.1 * x2**2
return c
# initial state value
x0 = [np.pi, 0]
t = 5 # simulation time
dt = 0.05 # time step-size
env = Pendulum(c_k, x0, dt)
path = '../../../results/pendulum/ilqr/' # path, where results are saved
algorithm = iLQR(env,
t,
dt,
fcost=c_N,
constrained=True,
path=path,
finite_diff=True) # instance of the iLQR algorithm
algorithm.run_optim() # run trajectory optimization
# plot trajectories
algorithm.plot()
plt.show()
# create an animation
algorithm.animation()