def examplePendulumLQR():
Q = np.diag([1, 0, 0])
def random_action():
return 4 * rd.random() - 2
def convert_obs(obs):
a = np.matrix(obs).T
a.put(0, a.item(0) - 1)
return a
def convert_control(u):
a = u.item(0)
a = max(-2, min(2, a))
return a
env = gym.make('Pendulum-v0')
agent = linear.LinearAgent(
n=3,
Q=Q,
epsilon=0.05,
N=20,
convert_control=convert_control,
convert_obs=convert_obs,
random_action=random_action,
)
ctl = ctrl.Controller(env, agent)
results = ctl.run_episodes(1000)
env.close()
return results
def exampleCartpoleLQR():
c_x = 1 # 2.4
c_theta = 0.2 # 0.2
Q = np.diag([1. / (c_x**2), 0, 1. / (c_theta**2), 0])
def random_action():
return rd.choice([-1, 1])
def convert_obs(obs):
a = np.matrix(obs).T
a.put(0, a.item(0) - 2)
return a
def convert_action(a):
return (a + 1) // 2
def convert_control(u):
if u.item(0) >= 0:
return 1
return -1
env = gym.make('CartPole-v1')
agent = linear.LinearAgent(
n=4,
Q=Q,
epsilon=0.05,
N=20,
convert_action=convert_action,
convert_control=convert_control,
convert_obs=convert_obs,
random_action=random_action,
)
ctl = ctrl.Controller(env, agent)
results = ctl.run_episodes(1000)
env.close()
return results