def test_network_distribution():
T1 = np.array([[0, 0.5, 0.5], [0, 1, 0], [0, 0, 1]])
Z2 = np.array([[1, 0], [1, 0], [0, 1]])
pomdp1 = POMDP([T1], [Z2],
input_names=['u1'],
state_name='x1',
output_name='z1')
T21 = np.array([[0, 1, 0], [0, 1, 0], [0, 0, 1]])
T22 = np.array([[0, 0, 1], [0, 1, 0], [0, 0, 1]])
pomdp2 = POMDP([T21, T22], [np.eye(3)],
input_names=['u2'],
state_name='x2',
output_name='z2')
network = POMDPNetwork([pomdp1, pomdp2])
network.add_connection(['z1'], 'u2', lambda z1: {z1})
# distribution over u1 x1 x2
D_ux = sparse.COO([[0], [0], [0]], [1], shape=(1, 3, 3))
D_xz = propagate_network_distribution(network, D_ux)
D_xz_r = sparse.COO([[1, 2], [1, 2], [0, 1], [1, 2]], [0.5, 0.5],
shape=(3, 3, 2, 3))
np.testing.assert_equal(D_xz.todense(), D_xz_r.todense())
def plan_mission(prob):
rob_abstr = Grid(prob['xmin'],
prob['xmax'],
prob['discretization'],
name_prefix='c')
env_list = [
environment_belief_model(info[1], name)
for (name, info) in prob['regs'].items()
]
# Construct rob-env network
rob_env_network = POMDPNetwork([rob_abstr.pomdp] + env_list)
for item in prob['regs'].items():
rob_env_network.add_connection(['c_x'], '{}_u'.format(item[0]),
get_rob_env_conn(item))
# solve rob LTL problem
predicates = get_predicates(prob['regs'])
rob_ltlpol = solve_ltl_cosafe(rob_env_network,
prob['formula'],
predicates,
horizon=prob['cas_T'],
delta=prob['step_margin'],
verbose=False)
return CassiePolicy(rob_ltlpol, rob_abstr)
def plan_exploration(prob, rob_policy):
rob_abstr = rob_policy.abstraction
rob_ltlpol = rob_policy.ltlpol
informed_samples = [r.chebXc for r, _, _ in prob['regs'].values()
] + [prob['uav_x0'], prob['uav_xT']]
uav_prm = PRM(prob['xmin'],
prob['xmax'],
num_nodes=40,
min_dist=2,
max_dist=5,
informed_samples=informed_samples,
name_prefix='u')
env_list = [
environment_belief_model(info[1], name)
for (name, info) in prob['regs'].items()
]
# Construct uav-env network
uav_env_network = POMDPNetwork([uav_prm.mdp] + env_list)
for item in prob['regs'].items():
uav_env_network.add_connection(['u_x'], '{}_u'.format(item[0]),
get_uav_env_conn(item))
# solve uav exploration problem
idx = np.logical_or(
rob_ltlpol.val[0][rob_abstr.x_to_s(prob['cas_x0']), ...,
rob_ltlpol.dfsa_init] > prob['accept_margin'],
rob_ltlpol.val[0][rob_abstr.x_to_s(prob['cas_x0']), ...,
rob_ltlpol.dfsa_init] < prob['reject_margin'])
target = np.zeros(uav_env_network.N)
target[uav_prm.x_to_s(prob['uav_xT'])][idx] = 1
costs = uav_prm.costs.reshape(uav_prm.costs.shape + (1, ) *
(1 + len(uav_env_network.N) - 2))
val_uav, pol_uav = solve_ssp(uav_env_network,
costs,
target,
M=500,
verbose=False)
return UAVPolicy(pol_uav, val_uav, uav_prm)
def test_evaluate_Q():
T1 = np.array([[0, 0.5, 0.5], [0, 1, 0], [0, 0, 1]])
Z2 = np.array([[1, 0], [1, 0], [0, 1]])
pomdp1 = POMDP([T1], [Z2],
input_names=['u1'],
state_name='x1',
output_name='z1')
T21 = np.array([[0, 1, 0], [0, 1, 0], [0, 0, 1]])
T22 = np.array([[0, 0, 1], [0, 1, 0], [0, 0, 1]])
pomdp2 = POMDP([T21, T22], [np.eye(3)],
input_names=['u2'],
state_name='x2',
output_name='z2')
network = POMDPNetwork([pomdp1, pomdp2])
network.add_connection(['z1'], 'u2', lambda z1: {z1})
V = np.array([[0, 0, 0], [0, 0, 0], [0, 0, 1]])
np.testing.assert_almost_equal(evaluate_Q(network, (0, ), (0, 0), V), 0.5)
def test_evolve():
'''test non-deterministic connection'''
T0 = np.array([[0, 1, 0], [0, 0, 1], [0, 0, 1]])
T1 = np.array([[1, 0, 0], [1, 0, 0], [0, 1, 0]])
mdp1 = POMDP([T0, T1], input_names=['u1'], state_name='x1')
mdp2 = POMDP([T0, T1], input_names=['u2'], state_name='x2')
network = POMDPNetwork()
network.add_pomdp(mdp1)
sp, _ = network.evolve([0], (0,))
np.testing.assert_equal(sp, [1])
network.add_pomdp(mdp2)
sp, _ = network.evolve([1,1], (0,1))
np.testing.assert_equal(sp, [2, 0])
network.add_connection(['x1'], 'u2', lambda x1: set([0, 1]))
n0 = 0
n2 = 0
for i in range(1000):
sp, _ = network.evolve([1,1], (0,))
np.testing.assert_equal(sp[0], 2)
if sp[1] == 0:
n0 += 1
if sp[1] == 2:
n2 += 1
np.testing.assert_equal(n0 + n2, 1000)
np.testing.assert_array_less(abs(n0 -n2), 100)