def test_observe_first_state_returns_valid_state(self):
sim = SimStub.make_valid_state_stub(self.task)
first_state = self.task.observe_first_state(sim)
self.assertEqual(len(first_state), len(self.task.state_variables))
self.assertIsInstance(first_state, tuple)
def test_observe_first_state_creates_target_heading_in_expected_range(
self):
sim = SimStub.make_valid_state_stub(self.task)
_ = self.task.observe_first_state(sim)
desired_heading = sim[HeadingControlTask.target_track_deg]
self.assertGreaterEqual(desired_heading, 0)
self.assertLessEqual(desired_heading, 360)
def setUp(self):
self.task = self.make_task()
sim = SimStub.make_valid_state_stub(self.task)
_ = self.task.observe_first_state(
sim) # causes task to init new-episode attributes
self.dummy_action = np.asarray(
[0 for _ in range(len(self.task.action_variables))])
def test_observe_first_state_changes_target_heading(self):
sim = SimStub.make_valid_state_stub(self.task)
_ = self.task.observe_first_state(sim)
desired_heading = sim[HeadingControlTask.target_track_deg]
_ = self.task.observe_first_state(sim)
new_desired_heading = sim[HeadingControlTask.target_track_deg]
self.assertNotEqual(desired_heading, new_desired_heading)
def test_task_first_observation_inputs_controls(self):
dummy_sim = SimStub.make_valid_state_stub(self.task)
_ = self.task.observe_first_state(dummy_sim)
# check engine as expected
self.assertAlmostEqual(self.task.THROTTLE_CMD,
dummy_sim[prp.throttle_cmd])
self.assertAlmostEqual(self.task.MIXTURE_CMD,
dummy_sim[prp.mixture_cmd])
self.assertAlmostEqual(1.0, dummy_sim[prp.engine_running])
def test_task_step_returns_terminal_time_equals_max(self):
sim = SimStub.make_valid_state_stub(self.task)
_ = self.task.observe_first_state(sim)
terminal_time = self.default_episode_time_s
sim[prp.sim_time_s] = terminal_time
steps = 1
_, _, is_terminal, _ = self.task.task_step(sim, self.dummy_action,
steps)
self.assertTrue(is_terminal)
def test_task_step_returns_reward_in_info(self):
sim = SimStub.make_valid_state_stub(self.task)
steps = 1
_ = self.task.observe_first_state(sim)
_, reward_scalar, _, info = self.task.task_step(
sim, self.dummy_action, steps)
reward_object = info['reward']
self.assertIsInstance(reward_object, rewards.Reward)
self.assertAlmostEqual(reward_object.agent_reward(), reward_scalar)
def test_observe_first_state_correct_track_error(self):
self.setUp()
sim = SimStub.make_valid_state_stub(self.task)
_ = self.task.observe_first_state(sim)
track_deg = prp.Vector2.from_sim(sim).heading_deg()
target_track_deg = sim[self.task.target_track_deg]
error_deg = track_deg - target_track_deg
expected_acute_error_deg = utils.reduce_reflex_angle_deg(error_deg)
actual_error_deg = sim[self.task.track_error_deg]
self.assertAlmostEqual(expected_acute_error_deg, actual_error_deg)
def get_initial_state_sim(self, task=None) -> SimStub:
if task is None:
task = self.task
sim = SimStub.make_valid_state_stub(task)
# set properties to reasonable initial episode values
sim[prp.sim_time_s] = 0.0
sim[prp.dist_travel_m] = 0.0
sim[prp.heading_deg] = task.INITIAL_HEADING_DEG
sim[prp.altitude_sl_ft] = task.get_initial_conditions()[
prp.initial_altitude_ft]
sim[prp.roll_rad] = 0
return sim
def test_task_step_correct_return_types(self):
sim = SimStub.make_valid_state_stub(self.task)
steps = 1
_ = self.task.observe_first_state(sim)
state, reward, is_terminal, info = self.task.task_step(
sim, self.dummy_action, steps)
self.assertIsInstance(state, tuple)
self.assertEqual(len(state), len(self.task.state_variables))
self.assertIsInstance(reward, float)
self.assertIsInstance(is_terminal, bool)
self.assertIsInstance(info, dict)
def get_perfect_state_sim(self, task=None, time_terminal=True) -> SimStub:
if task is None:
task = self.task
sim = SimStub.make_valid_state_stub(task)
task.observe_first_state(sim)
perfect_track = sim[task.target_track_deg]
perfect_altitude = task._get_target_altitude()
perfect_roll = 0.0
perfect_sideslip = 0.0
self.modify_sim_to_state_(sim,
task,
steps_terminal=time_terminal,
track_deg=perfect_track,
altitude_ft=perfect_altitude,
roll_rad=perfect_roll,
sideslip_deg=perfect_sideslip)
return sim
def get_initial_sim_with_state(self,
task: HeadingControlTask = None,
time_terminal=False,
track_deg=None,
altitude_ft=None,
roll_rad=0.0) -> SimStub:
if task is None:
task = self.task
sim = SimStub.make_valid_state_stub(task)
task.observe_first_state(sim)
if track_deg is None:
# get the target value and use that (i.e. perfect state)
track_deg = sim[task.target_track_deg]
perfect_initial_sim = self.modify_sim_to_state_(
sim, task, time_terminal, track_deg, altitude_ft, roll_rad)
# the task set its last_state attr when it observed the first state - reset this
# based on the perfect sim values we just set. Ugh.
perfect_state = task.State(*(sim[prop]
for prop in task.state_variables))
task.last_state = perfect_state
return perfect_initial_sim