def test_reward(self):
"""Check the reward function for different values.
The reward function should be a linear combination of the average speed
of all vehicles and a penalty on the requested accelerations by the
AVs.
"""
# create the environment
env = WaveAttenuationPOEnv(sim_params=self.sim_params,
scenario=self.scenario,
env_params=self.env_params)
env.reset()
# check the reward for no acceleration
env.k.vehicle.test_set_speed('human_0', 0)
env.k.vehicle.test_set_speed('rl_0', 0)
self.assertAlmostEqual(env.compute_reward(rl_actions=[0], fail=False),
0)
env.k.vehicle.test_set_speed('human_0', 0)
env.k.vehicle.test_set_speed('rl_0', 1)
self.assertAlmostEqual(env.compute_reward(rl_actions=[0], fail=False),
0.1)
env.k.vehicle.test_set_speed('human_0', 1)
env.k.vehicle.test_set_speed('rl_0', 1)
self.assertAlmostEqual(env.compute_reward(rl_actions=[0], fail=False),
0.2)
# check the fail option
env.k.vehicle.test_set_speed('human_0', 1)
env.k.vehicle.test_set_speed('rl_0', 1)
self.assertAlmostEqual(env.compute_reward(rl_actions=[0], fail=True),
0)
# check the effect of RL actions
env.k.vehicle.test_set_speed('human_0', 1)
env.k.vehicle.test_set_speed('rl_0', 1)
self.assertAlmostEqual(env.compute_reward(rl_actions=None, fail=False),
0)
env.k.vehicle.test_set_speed('human_0', 1)
env.k.vehicle.test_set_speed('rl_0', 1)
self.assertAlmostEqual(env.compute_reward(rl_actions=[1], fail=False),
-3.8)
def test_observation_action_space(self):
"""Tests the observation and action spaces upon initialization."""
# create the environment
env = WaveAttenuationPOEnv(
sumo_params=self.sumo_params,
scenario=self.scenario,
env_params=self.env_params
)
# check the observation space
self.assertTrue(test_space(
env.observation_space,
expected_size=3, expected_min=0, expected_max=1))
# check the action space
self.assertTrue(test_space(
env.action_space,
expected_size=1, expected_min=-1, expected_max=1))
env.terminate()
def test_observation_action_space(self):
"""Tests the observation and action spaces upon initialization."""
# create the environment
env = WaveAttenuationPOEnv(sim_params=self.sim_params,
network=self.network,
env_params=self.env_params)
# check the observation space
self.assertTrue(
test_space(env.observation_space,
expected_size=3,
expected_min=-float('inf'),
expected_max=float('inf')))
# check the action space
self.assertTrue(
test_space(env.action_space,
expected_size=1,
expected_min=-1,
expected_max=1))
env.terminate()
