def setUp(self):
"""Create the UR5 environment.
This follows the pendulum.py example located in example_designs/.
"""
self.env = Pendulum(
use_contexts=True,
random_contexts=True,
context_range=[(np.deg2rad(-16), np.deg2rad(16)), (-0.6, 0.6)]
)
self.env.reset()
if evaluate else UR5(use_contexts=True,
random_contexts=True,
context_range=[(-np.pi, np.pi), (-np.pi / 4, 0),
(-np.pi / 4, np.pi / 4)],
show=render),
},
"Pendulum": {
"meta_ac_space":
lambda relative_goals, multiagent: Box(low=np.array([-np.pi, -15]),
high=np.array([np.pi, 15]),
dtype=np.float32),
"state_indices":
lambda multiagent: [0, 2],
"env":
lambda evaluate, render, n_levels, multiagent, shared, maddpg:
Pendulum(use_contexts=True, context_range=[0, 0], show=render)
if evaluate else Pendulum(use_contexts=True,
random_contexts=True,
context_range=[(np.deg2rad(-16),
np.deg2rad(16)),
(-0.6, 0.6)],
show=render),
},
# ======================================================================= #
# Mixed autonomy traffic flow environments. #
# ======================================================================= #
"ring-v0": {
"meta_ac_space":
lambda relative_goals, multiagent: Box(low=-5 if relative_goals else 0,
high=5
(-np.pi / 4, np.pi / 4)],
show=render)
if evaluate else UR5(use_contexts=True,
random_contexts=True,
context_range=[(-np.pi, np.pi), (-np.pi / 4, 0),
(-np.pi / 4, np.pi / 4)],
show=render),
},
"Pendulum": {
"meta_ac_space":
lambda relative_goals: Box(low=np.array([-np.pi, -15]),
high=np.array([np.pi, 15]),
dtype=np.float32),
"state_indices": [0, 2],
"env":
lambda evaluate, render, multiagent, shared, maddpg: Pendulum(
use_contexts=True, context_range=[0, 0], show=render) if evaluate
else Pendulum(use_contexts=True,
random_contexts=True,
context_range=[(np.deg2rad(-16), np.deg2rad(16)),
(-0.6, 0.6)],
show=render),
},
# ======================================================================= #
# Mixed autonomy traffic flow environments. #
# ======================================================================= #
"ring": {
"meta_ac_space":
lambda relative_goals: Box(low=-10 if relative_goals else 0,
high=10 if relative_goals else 30,
shape=(5, ),
def create_env(env, render=False, evaluate=False):
"""Return, and potentially create, the environment.
Parameters
----------
env : str or gym.Env
the environment, or the name of a registered environment.
render : bool
whether to render the environment
evaluate : bool
specifies whether this is a training or evaluation environment
Returns
-------
gym.Env or list of gym.Env
gym-compatible environment(s)
"""
if env == "AntGather":
env = AntGatherEnv()
elif env == "AntMaze":
if evaluate:
env = [
AntMaze(use_contexts=True, context_range=[16, 0]),
AntMaze(use_contexts=True, context_range=[16, 16]),
AntMaze(use_contexts=True, context_range=[0, 16])
]
else:
env = AntMaze(use_contexts=True,
random_contexts=True,
context_range=[(-4, 20), (-4, 20)])
elif env == "AntPush":
if evaluate:
env = AntPush(use_contexts=True, context_range=[0, 19])
else:
env = AntPush(use_contexts=True, context_range=[0, 19])
# env = AntPush(use_contexts=True,
# random_contexts=True,
# context_range=[(-16, 16), (-4, 20)])
elif env == "AntFall":
if evaluate:
env = AntFall(use_contexts=True, context_range=[0, 27, 4.5])
else:
env = AntFall(use_contexts=True, context_range=[0, 27, 4.5])
# env = AntFall(use_contexts=True,
# random_contexts=True,
# context_range=[(-4, 12), (-4, 28), (0, 5)])
elif env == "AntFourRooms":
if evaluate:
env = [
AntFourRooms(use_contexts=True, context_range=[30, 0]),
AntFourRooms(use_contexts=True, context_range=[0, 30]),
AntFourRooms(use_contexts=True, context_range=[30, 30])
]
else:
env = AntFourRooms(use_contexts=True,
random_contexts=False,
context_range=[[30, 0], [0, 30], [30, 30]])
elif env == "UR5":
if evaluate:
env = UR5(use_contexts=True,
random_contexts=True,
context_range=[(-np.pi, np.pi), (-np.pi / 4, 0),
(-np.pi / 4, np.pi / 4)],
show=render)
else:
env = UR5(use_contexts=True,
random_contexts=True,
context_range=[(-np.pi, np.pi), (-np.pi / 4, 0),
(-np.pi / 4, np.pi / 4)],
show=render)
elif env == "Pendulum":
if evaluate:
env = Pendulum(use_contexts=True,
context_range=[0, 0],
show=render)
else:
env = Pendulum(use_contexts=True,
random_contexts=True,
context_range=[(np.deg2rad(-16), np.deg2rad(16)),
(-0.6, 0.6)],
show=render)
elif env in [
"bottleneck0", "bottleneck1", "bottleneck2", "grid0", "grid1"
]:
# Import the benchmark and fetch its flow_params
benchmark = __import__("flow.benchmarks.{}".format(env),
fromlist=["flow_params"])
flow_params = benchmark.flow_params
# Get the env name and a creator for the environment.
create_env, _ = make_create_env(flow_params, version=0, render=render)
# Create the environment.
env = create_env()
elif env in ["ring0", "multi-ring0"]:
env = FlowEnv("ring", render=render) # FIXME
elif env in [
"merge0", "merge1", "merge2", "multi-merge0", "multi-merge1",
"multi-merge2"
]:
env_num = int(env[-1])
env = FlowEnv("merge",
env_params={
"exp_num": env_num,
"horizon": 6000,
"simulator": "traci",
"multiagent": env[:5] == "multi"
},
render=render)
elif env in [
"figureeight0", "figureeight1", "figureeight02",
"multi-figureeight0", "multi-figureeight1", "multi-figureeight02"
]:
env_num = int(env[-1])
env = FlowEnv("figure_eight",
env_params={
"num_automated": [1, 7, 14][env_num],
"horizon": 750,
"simulator": "traci",
"multiagent": env[:5] == "multi"
},
render=render)
elif env == "BipedalSoccer":
env = BipedalSoccer(render=render)
elif isinstance(env, str):
# This is assuming the environment is registered with OpenAI gym.
env = gym.make(env)
# Reset the environment.
if env is not None:
if isinstance(env, list):
for next_env in env:
next_env.reset()
else:
env.reset()
return env
class TestPendulum(unittest.TestCase):
"""Tests the Pendulum environment class."""
def setUp(self):
"""Create the UR5 environment.
This follows the pendulum.py example located in example_designs/.
"""
self.env = Pendulum(
use_contexts=True,
random_contexts=True,
context_range=[(np.deg2rad(-16), np.deg2rad(16)), (-0.6, 0.6)]
)
self.env.reset()
def tearDown(self):
del self.env
def test_init(self):
"""Ensure that all variables are being initialized properly."""
self.assertEqual(self.env.name, 'pendulum.xml')
self.assertEqual(self.env.observation_space.shape[0], 3)
self.assertEqual(self.env.action_space.shape[0], 1)
np.testing.assert_array_almost_equal(
(self.env.action_space.high - self.env.action_space.low) / 2, [2])
np.testing.assert_array_almost_equal(
(self.env.action_space.high + self.env.action_space.low) / 2, [0])
self.assertEqual(len(self.env.context_range), 2)
np.testing.assert_array_almost_equal(
self.env.end_goal_thresholds, [0.16580628, 0.6])
self.assertEqual(
self.env.context_range,
[(-0.2792526803190927, 0.2792526803190927), (-0.6, 0.6)])
self.assertEqual(self.env.max_actions, 1000)
self.assertEqual(self.env.visualize, False)
self.assertEqual(self.env.viewer, None)
self.assertEqual(self.env.num_frames_skip, 1)
np.testing.assert_array_almost_equal(
self.env.context_space.low, [-0.279253, -0.6])
np.testing.assert_array_almost_equal(
self.env.context_space.high, [0.279253, 0.6])
def test_step(self):
"""Ensure the step method is functioning properly.
This does the following tasks:
* checks that the simulation control is set to the action.
* Checks that a sufficient number of steps have passed.
"""
action = [1]
steps_before = self.env.num_steps
self.env.step(action)
steps_after = self.env.num_steps
# check the number of steps that have passed
self.assertEqual(steps_after - steps_before, self.env.num_frames_skip)
# check the control method
np.testing.assert_array_almost_equal(action, self.env.sim.data.ctrl[:])
def test_reset(self):
"""Ensure the state initialization is within the expected range."""
state = self.env.reset()
num_obj = len(state) // 3
for i in range(num_obj):
self.assertTrue(np.arccos(state[i])
>= self.env.initial_state_space[i][0])
self.assertTrue(np.arccos(state[i])
<= self.env.initial_state_space[i][1])
self.assertTrue(state[i + 2 * num_obj]
>= self.env.initial_state_space[i + num_obj][0])
self.assertTrue(state[i + 2 * num_obj]
<= self.env.initial_state_space[i + num_obj][1])
def test_display_end_goal(self):
pass
def test_get_next_goal(self):
pass
def test_display_subgoal(self):
pass