def test_embodied_teacher(self):
agents = ['youbot_noplugin', 'pioneer2dx_noplugin', 'pr2_noplugin']
with open(os.path.join(social_bot.get_model_dir(), "agent_cfg.json"),
'r') as cfg_file:
agent_cfgs = json.load(cfg_file)
for agent_type in agents:
for use_image_obs in [True, False]:
agent_cfg = agent_cfgs[agent_type]
test_tasks = [GoalTask, KickingBallTask]
if agent_cfg['camera_sensor'] == '' and use_image_obs:
continue
logging.info("Testing Case: Agent " + agent_type + ", Task " +
str(test_tasks) + ", UseImage: " +
str(use_image_obs))
env = EmbodiedTeacher(with_language=True,
use_image_observation=use_image_obs,
image_with_internal_states=True,
agent_type=agent_type,
tasks=test_tasks,
demo_by_human=False)
step_cnt = 0
last_done_time = time.time()
while step_cnt < 100 and (time.time() - last_done_time) < 5:
actions = env._control_space.sample()
actions = dict(control=actions, sentence="hello")
teacher_actions = env._teacher_control_space.sample()
actions = OrderedDict(learner=actions,
teacher=teacher_actions)
env.step(actions)
step_cnt += 1
step_per_sec = step_cnt / (time.time() - last_done_time)
logging.info("Test Passed, FPS: " + str(step_per_sec))
env.close()
gazebo.close()
def test_pr2(self):
for use_internal_states_only in [True, False]:
env = Pr2Gripper(world_config=PR2_WORLD_SETTING + [
"//sensor[@type='camera']<>visualize=true",
"//camera//format=R8G8B8",
],
max_steps=100,
use_internal_states_only=use_internal_states_only)
env.run(render=False)
env.close()
gazebo.close()
def test_play_ground(self):
with_language = True
with_agent_language = False
agents = [
'pioneer2dx_noplugin', 'pr2_noplugin', 'icub', 'icub_with_hands',
'youbot_noplugin'
]
with open(os.path.join(social_bot.get_model_dir(), "agent_cfg.json"),
'r') as cfg_file:
agent_cfgs = json.load(cfg_file)
tasks = [GoalTask, KickingBallTask, Reaching3D, PickAndPlace, Stack]
for task in tasks:
for agent_type in task.compatible_agents:
for with_language in [True, False]:
for use_image_obs in [True, False]:
for image_with_internal_states in [True, False]:
agent_cfg = agent_cfgs[agent_type]
test_tasks = [task]
if agent_type.find('icub') != -1:
test_tasks.append(ICubAuxiliaryTask)
if agent_cfg[
'camera_sensor'] == '' and use_image_obs:
continue
logging.info("Testing Case: Agent " + agent_type +
", Task " + str(test_tasks) +
", UseImage: " + str(use_image_obs))
env = PlayGround(
with_language=with_language,
with_agent_language=with_agent_language,
use_image_observation=use_image_obs,
image_with_internal_states=
image_with_internal_states,
agent_type=agent_type,
tasks=test_tasks)
step_cnt = 0
last_done_time = time.time()
while step_cnt < 100 and (time.time() -
last_done_time) < 5:
actions = env._control_space.sample()
if with_agent_language:
actions = dict(control=actions,
sentence="hello")
env.step(actions)
step_cnt += 1
step_per_sec = step_cnt / (time.time() -
last_done_time)
logging.info("Test Passed, FPS: " +
str(step_per_sec))
env.close()
gazebo.close()
def run(self):
port = os.environ.get('PYGAZEBO_PORT', 11345)
env = SimpleNavigationLanguage(port=port + self.env_id + 1)
env.reset()
for _ in range(500):
control = [
random.random() * 0.2 - 0.1,
random.random() * 0.2 - 0.1
]
obs, reward, done, info = env.step(
dict(control=control, sentence="hello"))
if done:
env.reset()
env.close()
gazebo.close()
self.ok.value = 1
def close(self):
super().close()
gazebo.close()