def run_task(*_):
initial_goal = np.array([0.6, -0.1, 0.80])
rospy.init_node('trpo_real_sawyer_pnp_exp', anonymous=True)
pnp_env = TheanoEnv(
PickAndPlaceEnv(initial_goal,
initial_joint_pos=INITIAL_ROBOT_JOINT_POS,
simulated=False))
rospy.on_shutdown(pnp_env.shutdown)
pnp_env.initialize()
env = pnp_env
policy = GaussianMLPPolicy(env_spec=env.spec, hidden_sizes=(32, 32))
baseline = LinearFeatureBaseline(env_spec=env.spec)
algo = TRPO(
env=env,
policy=policy,
baseline=baseline,
batch_size=4000,
max_path_length=100,
n_itr=100,
discount=0.99,
step_size=0.01,
plot=False,
force_batch_sampler=True,
)
algo.train()
logger.set_snapshot_gap(args.snapshot_gap)
logger.log_parameters_lite(params_log_file, args)
if trial > 0:
old_log_dir = args.log_dir + '/' + str(trial - 1)
logger.pop_prefix()
logger.remove_text_output(osp.join(old_log_dir, 'text.txt'))
logger.remove_tabular_output(osp.join(old_log_dir, 'process.csv'))
logger.add_text_output(text_log_file)
logger.add_tabular_output(tabular_log_file)
logger.push_prefix("[" + args.exp_name + '_trial ' + str(trial) + "]")
np.random.seed(trial)
params = policy.get_params()
sess.run(tf.variables_initializer(params))
baseline = LinearFeatureBaseline(env_spec=env.spec)
# optimizer = ConjugateGradientOptimizer(hvp_approach=FiniteDifferenceHvp(base_eps=1e-5))
top_paths = BPQ.BoundedPriorityQueue(top_k)
algo = TRPO(
env=env,
policy=policy,
baseline=baseline,
batch_size=args.batch_size,
step_size=args.step_size,
n_itr=args.n_itr,
store_paths=True,
# optimizer= optimizer,
max_path_length=max_path_length,
top_paths=top_paths,
plot=False,
