import common.wrapper as wrapper
import common.gym_interface as gym_interface
import common.callbacks as callbacks
from common.activation_fn import MyThreshold
if __name__ == "__main__":
args = common.args
print(args)
# SAC.learn need this. If use SubprocVecEnv instead of DummyVecEnv, you need to seed in each subprocess.
set_random_seed(common.seed)
saved_model_filename = common.build_model_filename(args)
hyperparams = common.load_hyperparameters(conf_name="PPO")
print(hyperparams)
# Make every env has the same obs space and action space
default_wrapper = []
# if padding zero:
# default_wrapper.append(wrapper.WalkerWrapper)
if args.realign_method != "":
default_wrapper.append(wrapper.ReAlignedWrapper)
assert len(args.train_bodies) > 0, "No body to train."
if args.with_bodyinfo:
default_wrapper.append(wrapper.BodyinfoWrapper)
print("Making train environments...")
from common.cnspns import CNSPNSPPO, CNSPNSPolicy
if __name__ == "__main__":
args = common.args
print(args)
args.test_bodies = args.train_bodies # omit test_bodies in command from now on.
args.initialize_weights_from = args.model_filename
# SAC.learn need this. If use SubprocVecEnv instead of DummyVecEnv, you need to seed in each subprocess.
set_random_seed(common.seed)
saved_model_filename = common.build_model_filename(args)
hyperparams = common.load_hyperparameters(conf_name=args.rl_hyperparameter)
print(hyperparams)
# Make every env has the same obs space and action space
default_wrapper = []
# if padding zero:
# default_wrapper.append(wrapper.WalkerWrapper)
if args.topology_wrapper == "same":
body_type = 0
for body in args.train_bodies + args.test_bodies:
if body_type == 0:
body_type = body//100
else:
assert body_type == body//100, "Training on different body types."
if args.realign_method != "":
import common.wrapper as wrapper
import common.gym_interface as gym_interface
import common.callbacks as callbacks
from common.activation_fn import MyThreshold
if __name__ == "__main__":
args = common.args
print(args)
# PPO.learn need this. If use SubprocVecEnv instead of DummyVecEnv, you need to seed in each subprocess.
set_random_seed(common.seed)
saved_model_filename = common.build_model_filename(args)
hyperparams = common.load_hyperparameters()
print(hyperparams)
# Make every env has the same obs space and action space
default_wrapper = [wrapper.WalkerWrapper]
assert len(args.train_bodies) > 0, "No body to train."
if args.with_bodyinfo:
default_wrapper += [wrapper.BodyinfoWrapper]
venv = DummyVecEnv([
gym_interface.make_env(
rank=i,
seed=common.seed,
wrappers=default_wrapper,
render=args.render,
robot_body=args.train_bodies[i % len(args.train_bodies)])
import common.wrapper as wrapper
import common.gym_interface as gym_interface
import common.callbacks as callbacks
from common.activation_fn import MyThreshold
from common.pns import PNSPPO, PNSMlpPolicy
if __name__ == "__main__":
args = common.args
print(args)
# args.vec_normalize = True # Robo need normalization.
saved_model_filename = common.build_model_filename(args)
hyperparams = common.load_hyperparameters(conf_name="Robo")
# Overwrite learning_rate using args:
# hyperparams["learning_rate"] = common.args.learning_rate
print(hyperparams)
# Make every env has the same obs space and action space
default_wrapper = []
# if padding zero:
# default_wrapper.append(wrapper.WalkerWrapper)
if args.topology_wrapper == "same":
body_type = 0
for body in args.train_bodies + args.test_bodies:
if body_type == 0:
body_type = body // 100