def td3_experiment_online_vae_exploring(variant):
import railrl.samplers.rollout_functions as rf
import railrl.torch.pytorch_util as ptu
from railrl.data_management.online_vae_replay_buffer import \
OnlineVaeRelabelingBuffer
from railrl.exploration_strategies.base import (
PolicyWrappedWithExplorationStrategy)
from railrl.torch.her.online_vae_joint_algo import OnlineVaeHerJointAlgo
from railrl.torch.networks import FlattenMlp, TanhMlpPolicy
from railrl.torch.td3.td3 import TD3
from railrl.torch.vae.vae_trainer import ConvVAETrainer
preprocess_rl_variant(variant)
env = get_envs(variant)
es = get_exploration_strategy(variant, env)
observation_key = variant.get('observation_key', 'latent_observation')
desired_goal_key = variant.get('desired_goal_key', 'latent_desired_goal')
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
obs_dim = (env.observation_space.spaces[observation_key].low.size +
env.observation_space.spaces[desired_goal_key].low.size)
action_dim = env.action_space.low.size
qf1 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'],
)
qf2 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'],
)
policy = TanhMlpPolicy(
input_size=obs_dim,
output_size=action_dim,
**variant['policy_kwargs'],
)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
exploring_qf1 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'],
)
exploring_qf2 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'],
)
exploring_policy = TanhMlpPolicy(
input_size=obs_dim,
output_size=action_dim,
**variant['policy_kwargs'],
)
exploring_exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=exploring_policy,
)
vae = env.vae
replay_buffer = OnlineVaeRelabelingBuffer(
vae=vae,
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
variant["algo_kwargs"]["replay_buffer"] = replay_buffer
if variant.get('use_replay_buffer_goals', False):
env.replay_buffer = replay_buffer
env.use_replay_buffer_goals = True
vae_trainer_kwargs = variant.get('vae_trainer_kwargs')
t = ConvVAETrainer(variant['vae_train_data'],
variant['vae_test_data'],
vae,
beta=variant['online_vae_beta'],
**vae_trainer_kwargs)
control_algorithm = TD3(env=env,
training_env=env,
qf1=qf1,
qf2=qf2,
policy=policy,
exploration_policy=exploration_policy,
**variant['algo_kwargs'])
exploring_algorithm = TD3(env=env,
training_env=env,
qf1=exploring_qf1,
qf2=exploring_qf2,
policy=exploring_policy,
exploration_policy=exploring_exploration_policy,
**variant['algo_kwargs'])
assert 'vae_training_schedule' not in variant,\
"Just put it in joint_algo_kwargs"
algorithm = OnlineVaeHerJointAlgo(vae=vae,
vae_trainer=t,
env=env,
training_env=env,
policy=policy,
exploration_policy=exploration_policy,
replay_buffer=replay_buffer,
algo1=control_algorithm,
algo2=exploring_algorithm,
algo1_prefix="Control_",
algo2_prefix="VAE_Exploration_",
observation_key=observation_key,
desired_goal_key=desired_goal_key,
**variant['joint_algo_kwargs'])
algorithm.to(ptu.device)
vae.to(ptu.device)
if variant.get("save_video", True):
policy.train(False)
rollout_function = rf.create_rollout_function(
rf.multitask_rollout,
max_path_length=algorithm.max_path_length,
observation_key=algorithm.observation_key,
desired_goal_key=algorithm.desired_goal_key,
)
video_func = get_video_save_func(
rollout_function,
env,
algorithm.eval_policy,
variant,
)
algorithm.post_train_funcs.append(video_func)
algorithm.train()
def td3_experiment(variant):
import railrl.samplers.rollout_functions as rf
import railrl.torch.pytorch_util as ptu
from railrl.data_management.obs_dict_replay_buffer import \
ObsDictRelabelingBuffer
from railrl.exploration_strategies.base import (
PolicyWrappedWithExplorationStrategy)
from railrl.torch.td3.td3 import TD3 as TD3Trainer
from railrl.torch.torch_rl_algorithm import TorchBatchRLAlgorithm
from railrl.torch.networks import FlattenMlp, TanhMlpPolicy
# preprocess_rl_variant(variant)
env = get_envs(variant)
expl_env = env
eval_env = env
es = get_exploration_strategy(variant, env)
if variant.get("use_masks", False):
mask_wrapper_kwargs = variant.get("mask_wrapper_kwargs", dict())
expl_mask_distribution_kwargs = variant[
"expl_mask_distribution_kwargs"]
expl_mask_distribution = DiscreteDistribution(
**expl_mask_distribution_kwargs)
expl_env = RewardMaskWrapper(env, expl_mask_distribution,
**mask_wrapper_kwargs)
eval_mask_distribution_kwargs = variant[
"eval_mask_distribution_kwargs"]
eval_mask_distribution = DiscreteDistribution(
**eval_mask_distribution_kwargs)
eval_env = RewardMaskWrapper(env, eval_mask_distribution,
**mask_wrapper_kwargs)
env = eval_env
max_path_length = variant['max_path_length']
observation_key = variant.get('observation_key', 'latent_observation')
desired_goal_key = variant.get('desired_goal_key', 'latent_desired_goal')
achieved_goal_key = variant.get('achieved_goal_key',
'latent_achieved_goal')
# achieved_goal_key = desired_goal_key.replace("desired", "achieved")
obs_dim = (env.observation_space.spaces[observation_key].low.size +
env.observation_space.spaces[desired_goal_key].low.size)
action_dim = env.action_space.low.size
qf1 = FlattenMlp(input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
qf2 = FlattenMlp(input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
policy = TanhMlpPolicy(input_size=obs_dim,
output_size=action_dim,
**variant['policy_kwargs'])
target_qf1 = FlattenMlp(input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
target_qf2 = FlattenMlp(input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
target_policy = TanhMlpPolicy(input_size=obs_dim,
output_size=action_dim,
**variant['policy_kwargs'])
if variant.get("use_subgoal_policy", False):
from railrl.policies.timed_policy import SubgoalPolicyWrapper
subgoal_policy_kwargs = variant.get('subgoal_policy_kwargs', {})
policy = SubgoalPolicyWrapper(wrapped_policy=policy,
env=env,
episode_length=max_path_length,
**subgoal_policy_kwargs)
target_policy = SubgoalPolicyWrapper(wrapped_policy=target_policy,
env=env,
episode_length=max_path_length,
**subgoal_policy_kwargs)
expl_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
replay_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
# use_masks=variant.get("use_masks", False),
**variant['replay_buffer_kwargs'])
trainer = TD3Trainer(policy=policy,
qf1=qf1,
qf2=qf2,
target_qf1=target_qf1,
target_qf2=target_qf2,
target_policy=target_policy,
**variant['td3_trainer_kwargs'])
# if variant.get("use_masks", False):
# from railrl.torch.her.her import MaskedHERTrainer
# trainer = MaskedHERTrainer(trainer)
# else:
trainer = HERTrainer(trainer)
if variant.get("do_state_exp", False):
eval_path_collector = GoalConditionedPathCollector(
eval_env,
policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
# use_masks=variant.get("use_masks", False),
# full_mask=True,
)
expl_path_collector = GoalConditionedPathCollector(
expl_env,
expl_policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
# use_masks=variant.get("use_masks", False),
)
else:
eval_path_collector = VAEWrappedEnvPathCollector(
env,
policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
goal_sampling_mode=['evaluation_goal_sampling_mode'],
)
expl_path_collector = VAEWrappedEnvPathCollector(
env,
expl_policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
goal_sampling_mode=['exploration_goal_sampling_mode'],
)
algorithm = TorchBatchRLAlgorithm(
trainer=trainer,
exploration_env=env,
evaluation_env=env,
exploration_data_collector=expl_path_collector,
evaluation_data_collector=eval_path_collector,
replay_buffer=replay_buffer,
max_path_length=max_path_length,
**variant['algo_kwargs'])
vis_variant = variant.get('vis_kwargs', {})
vis_list = vis_variant.get('vis_list', [])
if variant.get("save_video", True):
if variant.get("do_state_exp", False):
rollout_function = rf.create_rollout_function(
rf.multitask_rollout,
max_path_length=max_path_length,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
# use_masks=variant.get("use_masks", False),
# full_mask=True,
# vis_list=vis_list,
)
video_func = get_video_save_func(
rollout_function,
env,
policy,
variant,
)
else:
video_func = VideoSaveFunction(
env,
variant,
)
algorithm.post_train_funcs.append(video_func)
algorithm.to(ptu.device)
if not variant.get("do_state_exp", False):
env.vae.to(ptu.device)
algorithm.train()
def experiment(variant):
feat_points = 16
history = 1
latent_obs_dim = feat_points * 2 * history
imsize = 64
downsampled_size = 32
env = SawyerXYZEnv()
extra_fc_size = env.obs_dim
env = ImageMujocoWithObsEnv(env,
imsize=imsize,
normalize=True,
grayscale=True,
keep_prev=history - 1,
init_camera=camera.sawyer_init_camera)
"""env = ImageMujocoEnv(env,
imsize=imsize,
keep_prev=history-1,
init_camera=camera.sawyer_init_camera)"""
es = GaussianStrategy(action_space=env.action_space, )
obs_dim = env.observation_space.low.size
action_dim = env.action_space.low.size
ae = FeatPointMlp(input_size=imsize,
downsample_size=downsampled_size,
input_channels=1,
num_feat_points=feat_points)
replay_buffer = AEEnvReplayBuffer(int(1e4),
env,
imsize=imsize,
history_length=history,
downsampled_size=downsampled_size)
qf = FlattenMlp(input_size=latent_obs_dim + extra_fc_size + action_dim,
output_size=1,
hidden_sizes=[400, 300])
policy = AETanhPolicy(
input_size=latent_obs_dim + extra_fc_size,
ae=ae,
env=env,
history_length=history,
output_size=action_dim,
hidden_sizes=[400, 300],
)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
algorithm = FeatPointDDPG(ae,
history,
env=env,
qf=qf,
policy=policy,
exploration_policy=exploration_policy,
replay_buffer=replay_buffer,
extra_fc_size=extra_fc_size,
imsize=imsize,
downsampled_size=downsampled_size,
**variant['algo_params'])
algorithm.to(ptu.device)
algorithm.train()
'e': np.array([-1, -1, 0, 0]),
'z': np.array([1, 1, 0, 0]),
'c': np.array([-1, 1, 0, 0]),
'x': 'toggle',
'r': 'reset',
}
# np.random.seed(100)
env = SawyerDoorPullOpenActionLimitedEnv(fix_goal=True, min_y_pos=.3)
policy = ZeroPolicy(env.action_space.low.size)
es = OUStrategy(env.action_space, theta=1)
es = EpsilonGreedy(
action_space=env.action_space,
prob_random_action=0.1,
)
policy = exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
env.reset()
ACTION_FROM = 'hardcoded'
# ACTION_FROM = 'pd'
# ACTION_FROM = 'random'
H = 10000
# H = 300
# H = 50
goal = .25
while True:
lock_action = False
obs = env.reset()
last_reward_t = 0
def experiment(variant):
expl_env = variant['env_class'](**variant['env_kwargs'])
eval_env = variant['env_class'](**variant['env_kwargs'])
observation_key = 'state_observation'
desired_goal_key = 'state_desired_goal'
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
es = GaussianAndEpislonStrategy(
action_space=expl_env.action_space,
max_sigma=.2,
min_sigma=.2, # constant sigma
epsilon=.3,
)
obs_dim = expl_env.observation_space.spaces['observation'].low.size
goal_dim = expl_env.observation_space.spaces['desired_goal'].low.size
action_dim = expl_env.action_space.low.size
qf1 = FlattenMlp(
input_size=obs_dim + goal_dim + action_dim,
output_size=1,
**variant['qf_kwargs']
)
qf2 = FlattenMlp(
input_size=obs_dim + goal_dim + action_dim,
output_size=1,
**variant['qf_kwargs']
)
target_qf1 = FlattenMlp(
input_size=obs_dim + goal_dim + action_dim,
output_size=1,
**variant['qf_kwargs']
)
target_qf2 = FlattenMlp(
input_size=obs_dim + goal_dim + action_dim,
output_size=1,
**variant['qf_kwargs']
)
policy = TanhMlpPolicy(
input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs']
)
target_policy = TanhMlpPolicy(
input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs']
)
expl_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
replay_buffer = ObsDictRelabelingBuffer(
env=eval_env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs']
)
trainer = TD3(
policy=policy,
qf1=qf1,
qf2=qf2,
target_qf1=target_qf1,
target_qf2=target_qf2,
target_policy=target_policy,
**variant['trainer_kwargs']
)
trainer = HERTrainer(trainer)
eval_path_collector = GoalConditionedPathCollector(
eval_env,
policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
)
expl_path_collector = GoalConditionedPathCollector(
expl_env,
expl_policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
)
algorithm = TorchBatchRLAlgorithm(
trainer=trainer,
exploration_env=expl_env,
evaluation_env=eval_env,
exploration_data_collector=expl_path_collector,
evaluation_data_collector=eval_path_collector,
replay_buffer=replay_buffer,
**variant['algo_kwargs']
)
algorithm.to(ptu.device)
algorithm.train()
def experiment(variant):
rdim = variant["rdim"]
vae_paths = {
2:
"/home/ashvin/data/s3doodad/ashvin/vae/pusher2d-conv-sweep2/run1/id0/params.pkl",
4:
"/home/ashvin/data/s3doodad/ashvin/vae/pusher2d-conv-sweep2/run1/id1/params.pkl",
8:
"/home/ashvin/data/s3doodad/ashvin/vae/pusher2d-conv-sweep2/run1/id2/params.pkl",
16:
"/home/ashvin/data/s3doodad/ashvin/vae/pusher2d-conv-sweep2/run1/id3/params.pkl"
}
vae_path = vae_paths[rdim]
vae = torch.load(vae_path)
print("loaded", vae_path)
if variant['multitask']:
env = CylinderXYPusher2DEnv(**variant["env_kwargs"])
env = ImageMujocoEnv(env, 84, camera_name="topview", transpose=True)
env = VAEWrappedEnv(env,
vae,
use_vae_obs=True,
use_vae_reward=True,
use_vae_goals=True)
env = MultitaskToFlatEnv(env)
# else:
# env = Pusher2DEnv(**variant['env_kwargs'])
if variant['normalize']:
env = NormalizedBoxEnv(env)
exploration_type = variant['exploration_type']
if exploration_type == 'ou':
es = OUStrategy(action_space=env.action_space)
elif exploration_type == 'gaussian':
es = GaussianStrategy(
action_space=env.action_space,
max_sigma=0.1,
min_sigma=0.1, # Constant sigma
)
elif exploration_type == 'epsilon':
es = EpsilonGreedy(
action_space=env.action_space,
prob_random_action=0.1,
)
else:
raise Exception("Invalid type: " + exploration_type)
obs_dim = env.observation_space.low.size
action_dim = env.action_space.low.size
qf1 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=[400, 300],
)
qf2 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=[400, 300],
)
policy = TanhMlpPolicy(
input_size=obs_dim,
output_size=action_dim,
hidden_sizes=[400, 300],
)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
algorithm = TD3(env,
training_env=env,
qf1=qf1,
qf2=qf2,
policy=policy,
exploration_policy=exploration_policy,
**variant['algo_kwargs'])
print("use_gpu", variant["use_gpu"], bool(variant["use_gpu"]))
if variant["use_gpu"]:
gpu_id = variant["gpu_id"]
ptu.set_gpu_mode(True)
ptu.set_device(gpu_id)
algorithm.to(ptu.device)
env._wrapped_env.vae.to(ptu.device)
algorithm.train()
def her_td3_experiment(variant):
import gym
import railrl.torch.pytorch_util as ptu
from railrl.data_management.obs_dict_replay_buffer import ObsDictRelabelingBuffer
from railrl.exploration_strategies.base import \
PolicyWrappedWithExplorationStrategy
from railrl.exploration_strategies.gaussian_and_epislon import \
GaussianAndEpislonStrategy
from railrl.launchers.launcher_util import setup_logger
from railrl.samplers.data_collector import GoalConditionedPathCollector
from railrl.torch.her.her import HERTrainer
from railrl.torch.networks import FlattenMlp, TanhMlpPolicy
from railrl.torch.td3.td3 import TD3
from railrl.torch.torch_rl_algorithm import TorchBatchRLAlgorithm
import railrl.samplers.rollout_functions as rf
from railrl.torch.grill.launcher import get_state_experiment_video_save_function
if 'env_id' in variant:
eval_env = gym.make(variant['env_id'])
expl_env = gym.make(variant['env_id'])
else:
eval_env_kwargs = variant.get('eval_env_kwargs', variant['env_kwargs'])
eval_env = variant['env_class'](**eval_env_kwargs)
expl_env = variant['env_class'](**variant['env_kwargs'])
observation_key = 'state_observation'
desired_goal_key = 'state_desired_goal'
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
es = GaussianAndEpislonStrategy(
action_space=expl_env.action_space,
max_sigma=.2,
min_sigma=.2, # constant sigma
epsilon=.3,
)
obs_dim = expl_env.observation_space.spaces['observation'].low.size
goal_dim = expl_env.observation_space.spaces['desired_goal'].low.size
action_dim = expl_env.action_space.low.size
qf1 = FlattenMlp(input_size=obs_dim + goal_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
qf2 = FlattenMlp(input_size=obs_dim + goal_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
target_qf1 = FlattenMlp(input_size=obs_dim + goal_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
target_qf2 = FlattenMlp(input_size=obs_dim + goal_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
policy = TanhMlpPolicy(input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs'])
target_policy = TanhMlpPolicy(input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs'])
expl_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
replay_buffer = ObsDictRelabelingBuffer(
env=eval_env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
trainer = TD3(policy=policy,
qf1=qf1,
qf2=qf2,
target_qf1=target_qf1,
target_qf2=target_qf2,
target_policy=target_policy,
**variant['trainer_kwargs'])
trainer = HERTrainer(trainer)
eval_path_collector = GoalConditionedPathCollector(
eval_env,
policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
)
expl_path_collector = GoalConditionedPathCollector(
expl_env,
expl_policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
)
algorithm = TorchBatchRLAlgorithm(
trainer=trainer,
exploration_env=expl_env,
evaluation_env=eval_env,
exploration_data_collector=expl_path_collector,
evaluation_data_collector=eval_path_collector,
replay_buffer=replay_buffer,
**variant['algo_kwargs'])
if variant.get("save_video", False):
rollout_function = rf.create_rollout_function(
rf.multitask_rollout,
max_path_length=algorithm.max_path_length,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
)
video_func = get_state_experiment_video_save_function(
rollout_function,
eval_env,
policy,
variant,
)
algorithm.post_epoch_funcs.append(video_func)
algorithm.to(ptu.device)
algorithm.train()
def experiment(variant):
if variant.get("pretrained_algorithm_path", False):
resume(variant)
return
if 'env' in variant:
env_params = ENV_PARAMS[variant['env']]
variant.update(env_params)
if 'env_id' in env_params:
if env_params['env_id'] in [
'pen-v0', 'pen-sparse-v0', 'door-v0', 'relocate-v0',
'hammer-v0', 'pen-sparse-v0', 'door-sparse-v0',
'relocate-sparse-v0', 'hammer-sparse-v0'
]:
import mj_envs
expl_env = gym.make(env_params['env_id'])
eval_env = gym.make(env_params['env_id'])
else:
expl_env = NormalizedBoxEnv(variant['env_class']())
eval_env = NormalizedBoxEnv(variant['env_class']())
if variant.get('sparse_reward', False):
expl_env = RewardWrapperEnv(expl_env, compute_hand_sparse_reward)
eval_env = RewardWrapperEnv(eval_env, compute_hand_sparse_reward)
if variant.get('add_env_demos', False):
variant["path_loader_kwargs"]["demo_paths"].append(
variant["env_demo_path"])
if variant.get('add_env_offpolicy_data', False):
variant["path_loader_kwargs"]["demo_paths"].append(
variant["env_offpolicy_data_path"])
else:
expl_env = encoder_wrapped_env(variant)
eval_env = encoder_wrapped_env(variant)
path_loader_kwargs = variant.get("path_loader_kwargs", {})
stack_obs = path_loader_kwargs.get("stack_obs", 1)
if stack_obs > 1:
expl_env = StackObservationEnv(expl_env, stack_obs=stack_obs)
eval_env = StackObservationEnv(eval_env, stack_obs=stack_obs)
obs_dim = expl_env.observation_space.low.size
action_dim = eval_env.action_space.low.size
if hasattr(expl_env, 'info_sizes'):
env_info_sizes = expl_env.info_sizes
else:
env_info_sizes = dict()
M = variant['layer_size']
qf1 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=[M, M],
)
qf2 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=[M, M],
)
target_qf1 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=[M, M],
)
target_qf2 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=[M, M],
)
policy_class = variant.get("policy_class", TanhGaussianPolicy)
policy = policy_class(
obs_dim=obs_dim,
action_dim=action_dim,
**variant['policy_kwargs'],
)
buffer_policy = policy_class(
obs_dim=obs_dim,
action_dim=action_dim,
**variant['policy_kwargs'],
)
eval_policy = MakeDeterministic(policy)
eval_path_collector = MdpPathCollector(
eval_env,
eval_policy,
)
expl_policy = policy
exploration_kwargs = variant.get('exploration_kwargs', {})
if exploration_kwargs:
if exploration_kwargs.get("deterministic_exploration", False):
expl_policy = MakeDeterministic(policy)
exploration_strategy = exploration_kwargs.get("strategy", None)
if exploration_strategy is None:
pass
elif exploration_strategy == 'ou':
es = OUStrategy(
action_space=expl_env.action_space,
max_sigma=exploration_kwargs['noise'],
min_sigma=exploration_kwargs['noise'],
)
expl_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=expl_policy,
)
elif exploration_strategy == 'gauss_eps':
es = GaussianAndEpislonStrategy(
action_space=expl_env.action_space,
max_sigma=exploration_kwargs['noise'],
min_sigma=exploration_kwargs['noise'], # constant sigma
epsilon=0,
)
expl_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=expl_policy,
)
else:
error
if variant.get('replay_buffer_class',
EnvReplayBuffer) == AWREnvReplayBuffer:
main_replay_buffer_kwargs = variant['replay_buffer_kwargs']
main_replay_buffer_kwargs['env'] = expl_env
main_replay_buffer_kwargs['qf1'] = qf1
main_replay_buffer_kwargs['qf2'] = qf2
main_replay_buffer_kwargs['policy'] = policy
else:
main_replay_buffer_kwargs = dict(
max_replay_buffer_size=variant['replay_buffer_size'],
env=expl_env,
)
replay_buffer_kwargs = dict(
max_replay_buffer_size=variant['replay_buffer_size'],
env=expl_env,
)
replay_buffer = variant.get('replay_buffer_class',
EnvReplayBuffer)(**main_replay_buffer_kwargs, )
trainer = AWRSACTrainer(env=eval_env,
policy=policy,
qf1=qf1,
qf2=qf2,
target_qf1=target_qf1,
target_qf2=target_qf2,
buffer_policy=buffer_policy,
**variant['trainer_kwargs'])
if variant['collection_mode'] == 'online':
expl_path_collector = MdpStepCollector(
expl_env,
policy,
)
algorithm = TorchOnlineRLAlgorithm(
trainer=trainer,
exploration_env=expl_env,
evaluation_env=eval_env,
exploration_data_collector=expl_path_collector,
evaluation_data_collector=eval_path_collector,
replay_buffer=replay_buffer,
max_path_length=variant['max_path_length'],
batch_size=variant['batch_size'],
num_epochs=variant['num_epochs'],
num_eval_steps_per_epoch=variant['num_eval_steps_per_epoch'],
num_expl_steps_per_train_loop=variant[
'num_expl_steps_per_train_loop'],
num_trains_per_train_loop=variant['num_trains_per_train_loop'],
min_num_steps_before_training=variant[
'min_num_steps_before_training'],
)
else:
expl_path_collector = MdpPathCollector(
expl_env,
expl_policy,
)
algorithm = TorchBatchRLAlgorithm(
trainer=trainer,
exploration_env=expl_env,
evaluation_env=eval_env,
exploration_data_collector=expl_path_collector,
evaluation_data_collector=eval_path_collector,
replay_buffer=replay_buffer,
max_path_length=variant['max_path_length'],
batch_size=variant['batch_size'],
num_epochs=variant['num_epochs'],
num_eval_steps_per_epoch=variant['num_eval_steps_per_epoch'],
num_expl_steps_per_train_loop=variant[
'num_expl_steps_per_train_loop'],
num_trains_per_train_loop=variant['num_trains_per_train_loop'],
min_num_steps_before_training=variant[
'min_num_steps_before_training'],
)
algorithm.to(ptu.device)
demo_train_buffer = EnvReplayBuffer(**replay_buffer_kwargs, )
demo_test_buffer = EnvReplayBuffer(**replay_buffer_kwargs, )
if variant.get('save_paths', False):
algorithm.post_train_funcs.append(save_paths)
if variant.get('load_demos', False):
path_loader_class = variant.get('path_loader_class', MDPPathLoader)
path_loader = path_loader_class(trainer,
replay_buffer=replay_buffer,
demo_train_buffer=demo_train_buffer,
demo_test_buffer=demo_test_buffer,
**path_loader_kwargs)
path_loader.load_demos()
if variant.get('pretrain_policy', False):
trainer.pretrain_policy_with_bc()
if variant.get('pretrain_rl', False):
trainer.pretrain_q_with_bc_data()
if variant.get('save_pretrained_algorithm', False):
p_path = osp.join(logger.get_snapshot_dir(), 'pretrain_algorithm.p')
pt_path = osp.join(logger.get_snapshot_dir(), 'pretrain_algorithm.pt')
data = algorithm._get_snapshot()
data['algorithm'] = algorithm
torch.save(data, open(pt_path, "wb"))
torch.save(data, open(p_path, "wb"))
if variant.get('train_rl', True):
algorithm.train()
def her_td3_experiment(variant):
env = variant['env_class'](**variant['env_kwargs'])
her_kwargs = variant['algo_kwargs']['her_kwargs']
observation_key = her_kwargs['observation_key']
desired_goal_key = her_kwargs['desired_goal_key']
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
replay_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs']
)
obs_dim = env.observation_space.spaces['observation'].low.size
action_dim = env.action_space.low.size
goal_dim = env.observation_space.spaces['desired_goal'].low.size
exploration_type = variant['exploration_type']
if exploration_type == 'ou':
es = OUStrategy(
action_space=env.action_space,
**variant['es_kwargs']
)
elif exploration_type == 'gaussian':
es = GaussianStrategy(
action_space=env.action_space,
**variant['es_kwargs'],
)
elif exploration_type == 'epsilon':
es = EpsilonGreedy(
action_space=env.action_space,
**variant['es_kwargs'],
)
else:
raise Exception("Invalid type: " + exploration_type)
qf1 = FlattenMlp(
input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs']
)
qf2 = FlattenMlp(
input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs']
)
num_ensemble_qs = variant.get("num_ensemble_qs", 0)
ensemble_qs = [FlattenMlp(
input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs']
) for _ in range(num_ensemble_qs)]
policy = TanhMlpPolicy(
input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs']
)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
render = variant.get("render", False)
algorithm = HerExplorationTd3(
env,
qf1=qf1,
qf2=qf2,
policy=policy,
exploration_policy=exploration_policy,
replay_buffer=replay_buffer,
render=render,
render_during_eval=render,
ensemble_qs=ensemble_qs,
**variant['algo_kwargs']
)
algorithm.to(ptu.device)
algorithm.train()
def encoder_wrapped_td3bc_experiment(variant):
representation_size = 128
output_classes = 20
model_class = variant.get('model_class', TimestepPredictionModel)
model = model_class(
representation_size,
# decoder_output_activation=decoder_activation,
output_classes=output_classes,
**variant['model_kwargs'],
)
# model = torch.nn.DataParallel(model)
model_path = variant.get("model_path")
# model = load_local_or_remote_file(model_path)
state_dict = torch.load(model_path)
model.load_state_dict(state_dict)
model.to(ptu.device)
model.eval()
traj = np.load(variant.get("desired_trajectory"), allow_pickle=True)[0]
goal_image = traj["observations"][-1]["image_observation"]
goal_image = goal_image.reshape(1, 3, 500, 300).transpose([0, 1, 3, 2
]) / 255.0
# goal_image = goal_image.reshape(1, 300, 500, 3).transpose([0, 3, 1, 2]) / 255.0 # BECAUSE RLBENCH DEMOS ARENT IMAGE_ENV WRAPPED
# goal_image = goal_image[:, :, :240, 60:500]
goal_image = goal_image[:, :, 60:, 60:500]
goal_image_pt = ptu.from_numpy(goal_image)
save_image(goal_image_pt.data.cpu(), 'demos/goal.png', nrow=1)
goal_latent = model.encode(goal_image_pt).detach().cpu().numpy().flatten()
initial_image = traj["observations"][0]["image_observation"]
initial_image = initial_image.reshape(1, 3, 500, 300).transpose(
[0, 1, 3, 2]) / 255.0
# initial_image = initial_image.reshape(1, 300, 500, 3).transpose([0, 3, 1, 2]) / 255.0
# initial_image = initial_image[:, :, :240, 60:500]
initial_image = initial_image[:, :, 60:, 60:500]
initial_image_pt = ptu.from_numpy(initial_image)
save_image(initial_image_pt.data.cpu(), 'demos/initial.png', nrow=1)
initial_latent = model.encode(
initial_image_pt).detach().cpu().numpy().flatten()
# Move these to td3_bc and bc_v3 (or at least type for reward_params)
reward_params = dict(
goal_latent=goal_latent,
initial_latent=initial_latent,
type=variant["reward_params_type"],
)
config_params = variant.get("config_params")
env = variant['env_class'](**variant['env_kwargs'])
env = ImageEnv(
env,
recompute_reward=False,
transpose=True,
image_length=450000,
reward_type="image_distance",
# init_camera=sawyer_pusher_camera_upright_v2,
)
env = EncoderWrappedEnv(
env, model, reward_params, config_params,
**variant.get("encoder_wrapped_env_kwargs", dict()))
expl_env = env # variant['env_class'](**variant['env_kwargs'])
eval_env = env # variant['env_class'](**variant['env_kwargs'])
observation_key = variant.get("observation_key", 'state_observation')
# one of 'state_observation', 'latent_observation', 'concat_observation'
desired_goal_key = 'latent_desired_goal'
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
es = GaussianAndEpislonStrategy(
action_space=expl_env.action_space,
**variant["exploration_kwargs"],
)
obs_dim = expl_env.observation_space.spaces[observation_key].low.size
goal_dim = expl_env.observation_space.spaces[desired_goal_key].low.size
action_dim = expl_env.action_space.low.size
qf1 = FlattenMlp(
input_size=obs_dim + goal_dim + action_dim,
output_size=1,
# output_activation=TorchMaxClamp(0.0),
**variant['qf_kwargs'])
qf2 = FlattenMlp(
input_size=obs_dim + goal_dim + action_dim,
output_size=1,
# output_activation=TorchMaxClamp(0.0),
**variant['qf_kwargs'])
target_qf1 = FlattenMlp(
input_size=obs_dim + goal_dim + action_dim,
output_size=1,
# output_activation=TorchMaxClamp(0.0),
**variant['qf_kwargs'])
target_qf2 = FlattenMlp(
input_size=obs_dim + goal_dim + action_dim,
output_size=1,
# output_activation=TorchMaxClamp(0.0),
**variant['qf_kwargs'])
# Support for CNNPolicy based policy/target policy
# Defaults to TanhMlpPolicy unless cnn_params is supplied in variant
if 'cnn_params' in variant.keys():
imsize = 48
policy = CNNPolicy(
input_width=imsize,
input_height=imsize,
output_size=action_dim,
input_channels=3,
**variant['cnn_params'],
output_activation=torch.tanh,
)
target_policy = CNNPolicy(
input_width=imsize,
input_height=imsize,
output_size=action_dim,
input_channels=3,
**variant['cnn_params'],
output_activation=torch.tanh,
)
else:
policy = TanhMlpPolicy(input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs'])
target_policy = TanhMlpPolicy(input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs'])
expl_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
replay_buffer = ObsDictRelabelingBuffer(
env=eval_env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
demo_train_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
demo_test_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
td3bc_trainer = TD3BCTrainer(env=env,
policy=policy,
qf1=qf1,
qf2=qf2,
replay_buffer=replay_buffer,
demo_train_buffer=demo_train_buffer,
demo_test_buffer=demo_test_buffer,
target_qf1=target_qf1,
target_qf2=target_qf2,
target_policy=target_policy,
**variant['trainer_kwargs'])
trainer = HERTrainer(td3bc_trainer)
eval_path_collector = GoalConditionedPathCollector(
eval_env,
policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
)
expl_path_collector = GoalConditionedPathCollector(
expl_env,
expl_policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
)
algorithm = TorchBatchRLAlgorithm(
trainer=trainer,
exploration_env=expl_env,
evaluation_env=eval_env,
exploration_data_collector=expl_path_collector,
evaluation_data_collector=eval_path_collector,
replay_buffer=replay_buffer,
**variant['algo_kwargs'])
if variant.get("save_video", True):
video_func = VideoSaveFunction(
env,
variant,
)
algorithm.post_train_funcs.append(video_func)
algorithm.to(ptu.device)
td3bc_trainer.load_demos()
td3bc_trainer.pretrain_policy_with_bc()
td3bc_trainer.pretrain_q_with_bc_data()
algorithm.train()
def experiment(variant):
import railrl.samplers.rollout_functions as rf
import railrl.torch.pytorch_util as ptu
from railrl.data_management.obs_dict_replay_buffer import \
ObsDictRelabelingBuffer
from railrl.exploration_strategies.base import (
PolicyWrappedWithExplorationStrategy)
from railrl.torch.her.her import HERTrainer
from railrl.torch.td3.td3 import TD3 as TD3Trainer
from railrl.torch.networks import FlattenMlp, TanhMlpPolicy
from railrl.torch.torch_rl_algorithm import TorchBatchRLAlgorithm
from railrl.samplers.data_collector import GoalConditionedPathCollector
from railrl.torch.grill.launcher import (
grill_preprocess_variant,
get_envs,
get_exploration_strategy,
full_experiment_variant_preprocess,
train_vae_and_update_variant,
get_video_save_func,
)
full_experiment_variant_preprocess(variant)
if not variant['grill_variant'].get('do_state_exp', False):
train_vae_and_update_variant(variant)
variant = variant['grill_variant']
grill_preprocess_variant(variant)
eval_env = get_envs(variant)
expl_env = get_envs(variant)
es = get_exploration_strategy(variant, expl_env)
observation_key = variant.get('observation_key', 'latent_observation')
desired_goal_key = variant.get('desired_goal_key', 'latent_desired_goal')
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
obs_dim = (expl_env.observation_space.spaces[observation_key].low.size +
expl_env.observation_space.spaces[desired_goal_key].low.size)
action_dim = expl_env.action_space.low.size
qf1 = FlattenMlp(input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
qf2 = FlattenMlp(input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
policy = TanhMlpPolicy(input_size=obs_dim,
output_size=action_dim,
**variant['policy_kwargs'])
target_qf1 = FlattenMlp(input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
target_qf2 = FlattenMlp(input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
target_policy = TanhMlpPolicy(input_size=obs_dim,
output_size=action_dim,
**variant['policy_kwargs'])
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
replay_buffer = ObsDictRelabelingBuffer(
env=eval_env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
trainer = TD3Trainer(policy=policy,
qf1=qf1,
qf2=qf2,
target_qf1=target_qf1,
target_qf2=target_qf2,
target_policy=target_policy,
**variant['td3_kwargs'])
trainer = HERTrainer(trainer)
eval_path_collector = GoalConditionedPathCollector(
eval_env,
policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
)
expl_path_collector = GoalConditionedPathCollector(
expl_env,
exploration_policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
)
algorithm = TorchBatchRLAlgorithm(
trainer=trainer,
exploration_env=expl_env,
evaluation_env=eval_env,
exploration_data_collector=expl_path_collector,
evaluation_data_collector=eval_path_collector,
replay_buffer=replay_buffer,
**variant['algo_kwargs'])
if variant.get("save_video", True): # Does not work
rollout_function = rf.create_rollout_function(
rf.multitask_rollout,
max_path_length=algorithm.max_path_length,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
)
video_func = get_video_save_func(
rollout_function,
eval_env,
policy,
variant,
)
algorithm.post_epoch_funcs.append(video_func)
algorithm.to(ptu.device)
if not variant.get("do_state_exp", False):
eval_env.vae.to(ptu.device)
expl_env.vae.to(ptu.device)
algorithm.train()
def HER_baseline_td3_experiment(variant):
import railrl.torch.pytorch_util as ptu
from railrl.data_management.obs_dict_replay_buffer import \
ObsDictRelabelingBuffer
from railrl.exploration_strategies.base import (
PolicyWrappedWithExplorationStrategy)
from railrl.torch.her.her_td3 import HerTd3
from railrl.torch.networks import MergedCNN, CNNPolicy
import torch
from multiworld.core.image_env import ImageEnv
from railrl.misc.asset_loader import load_local_or_remote_file
init_camera = variant.get("init_camera", None)
presample_goals = variant.get('presample_goals', False)
presampled_goals_path = get_presampled_goals_path(
variant.get('presampled_goals_path', None))
if 'env_id' in variant:
import gym
import multiworld
multiworld.register_all_envs()
env = gym.make(variant['env_id'])
else:
env = variant["env_class"](**variant['env_kwargs'])
image_env = ImageEnv(
env,
variant.get('imsize'),
reward_type='image_sparse',
init_camera=init_camera,
transpose=True,
normalize=True,
)
if presample_goals:
if presampled_goals_path is None:
image_env.non_presampled_goal_img_is_garbage = True
presampled_goals = variant['generate_goal_dataset_fctn'](
env=image_env, **variant['goal_generation_kwargs'])
else:
presampled_goals = load_local_or_remote_file(
presampled_goals_path).item()
del image_env
env = ImageEnv(
env,
variant.get('imsize'),
reward_type='image_distance',
init_camera=init_camera,
transpose=True,
normalize=True,
presampled_goals=presampled_goals,
)
else:
env = image_env
es = get_exploration_strategy(variant, env)
observation_key = variant.get('observation_key', 'image_observation')
desired_goal_key = variant.get('desired_goal_key', 'image_desired_goal')
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
imsize = variant['imsize']
action_dim = env.action_space.low.size
qf1 = MergedCNN(input_width=imsize,
input_height=imsize,
output_size=1,
input_channels=3 * 2,
added_fc_input_size=action_dim,
**variant['cnn_params'])
qf2 = MergedCNN(input_width=imsize,
input_height=imsize,
output_size=1,
input_channels=3 * 2,
added_fc_input_size=action_dim,
**variant['cnn_params'])
policy = CNNPolicy(
input_width=imsize,
input_height=imsize,
added_fc_input_size=0,
output_size=action_dim,
input_channels=3 * 2,
output_activation=torch.tanh,
**variant['cnn_params'],
)
target_qf1 = MergedCNN(input_width=imsize,
input_height=imsize,
output_size=1,
input_channels=3 * 2,
added_fc_input_size=action_dim,
**variant['cnn_params'])
target_qf2 = MergedCNN(input_width=imsize,
input_height=imsize,
output_size=1,
input_channels=3 * 2,
added_fc_input_size=action_dim,
**variant['cnn_params'])
target_policy = CNNPolicy(
input_width=imsize,
input_height=imsize,
added_fc_input_size=0,
output_size=action_dim,
input_channels=3 * 2,
output_activation=torch.tanh,
**variant['cnn_params'],
)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
replay_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
algo_kwargs = variant['algo_kwargs']
algo_kwargs['replay_buffer'] = replay_buffer
base_kwargs = algo_kwargs['base_kwargs']
base_kwargs['training_env'] = env
base_kwargs['render'] = variant["render"]
base_kwargs['render_during_eval'] = variant["render"]
her_kwargs = algo_kwargs['her_kwargs']
her_kwargs['observation_key'] = observation_key
her_kwargs['desired_goal_key'] = desired_goal_key
algorithm = HerTd3(env,
qf1=qf1,
qf2=qf2,
policy=policy,
target_qf1=target_qf1,
target_qf2=target_qf2,
target_policy=target_policy,
exploration_policy=exploration_policy,
**variant['algo_kwargs'])
algorithm.to(ptu.device)
algorithm.train()
def tdm_td3_experiment_online_vae(variant):
import railrl.samplers.rollout_functions as rf
import railrl.torch.pytorch_util as ptu
from railrl.data_management.online_vae_replay_buffer import \
OnlineVaeRelabelingBuffer
from railrl.exploration_strategies.base import (
PolicyWrappedWithExplorationStrategy)
from railrl.state_distance.tdm_networks import TdmQf, TdmPolicy
from railrl.torch.vae.vae_trainer import ConvVAETrainer
from railrl.torch.online_vae.online_vae_tdm_td3 import OnlineVaeTdmTd3
preprocess_rl_variant(variant)
env = get_envs(variant)
es = get_exploration_strategy(variant, env)
vae_trainer_kwargs = variant.get('vae_trainer_kwargs')
observation_key = variant.get('observation_key', 'latent_observation')
desired_goal_key = variant.get('desired_goal_key', 'latent_desired_goal')
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
obs_dim = (env.observation_space.spaces[observation_key].low.size)
goal_dim = (env.observation_space.spaces[desired_goal_key].low.size)
action_dim = env.action_space.low.size
vectorized = 'vectorized' in env.reward_type
variant['algo_kwargs']['tdm_td3_kwargs']['tdm_kwargs'][
'vectorized'] = vectorized
norm_order = env.norm_order
# variant['algo_kwargs']['tdm_td3_kwargs']['tdm_kwargs'][
# 'norm_order'] = norm_order
qf1 = TdmQf(env=env,
vectorized=vectorized,
norm_order=norm_order,
observation_dim=obs_dim,
goal_dim=goal_dim,
action_dim=action_dim,
**variant['qf_kwargs'])
qf2 = TdmQf(env=env,
vectorized=vectorized,
norm_order=norm_order,
observation_dim=obs_dim,
goal_dim=goal_dim,
action_dim=action_dim,
**variant['qf_kwargs'])
policy = TdmPolicy(env=env,
observation_dim=obs_dim,
goal_dim=goal_dim,
action_dim=action_dim,
**variant['policy_kwargs'])
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
vae = env.vae
replay_buffer = OnlineVaeRelabelingBuffer(
vae=vae,
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
algo_kwargs = variant['algo_kwargs']['tdm_td3_kwargs']
td3_kwargs = algo_kwargs['td3_kwargs']
td3_kwargs['training_env'] = env
tdm_kwargs = algo_kwargs['tdm_kwargs']
tdm_kwargs['observation_key'] = observation_key
tdm_kwargs['desired_goal_key'] = desired_goal_key
algo_kwargs["replay_buffer"] = replay_buffer
t = ConvVAETrainer(variant['vae_train_data'],
variant['vae_test_data'],
vae,
beta=variant['online_vae_beta'],
**vae_trainer_kwargs)
render = variant["render"]
assert 'vae_training_schedule' not in variant, "Just put it in algo_kwargs"
algorithm = OnlineVaeTdmTd3(
online_vae_kwargs=dict(vae=vae,
vae_trainer=t,
**variant['algo_kwargs']['online_vae_kwargs']),
tdm_td3_kwargs=dict(env=env,
qf1=qf1,
qf2=qf2,
policy=policy,
exploration_policy=exploration_policy,
**variant['algo_kwargs']['tdm_td3_kwargs']),
)
algorithm.to(ptu.device)
vae.to(ptu.device)
if variant.get("save_video", True):
policy.train(False)
rollout_function = rf.create_rollout_function(
rf.tdm_rollout,
init_tau=algorithm._sample_max_tau_for_rollout(),
decrement_tau=algorithm.cycle_taus_for_rollout,
cycle_tau=algorithm.cycle_taus_for_rollout,
max_path_length=algorithm.max_path_length,
observation_key=algorithm.observation_key,
desired_goal_key=algorithm.desired_goal_key,
)
video_func = get_video_save_func(
rollout_function,
env,
algorithm.eval_policy,
variant,
)
algorithm.post_train_funcs.append(video_func)
algorithm.to(ptu.device)
if not variant.get("do_state_exp", False):
env.vae.to(ptu.device)
algorithm.train()
def tdm_td3_experiment(variant):
import railrl.samplers.rollout_functions as rf
import railrl.torch.pytorch_util as ptu
from railrl.core import logger
from railrl.data_management.obs_dict_replay_buffer import \
ObsDictRelabelingBuffer
from railrl.exploration_strategies.base import (
PolicyWrappedWithExplorationStrategy)
from railrl.state_distance.tdm_networks import TdmQf, TdmPolicy
from railrl.state_distance.tdm_td3 import TdmTd3
preprocess_rl_variant(variant)
env = get_envs(variant)
es = get_exploration_strategy(variant, env)
observation_key = variant.get('observation_key', 'latent_observation')
desired_goal_key = variant.get('desired_goal_key', 'latent_desired_goal')
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
obs_dim = (env.observation_space.spaces[observation_key].low.size)
goal_dim = (env.observation_space.spaces[desired_goal_key].low.size)
action_dim = env.action_space.low.size
vectorized = 'vectorized' in env.reward_type
norm_order = env.norm_order
variant['algo_kwargs']['tdm_kwargs']['vectorized'] = vectorized
variant['qf_kwargs']['vectorized'] = vectorized
variant['qf_kwargs']['norm_order'] = norm_order
qf1 = TdmQf(env=env,
observation_dim=obs_dim,
goal_dim=goal_dim,
action_dim=action_dim,
**variant['qf_kwargs'])
qf2 = TdmQf(env=env,
observation_dim=obs_dim,
goal_dim=goal_dim,
action_dim=action_dim,
**variant['qf_kwargs'])
policy = TdmPolicy(env=env,
observation_dim=obs_dim,
goal_dim=goal_dim,
action_dim=action_dim,
**variant['policy_kwargs'])
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
variant['replay_buffer_kwargs']['vectorized'] = vectorized
replay_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
algo_kwargs = variant['algo_kwargs']
algo_kwargs['replay_buffer'] = replay_buffer
base_kwargs = algo_kwargs['base_kwargs']
base_kwargs['training_env'] = env
base_kwargs['render'] = variant["render"]
base_kwargs['render_during_eval'] = variant["render"]
tdm_kwargs = algo_kwargs['tdm_kwargs']
tdm_kwargs['observation_key'] = observation_key
tdm_kwargs['desired_goal_key'] = desired_goal_key
algorithm = TdmTd3(env,
qf1=qf1,
qf2=qf2,
policy=policy,
exploration_policy=exploration_policy,
**variant['algo_kwargs'])
algorithm.to(ptu.device)
if not variant.get("do_state_exp", False):
env.vae.to(ptu.device)
if variant.get("save_video", True):
logdir = logger.get_snapshot_dir()
policy.train(False)
rollout_function = rf.create_rollout_function(
rf.tdm_rollout,
init_tau=algorithm.max_tau,
max_path_length=algorithm.max_path_length,
observation_key=algorithm.observation_key,
desired_goal_key=algorithm.desired_goal_key,
)
video_func = get_video_save_func(
rollout_function,
env,
policy,
variant,
)
algorithm.post_train_funcs.append(video_func)
algorithm.train()
def experiment(variant):
vectorized = variant['td3_tdm_kwargs']['tdm_kwargs']['vectorized']
env = NormalizedBoxEnv(variant['env_class'](**variant['env_kwargs']))
max_tau = variant['td3_tdm_kwargs']['tdm_kwargs']['max_tau']
qf1 = TdmQf(env, vectorized=vectorized, **variant['qf_kwargs'])
qf2 = TdmQf(env, vectorized=vectorized, **variant['qf_kwargs'])
tdm_normalizer = TdmNormalizer(env,
vectorized,
max_tau=max_tau,
**variant['tdm_normalizer_kwargs'])
implicit_model = TdmToImplicitModel(
env,
qf1,
tau=0,
)
vf = TdmVf(env=env,
vectorized=vectorized,
tdm_normalizer=tdm_normalizer,
**variant['vf_kwargs'])
policy = TdmPolicy(env=env,
tdm_normalizer=tdm_normalizer,
**variant['policy_kwargs'])
replay_buffer = HerReplayBuffer(env=env,
**variant['her_replay_buffer_kwargs'])
goal_slice = env.ob_to_goal_slice
lbfgs_mpc_controller = TdmLBfgsBCMC(implicit_model,
env,
goal_slice=goal_slice,
multitask_goal_slice=goal_slice,
tdm_policy=policy,
**variant['mpc_controller_kwargs'])
state_only_mpc_controller = TdmLBfgsBStateOnlyCMC(
vf,
policy,
env,
goal_slice=goal_slice,
multitask_goal_slice=goal_slice,
**variant['state_only_mpc_controller_kwargs'])
es = GaussianStrategy(action_space=env.action_space,
**variant['es_kwargs'])
if variant['explore_with'] == 'TdmLBfgsBCMC':
raw_exploration_policy = lbfgs_mpc_controller
elif variant['explore_with'] == 'TdmLBfgsBStateOnlyCMC':
raw_exploration_policy = state_only_mpc_controller
else:
raw_exploration_policy = policy
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=raw_exploration_policy,
)
if variant['eval_with'] == 'TdmLBfgsBCMC':
eval_policy = lbfgs_mpc_controller
elif variant['eval_with'] == 'TdmLBfgsBStateOnlyCMC':
eval_policy = state_only_mpc_controller
else:
eval_policy = policy
# variant['td3_tdm_kwargs']['base_kwargs']['eval_policy'] = eval_policy
algorithm = TdmTd3(env=env,
policy=policy,
qf1=qf1,
qf2=qf2,
vf=vf,
exploration_policy=exploration_policy,
eval_policy=eval_policy,
replay_buffer=replay_buffer,
**variant['td3_tdm_kwargs'])
algorithm.to(ptu.device)
algorithm.train()
def her_td3_experiment(variant):
env = variant['env_class'](**variant['env_kwargs'])
observation_key = variant['observation_key']
desired_goal_key = variant['desired_goal_key']
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
variant['algo_kwargs']['her_kwargs']['observation_key'] = observation_key
variant['algo_kwargs']['her_kwargs']['desired_goal_key'] = desired_goal_key
replay_buffer = variant['replay_buffer_class'](
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs']
)
variant['count_based_sampler_kwargs']['replay_buffer'] = replay_buffer
env = CountBasedGoalSamplingEnv(wrapped_env=env, **variant['count_based_sampler_kwargs'])
obs_dim = env.observation_space.spaces['observation'].low.size
action_dim = env.action_space.low.size
goal_dim = env.observation_space.spaces['desired_goal'].low.size
exploration_type = variant['exploration_type']
if exploration_type == 'ou':
es = OUStrategy(
action_space=env.action_space,
**variant['es_kwargs']
)
elif exploration_type == 'gaussian':
es = GaussianStrategy(
action_space=env.action_space,
**variant['es_kwargs'],
)
elif exploration_type == 'epsilon':
es = EpsilonGreedy(
action_space=env.action_space,
**variant['es_kwargs'],
)
else:
raise Exception("Invalid type: " + exploration_type)
qf1 = FlattenMlp(
input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs']
)
qf2 = FlattenMlp(
input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs']
)
policy = TanhMlpPolicy(
input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs']
)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
algorithm = HerTd3(
env,
qf1=qf1,
qf2=qf2,
policy=policy,
training_env=env,
exploration_policy=exploration_policy,
replay_buffer=replay_buffer,
**variant['algo_kwargs']
)
if variant.get("save_video", False):
rollout_function = rf.create_rollout_function(
rf.multitask_rollout,
max_path_length=algorithm.max_path_length,
observation_key=algorithm.observation_key,
desired_goal_key=algorithm.desired_goal_key,
)
video_func = get_video_save_func(
rollout_function,
env,
policy,
variant,
)
algorithm.post_epoch_funcs.append(video_func)
if ptu.gpu_enabled():
algorithm.cuda()
algorithm.train()
def her_td3_experiment(variant):
import multiworld.envs.mujoco
import multiworld.envs.pygame
import railrl.samplers.rollout_functions as rf
import railrl.torch.pytorch_util as ptu
from railrl.exploration_strategies.base import (
PolicyWrappedWithExplorationStrategy)
from railrl.exploration_strategies.epsilon_greedy import EpsilonGreedy
from railrl.exploration_strategies.gaussian_strategy import GaussianStrategy
from railrl.exploration_strategies.ou_strategy import OUStrategy
from railrl.torch.grill.launcher import get_video_save_func
from railrl.torch.her.her_td3 import HerTd3
from railrl.data_management.obs_dict_replay_buffer import (
ObsDictRelabelingBuffer)
if 'env_id' in variant:
env = gym.make(variant['env_id'])
else:
env = variant['env_class'](**variant['env_kwargs'])
imsize = 84
env = GymToMultiEnv(env.env) # unwrap TimeLimit
env = ImageEnv(env, non_presampled_goal_img_is_garbage=True)
observation_key = variant['observation_key']
desired_goal_key = variant['desired_goal_key']
variant['algo_kwargs']['her_kwargs']['observation_key'] = observation_key
variant['algo_kwargs']['her_kwargs']['desired_goal_key'] = desired_goal_key
if variant.get('normalize', False):
raise NotImplementedError()
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
replay_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
obs_dim = env.observation_space.spaces[observation_key].low.size
action_dim = env.action_space.low.size
goal_dim = env.observation_space.spaces[desired_goal_key].low.size
exploration_type = variant['exploration_type']
if exploration_type == 'ou':
es = OUStrategy(action_space=env.action_space, **variant['es_kwargs'])
elif exploration_type == 'gaussian':
es = GaussianStrategy(
action_space=env.action_space,
**variant['es_kwargs'],
)
elif exploration_type == 'epsilon':
es = EpsilonGreedy(
action_space=env.action_space,
**variant['es_kwargs'],
)
else:
raise Exception("Invalid type: " + exploration_type)
use_images_for_q = variant["use_images_for_q"]
use_images_for_pi = variant["use_images_for_pi"]
qs = []
for i in range(2):
if use_images_for_q:
image_q = MergedCNN(input_width=imsize,
input_height=imsize,
output_size=1,
input_channels=3,
added_fc_input_size=action_dim,
**variant['cnn_params'])
q = ImageStateQ(image_q, None)
else:
state_q = FlattenMlp(input_size=action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs'])
q = ImageStateQ(None, state_q)
qs.append(q)
qf1, qf2 = qs
if use_images_for_pi:
image_policy = CNNPolicy(
input_width=imsize,
input_height=imsize,
output_size=action_dim,
input_channels=3,
**variant['cnn_params'],
output_activation=torch.tanh,
)
policy = ImageStatePolicy(image_policy, None)
else:
state_policy = TanhMlpPolicy(input_size=goal_dim,
output_size=action_dim,
**variant['policy_kwargs'])
policy = ImageStatePolicy(None, state_policy)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
algorithm = HerTd3(env,
qf1=qf1,
qf2=qf2,
policy=policy,
exploration_policy=exploration_policy,
replay_buffer=replay_buffer,
**variant['algo_kwargs'])
if variant.get("save_video", False):
rollout_function = rf.create_rollout_function(
rf.multitask_rollout,
max_path_length=algorithm.max_path_length,
observation_key=algorithm.observation_key,
desired_goal_key=algorithm.desired_goal_key,
)
video_func = get_video_save_func(
rollout_function,
env,
policy,
variant,
)
algorithm.post_epoch_funcs.append(video_func)
algorithm.to(ptu.device)
algorithm.train()
def grill_her_td3_experiment(variant):
env = variant["env_class"](**variant['env_kwargs'])
render = variant["render"]
rdim = variant["rdim"]
vae_path = variant["vae_paths"][str(rdim)]
reward_params = variant.get("reward_params", dict())
init_camera = variant.get("init_camera", None)
if init_camera is None:
camera_name = "topview"
else:
camera_name = None
env = ImageEnv(
env,
84,
init_camera=init_camera,
camera_name=camera_name,
transpose=True,
normalize=True,
)
env = VAEWrappedEnv(
env,
vae_path,
decode_goals=render,
render_goals=render,
render_rollouts=render,
reward_params=reward_params,
**variant.get('vae_wrapped_env_kwargs', {})
)
if variant['normalize']:
env = NormalizedBoxEnv(env)
exploration_type = variant['exploration_type']
exploration_noise = variant.get('exploration_noise', 0.1)
if exploration_type == 'ou':
es = OUStrategy(action_space=env.action_space)
elif exploration_type == 'gaussian':
es = GaussianStrategy(
action_space=env.action_space,
max_sigma=exploration_noise,
min_sigma=exploration_noise, # Constant sigma
)
elif exploration_type == 'epsilon':
es = EpsilonGreedy(
action_space=env.action_space,
prob_random_action=exploration_noise,
)
else:
raise Exception("Invalid type: " + exploration_type)
observation_key = variant.get('observation_key', 'latent_observation')
desired_goal_key = variant.get('desired_goal_key', 'latent_desired_goal')
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
obs_dim = (
env.observation_space.spaces[observation_key].low.size
+ env.observation_space.spaces[desired_goal_key].low.size
)
action_dim = env.action_space.low.size
hidden_sizes = variant.get('hidden_sizes', [400, 300])
qf1 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=hidden_sizes,
)
qf2 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=hidden_sizes,
)
policy = TanhMlpPolicy(
input_size=obs_dim,
output_size=action_dim,
hidden_sizes=hidden_sizes,
)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
training_mode = variant.get("training_mode", "train")
testing_mode = variant.get("testing_mode", "test")
testing_env = pickle.loads(pickle.dumps(env))
testing_env.mode(testing_mode)
training_env = pickle.loads(pickle.dumps(env))
training_env.mode(training_mode)
relabeling_env = pickle.loads(pickle.dumps(env))
relabeling_env.mode(training_mode)
relabeling_env.disable_render()
video_vae_env = pickle.loads(pickle.dumps(env))
video_vae_env.mode("video_vae")
video_goal_env = pickle.loads(pickle.dumps(env))
video_goal_env.mode("video_env")
replay_buffer = ObsDictRelabelingBuffer(
env=relabeling_env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_kwargs']
)
variant["algo_kwargs"]["replay_buffer"] = replay_buffer
algorithm = HerTd3(
testing_env,
training_env=training_env,
qf1=qf1,
qf2=qf2,
policy=policy,
exploration_policy=exploration_policy,
render=render,
render_during_eval=render,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
**variant['algo_kwargs']
)
if ptu.gpu_enabled():
print("using GPU")
algorithm.to(ptu.device)
for e in [testing_env, training_env, video_vae_env, video_goal_env]:
e.vae.to(ptu.device)
algorithm.train()
if variant.get("save_video", True):
logdir = logger.get_snapshot_dir()
policy.train(False)
filename = osp.join(logdir, 'video_final_env.mp4')
rollout_function = rf.create_rollout_function(
rf.multitask_rollout,
max_path_length=algorithm.max_path_length,
observation_key=algorithm.observation_key,
desired_goal_key=algorithm.desired_goal_key,
)
dump_video(video_goal_env, policy, filename, rollout_function)
filename = osp.join(logdir, 'video_final_vae.mp4')
dump_video(video_vae_env, policy, filename, rollout_function)
def her_td3_experiment(variant):
env = variant['env_class'](**variant['env_kwargs'])
observation_key = variant.get('observation_key', 'observation')
desired_goal_key = variant.get('desired_goal_key', 'desired_goal')
replay_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
**variant['replay_buffer_kwargs']
)
obs_dim = env.observation_space.spaces['observation'].low.size
action_dim = env.action_space.low.size
goal_dim = env.observation_space.spaces['desired_goal'].low.size
if variant['normalize']:
env = NormalizedBoxEnv(env)
exploration_type = variant['exploration_type']
if exploration_type == 'ou':
es = OUStrategy(
action_space=env.action_space,
max_sigma=0.1,
**variant['es_kwargs']
)
elif exploration_type == 'gaussian':
es = GaussianStrategy(
action_space=env.action_space,
max_sigma=0.1,
min_sigma=0.1, # Constant sigma
**variant['es_kwargs'],
)
elif exploration_type == 'epsilon':
es = EpsilonGreedy(
action_space=env.action_space,
prob_random_action=0.1,
**variant['es_kwargs'],
)
else:
raise Exception("Invalid type: " + exploration_type)
qf1 = FlattenMlp(
input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs']
)
qf2 = FlattenMlp(
input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs']
)
policy = TanhMlpPolicy(
input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs']
)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
algorithm = HerTd3(
env,
qf1=qf1,
qf2=qf2,
policy=policy,
exploration_policy=exploration_policy,
replay_buffer=replay_buffer,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
**variant['algo_kwargs']
)
if ptu.gpu_enabled():
qf1.to(ptu.device)
qf2.to(ptu.device)
policy.to(ptu.device)
algorithm.to(ptu.device)
algorithm.train()
def state_td3bc_experiment(variant):
if variant.get('env_id', None):
import gym
import multiworld
multiworld.register_all_envs()
eval_env = gym.make(variant['env_id'])
expl_env = gym.make(variant['env_id'])
else:
eval_env_kwargs = variant.get('eval_env_kwargs', variant['env_kwargs'])
eval_env = variant['env_class'](**eval_env_kwargs)
expl_env = variant['env_class'](**variant['env_kwargs'])
observation_key = 'state_observation'
desired_goal_key = 'state_desired_goal'
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
es_strat = variant.get('es', 'ou')
if es_strat == 'ou':
es = OUStrategy(
action_space=expl_env.action_space,
max_sigma=variant['exploration_noise'],
min_sigma=variant['exploration_noise'],
)
elif es_strat == 'gauss_eps':
es = GaussianAndEpislonStrategy(
action_space=expl_env.action_space,
max_sigma=.2,
min_sigma=.2, # constant sigma
epsilon=.3,
)
else:
raise ValueError("invalid exploration strategy provided")
obs_dim = expl_env.observation_space.spaces['observation'].low.size
goal_dim = expl_env.observation_space.spaces['desired_goal'].low.size
action_dim = expl_env.action_space.low.size
qf1 = FlattenMlp(
input_size=obs_dim + goal_dim + action_dim,
output_size=1,
**variant['qf_kwargs']
)
qf2 = FlattenMlp(
input_size=obs_dim + goal_dim + action_dim,
output_size=1,
**variant['qf_kwargs']
)
target_qf1 = FlattenMlp(
input_size=obs_dim + goal_dim + action_dim,
output_size=1,
**variant['qf_kwargs']
)
target_qf2 = FlattenMlp(
input_size=obs_dim + goal_dim + action_dim,
output_size=1,
**variant['qf_kwargs']
)
policy = TanhMlpPolicy(
input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs']
)
target_policy = TanhMlpPolicy(
input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs']
)
expl_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
replay_buffer = ObsDictRelabelingBuffer(
env=eval_env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs']
)
demo_train_buffer = ObsDictRelabelingBuffer(
env=eval_env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
max_size=variant['replay_buffer_kwargs']['max_size']
)
demo_test_buffer = ObsDictRelabelingBuffer(
env=eval_env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
max_size=variant['replay_buffer_kwargs']['max_size'],
)
if variant.get('td3_bc', True):
td3_trainer = TD3BCTrainer(
env=expl_env,
policy=policy,
qf1=qf1,
qf2=qf2,
replay_buffer=replay_buffer,
demo_train_buffer=demo_train_buffer,
demo_test_buffer=demo_test_buffer,
target_qf1=target_qf1,
target_qf2=target_qf2,
target_policy=target_policy,
**variant['td3_bc_trainer_kwargs']
)
else:
td3_trainer = TD3(
policy=policy,
qf1=qf1,
qf2=qf2,
target_qf1=target_qf1,
target_qf2=target_qf2,
target_policy=target_policy,
**variant['td3_trainer_kwargs']
)
trainer = HERTrainer(td3_trainer)
eval_path_collector = GoalConditionedPathCollector(
eval_env,
policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
)
expl_path_collector = GoalConditionedPathCollector(
expl_env,
expl_policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
)
algorithm = TorchBatchRLAlgorithm(
trainer=trainer,
exploration_env=expl_env,
evaluation_env=eval_env,
exploration_data_collector=expl_path_collector,
evaluation_data_collector=eval_path_collector,
replay_buffer=replay_buffer,
**variant['algo_kwargs']
)
if variant.get("save_video", True):
if variant.get("presampled_goals", None):
variant['image_env_kwargs']['presampled_goals'] = load_local_or_remote_file(variant['presampled_goals']).item()
image_eval_env = ImageEnv(eval_env, **variant["image_env_kwargs"])
image_eval_path_collector = GoalConditionedPathCollector(
image_eval_env,
policy,
observation_key='state_observation',
desired_goal_key='state_desired_goal',
)
image_expl_env = ImageEnv(expl_env, **variant["image_env_kwargs"])
image_expl_path_collector = GoalConditionedPathCollector(
image_expl_env,
expl_policy,
observation_key='state_observation',
desired_goal_key='state_desired_goal',
)
video_func = VideoSaveFunction(
image_eval_env,
variant,
image_expl_path_collector,
image_eval_path_collector,
)
algorithm.post_train_funcs.append(video_func)
algorithm.to(ptu.device)
if variant.get('load_demos', False):
td3_trainer.load_demos()
if variant.get('pretrain_policy', False):
td3_trainer.pretrain_policy_with_bc()
if variant.get('pretrain_rl', False):
td3_trainer.pretrain_q_with_bc_data()
algorithm.train()
def td3_experiment(variant):
import gym
import multiworld.envs.mujoco
import multiworld.envs.pygame
import railrl.samplers.rollout_functions as rf
import railrl.torch.pytorch_util as ptu
from railrl.exploration_strategies.base import (
PolicyWrappedWithExplorationStrategy)
from railrl.exploration_strategies.epsilon_greedy import EpsilonGreedy
from railrl.exploration_strategies.gaussian_strategy import GaussianStrategy
from railrl.exploration_strategies.ou_strategy import OUStrategy
from railrl.torch.grill.launcher import get_state_experiment_video_save_function
from railrl.torch.her.her_td3 import HerTd3
from railrl.torch.td3.td3 import TD3
from railrl.torch.networks import FlattenMlp, TanhMlpPolicy
from railrl.data_management.obs_dict_replay_buffer import (
ObsDictReplayBuffer)
from railrl.torch.torch_rl_algorithm import TorchBatchRLAlgorithm
from railrl.samplers.data_collector.path_collector import ObsDictPathCollector
if 'env_id' in variant:
eval_env = gym.make(variant['env_id'])
expl_env = gym.make(variant['env_id'])
else:
eval_env_kwargs = variant.get('eval_env_kwargs', variant['env_kwargs'])
eval_env = variant['env_class'](**eval_env_kwargs)
expl_env = variant['env_class'](**variant['env_kwargs'])
observation_key = variant['observation_key']
# desired_goal_key = variant['desired_goal_key']
# variant['algo_kwargs']['her_kwargs']['observation_key'] = observation_key
# variant['algo_kwargs']['her_kwargs']['desired_goal_key'] = desired_goal_key
if variant.get('normalize', False):
raise NotImplementedError()
# achieved_goal_key = desired_goal_key.replace("desired", "achieved")
replay_buffer = ObsDictReplayBuffer(
env=eval_env,
observation_key=observation_key,
# desired_goal_key=desired_goal_key,
# achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
obs_dim = eval_env.observation_space.spaces['observation'].low.size
action_dim = eval_env.action_space.low.size
goal_dim = eval_env.observation_space.spaces['desired_goal'].low.size
exploration_type = variant['exploration_type']
if exploration_type == 'ou':
es = OUStrategy(action_space=eval_env.action_space,
**variant['es_kwargs'])
elif exploration_type == 'gaussian':
es = GaussianStrategy(
action_space=eval_env.action_space,
**variant['es_kwargs'],
)
elif exploration_type == 'epsilon':
es = EpsilonGreedy(
action_space=eval_env.action_space,
**variant['es_kwargs'],
)
else:
raise Exception("Invalid type: " + exploration_type)
qf1 = FlattenMlp(input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs'])
qf2 = FlattenMlp(input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs'])
policy = TanhMlpPolicy(input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs'])
target_qf1 = FlattenMlp(input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs'])
target_qf2 = FlattenMlp(input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs'])
target_policy = TanhMlpPolicy(input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs'])
expl_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
trainer = TD3(policy=policy,
qf1=qf1,
qf2=qf2,
target_qf1=target_qf1,
target_qf2=target_qf2,
target_policy=target_policy,
**variant['trainer_kwargs'])
observation_key = 'observation'
desired_goal_key = 'desired_goal'
eval_path_collector = ObsDictPathCollector(
eval_env,
policy,
observation_key=observation_key,
# render=True,
# desired_goal_key=desired_goal_key,
)
expl_path_collector = ObsDictPathCollector(
expl_env,
expl_policy,
observation_key=observation_key,
# render=True,
# desired_goal_key=desired_goal_key,
)
algorithm = TorchBatchRLAlgorithm(
trainer=trainer,
exploration_env=expl_env,
evaluation_env=eval_env,
exploration_data_collector=expl_path_collector,
evaluation_data_collector=eval_path_collector,
replay_buffer=replay_buffer,
**variant['algo_kwargs'])
# if variant.get("save_video", False):
# rollout_function = rf.create_rollout_function(
# rf.multitask_rollout,
# max_path_length=algorithm.max_path_length,
# observation_key=observation_key,
# desired_goal_key=algorithm.desired_goal_key,
# )
# video_func = get_state_experiment_video_save_function(
# rollout_function,
# env,
# policy,
# variant,
# )
# algorithm.post_epoch_funcs.append(video_func)
algorithm.to(ptu.device)
algorithm.train()
def generate_vae_dataset(
N=10000, test_p=0.9, use_cached=True, imsize=84, show=False,
dataset_path=None, policy_path=None, action_space_sampling=False, env_class=SawyerPushAndPullDoorEnv, env_kwargs=None,
action_plus_random_sampling=False, init_camera=sawyer_door_env_camera, ratio_action_sample_to_random=1 / 2, env_id=None,
):
if policy_path is not None:
filename = "/tmp/sawyer_door_push_and_pull_open_oracle+random_policy_data_closer_zoom_action_limited" + str(N) + ".npy"
elif action_space_sampling:
filename = "/tmp/sawyer_door_push_and_pull_open_zoomed_in_action_space_sampling" + str(N) + ".npy"
else:
filename = "/tmp/sawyer_door_push_and_pull_open" + str(N) + ".npy"
info = {}
if dataset_path is not None:
filename = local_path_from_s3_or_local_path(dataset_path)
dataset = np.load(filename)
elif use_cached and osp.isfile(filename):
dataset = np.load(filename)
print("loaded data from saved file", filename)
elif action_plus_random_sampling:
if env_id is not None:
import gym
env = gym.make(env_id)
else:
env = env_class(**env_kwargs)
env = ImageEnv(
env, imsize,
transpose=True,
init_camera=init_camera,
normalize=True,
)
action_sampled_data = int(N*ratio_action_sample_to_random)
dataset = np.zeros((N, imsize * imsize * 3), dtype=np.uint8)
print('Action Space Sampling')
for i in range(action_sampled_data):
goal = env.sample_goal()
env.set_to_goal(goal)
img = env._get_flat_img()
dataset[i, :] = unormalize_image(img)
if show:
cv2.imshow('img', img.reshape(3, 84, 84).transpose())
cv2.waitKey(1)
print(i)
policy = RandomPolicy(env.action_space)
es = OUStrategy(action_space=env.action_space, theta=0)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
print('Random Sampling')
for i in range(action_sampled_data, N):
if i % 20==0:
env.reset()
exploration_policy.reset()
for _ in range(10):
action = exploration_policy.get_action()[0]
env.wrapped_env.step(
action
)
goal = env.sample_goal()
env.set_to_goal_angle(goal['state_desired_goal'])
img = env._get_flat_img()
dataset[i, :] = unormalize_image(img)
if show:
cv2.imshow('img', img.reshape(3, 84, 84).transpose())
cv2.waitKey(1)
print(i)
env._wrapped_env.min_y_pos = .5
info['env'] = env
else:
raise NotImplementedError()
n = int(N * test_p)
train_dataset = dataset[:n, :]
test_dataset = dataset[n:, :]
return train_dataset, test_dataset, info
def generate_vae_dataset(
N=10000,
test_p=0.9,
use_cached=True,
imsize=84,
show=False,
dataset_path=None,
action_space_sampling=False,
init_camera=None,
env_class=None,
env_kwargs=None,
):
filename = "/tmp/sawyer_xyz_pos_control_new_zoom_cam" + str(N) + '.npy'
info = {}
if dataset_path is not None:
filename = local_path_from_s3_or_local_path(dataset_path)
dataset = np.load(filename)
elif use_cached and osp.isfile(filename):
dataset = np.load(filename)
print("loaded data from saved file", filename)
else:
now = time.time()
if env_kwargs == None:
env_kwargs = dict()
env = env_class(**env_kwargs)
env = ImageEnv(
env,
imsize,
transpose=True,
init_camera=init_camera,
normalize=True,
)
dataset = np.zeros((N, imsize * imsize * 3), dtype=np.uint8)
if action_space_sampling:
action_space = Box(np.array([-.1, .5, 0]), np.array([.1, .7, .5]))
for i in range(N):
env.set_to_goal(env.sample_goal())
img = env._get_flat_img()
dataset[i, :] = unormalize_image(img)
if show:
cv2.imshow('img', img.reshape(3, 84, 84).transpose())
cv2.waitKey(1)
print(i)
info['env'] = env
else:
policy = RandomPolicy(env.action_space)
es = OUStrategy(action_space=env.action_space, theta=0)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
for i in range(N):
# Move the goal out of the image
env.wrapped_env.set_goal(np.array([100, 100, 100]))
if i % 50 == 0:
print('Reset')
env.reset()
exploration_policy.reset()
for _ in range(1):
action = exploration_policy.get_action()[0] * 10
env.wrapped_env.step(action)
img = env.step(env.action_space.sample())[0]
dataset[i, :] = img
if show:
cv2.imshow('img', img.reshape(3, 84, 84).transpose())
cv2.waitKey(1)
print(i)
print("done making training data", time.time() - now)
np.save(filename, dataset)
n = int(N * test_p)
train_dataset = dataset[:n, :]
test_dataset = dataset[n:, :]
return train_dataset, test_dataset, info
def generate_vae_dataset(
N=10000,
test_p=0.9,
use_cached=True,
imsize=84,
show=False,
dataset_path=None,
policy_path=None,
ratio_oracle_policy_data_to_random=1 / 2,
action_space_sampling=False,
env_class=None,
env_kwargs=None,
action_plus_random_sampling=False,
init_camera=sawyer_door_env_camera,
):
if policy_path is not None:
filename = "/tmp/sawyer_door_push_open_oracle+random_policy_data_closer_zoom_action_limited" + str(
N) + ".npy"
elif action_space_sampling:
filename = "/tmp/sawyer_door_push_open_zoomed_in_action_space_sampling" + str(
N) + ".npy"
else:
filename = "/tmp/sawyer_door_push_open" + str(N) + ".npy"
info = {}
if dataset_path is not None:
filename = local_path_from_s3_or_local_path(dataset_path)
dataset = np.load(filename)
elif use_cached and osp.isfile(filename):
dataset = np.load(filename)
print("loaded data from saved file", filename)
elif action_space_sampling:
env = SawyerDoorPushOpenEnv(**env_kwargs)
env = ImageEnv(
env,
imsize,
transpose=False,
init_camera=sawyer_door_env_camera,
normalize=False,
)
action_space = Box(np.array([-env.max_x_pos, .5, .06]),
np.array([env.max_x_pos, env.max_y_pos, .06]))
dataset = np.zeros((N, imsize * imsize * 3))
for i in range(N):
env.set_to_goal_pos(action_space.sample()) #move arm to spot
goal = env.sample_goal()
env.set_to_goal(goal)
img = env.get_image().flatten()
dataset[i, :] = img
if show:
cv2.imshow('img', img.reshape(3, 84, 84).transpose())
cv2.waitKey(1)
print(i)
info['env'] = env
elif action_plus_random_sampling:
env = env_class(**env_kwargs)
env = ImageEnv(
env,
imsize,
transpose=True,
init_camera=init_camera,
normalize=True,
)
action_space = Box(np.array([-env.max_x_pos, .5, .06]),
np.array([env.max_x_pos, .6, .06]))
action_sampled_data = int(N / 2)
dataset = np.zeros((N, imsize * imsize * 3))
print('Action Space Sampling')
for i in range(action_sampled_data):
env.set_to_goal_pos(action_space.sample()) # move arm to spot
goal = env.sample_goal()
env.set_to_goal(goal)
img = env._get_flat_img()
dataset[i, :] = img
if show:
cv2.imshow('img', img.reshape(3, 84, 84).transpose())
cv2.waitKey(1)
print(i)
env._wrapped_env.min_y_pos = .6
policy = RandomPolicy(env.action_space)
es = OUStrategy(action_space=env.action_space, theta=0)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
print('Random Sampling')
for i in range(action_sampled_data, N):
if i % 20 == 0:
env.reset()
exploration_policy.reset()
for _ in range(10):
action = exploration_policy.get_action()[0]
env.wrapped_env.step(action)
img = env._get_flat_img()
dataset[i, :] = img
if show:
cv2.imshow('img', img.reshape(3, 84, 84).transpose())
cv2.waitKey(1)
print(i)
env._wrapped_env.min_y_pos = .5
info['env'] = env
else:
now = time.time()
env = SawyerDoorPushOpenEnv(max_angle=.5)
env = ImageEnv(
env,
imsize,
transpose=True,
init_camera=sawyer_door_env_camera,
normalize=True,
)
info['env'] = env
policy = RandomPolicy(env.action_space)
es = OUStrategy(action_space=env.action_space, theta=0)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
dataset = np.zeros((N, imsize * imsize * 3))
for i in range(N):
if i % 100 == 0:
env.reset()
exploration_policy.reset()
for _ in range(25):
# env.wrapped_env.step(
# env.wrapped_env.action_space.sample()
# )
action = exploration_policy.get_action()[0]
env.wrapped_env.step(action)
goal = env.sample_goal_for_rollout()
env.set_to_goal(goal)
img = env.step(env.action_space.sample())[0]
dataset[i, :] = img
if show:
cv2.imshow('img', img.reshape(3, 84, 84).transpose())
cv2.waitKey(1)
print(i)
print("done making training data", filename, time.time() - now)
np.save(filename, dataset)
n = int(N * test_p)
train_dataset = dataset[:n, :]
test_dataset = dataset[n:, :]
return train_dataset, test_dataset, info
def grill_her_td3_experiment(variant):
import railrl.samplers.rollout_functions as rf
import railrl.torch.pytorch_util as ptu
from railrl.data_management.obs_dict_replay_buffer import \
ObsDictRelabelingBuffer
from railrl.exploration_strategies.base import (
PolicyWrappedWithExplorationStrategy)
from railrl.demos.her_td3bc import HerTD3BC
from railrl.torch.networks import FlattenMlp, TanhMlpPolicy
grill_preprocess_variant(variant)
env = get_envs(variant)
es = get_exploration_strategy(variant, env)
observation_key = variant.get('observation_key', 'latent_observation')
desired_goal_key = variant.get('desired_goal_key', 'latent_desired_goal')
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
obs_dim = (env.observation_space.spaces[observation_key].low.size +
env.observation_space.spaces[desired_goal_key].low.size)
action_dim = env.action_space.low.size
qf1 = FlattenMlp(input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
qf2 = FlattenMlp(input_size=obs_dim + action_dim,
output_size=1,
**variant['qf_kwargs'])
policy = TanhMlpPolicy(input_size=obs_dim,
output_size=action_dim,
**variant['policy_kwargs'])
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
replay_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
demo_train_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
demo_test_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
algo_kwargs = variant['algo_kwargs']
algo_kwargs['replay_buffer'] = replay_buffer
base_kwargs = algo_kwargs['base_kwargs']
base_kwargs['training_env'] = env
base_kwargs['render'] = variant["render"]
base_kwargs['render_during_eval'] = variant["render"]
her_kwargs = algo_kwargs['her_kwargs']
her_kwargs['observation_key'] = observation_key
her_kwargs['desired_goal_key'] = desired_goal_key
# algorithm = HerTd3(
# env,
# qf1=qf1,
# qf2=qf2,
# policy=policy,
# exploration_policy=exploration_policy,
# **variant['algo_kwargs']
# )
env.vae.to(ptu.device)
algorithm = HerTD3BC(env,
qf1=qf1,
qf2=qf2,
policy=policy,
exploration_policy=exploration_policy,
demo_train_buffer=demo_train_buffer,
demo_test_buffer=demo_test_buffer,
demo_path=variant["demo_path"],
add_demo_latents=True,
**variant['algo_kwargs'])
if variant.get("save_video", True):
rollout_function = rf.create_rollout_function(
rf.multitask_rollout,
max_path_length=algorithm.max_path_length,
observation_key=algorithm.observation_key,
desired_goal_key=algorithm.desired_goal_key,
)
video_func = get_video_save_func(
rollout_function,
env,
algorithm.eval_policy,
variant,
)
algorithm.post_epoch_funcs.append(video_func)
algorithm.to(ptu.device)
if not variant.get("do_state_exp", False):
env.vae.to(ptu.device)
algorithm.train()
def grill_her_td3_experiment_offpolicy_online_vae(variant):
import railrl.torch.pytorch_util as ptu
from railrl.data_management.online_vae_replay_buffer import \
OnlineVaeRelabelingBuffer
from railrl.torch.networks import FlattenMlp, TanhMlpPolicy
from railrl.torch.vae.vae_trainer import ConvVAETrainer
from railrl.torch.td3.td3 import TD3
from railrl.exploration_strategies.base import (
PolicyWrappedWithExplorationStrategy)
from railrl.exploration_strategies.gaussian_and_epislon import \
GaussianAndEpislonStrategy
from railrl.torch.vae.online_vae_offpolicy_algorithm import OnlineVaeOffpolicyAlgorithm
import gc
gc.collect() # Ashvin: this line for a GPU memory error
grill_preprocess_variant(variant)
env = get_envs(variant)
uniform_dataset_fn = variant.get('generate_uniform_dataset_fn', None)
if uniform_dataset_fn:
uniform_dataset = uniform_dataset_fn(
**variant['generate_uniform_dataset_kwargs'])
else:
uniform_dataset = None
observation_key = variant.get('observation_key', 'latent_observation')
desired_goal_key = variant.get('desired_goal_key', 'latent_desired_goal')
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
obs_dim = (env.observation_space.spaces[observation_key].low.size +
env.observation_space.spaces[desired_goal_key].low.size)
action_dim = env.action_space.low.size
hidden_sizes = variant.get('hidden_sizes', [400, 300])
qf1 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=hidden_sizes,
)
qf2 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=hidden_sizes,
)
target_qf1 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=hidden_sizes,
)
target_qf2 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=hidden_sizes,
)
policy = TanhMlpPolicy(
input_size=obs_dim,
output_size=action_dim,
hidden_sizes=hidden_sizes,
# **variant['policy_kwargs']
)
target_policy = TanhMlpPolicy(
input_size=obs_dim,
output_size=action_dim,
hidden_sizes=hidden_sizes,
# **variant['policy_kwargs']
)
#es = get_exploration_strategy(varient, env)
es = GaussianAndEpislonStrategy(
action_space=env.action_space,
max_sigma=.2,
min_sigma=.2, # constant sigma
epsilon=variant.get('exploration_noise', 0.1),
)
expl_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
vae = env.vae
replay_buffer_class = variant.get("replay_buffer_class",
OnlineVaeRelabelingBuffer)
replay_buffer = replay_buffer_class(vae=env.vae,
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
replay_buffer.representation_size = vae.representation_size
vae_trainer_class = variant.get("vae_trainer_class", ConvVAETrainer)
vae_trainer = vae_trainer_class(env.vae,
**variant['online_vae_trainer_kwargs'])
assert 'vae_training_schedule' not in variant, "Just put it in algo_kwargs"
max_path_length = variant['max_path_length']
trainer = TD3(policy=policy,
qf1=qf1,
qf2=qf2,
target_qf1=target_qf1,
target_qf2=target_qf2,
target_policy=target_policy,
**variant['td3_trainer_kwargs'])
trainer = HERTrainer(trainer)
eval_path_collector = VAEWrappedEnvPathCollector(
env,
policy,
max_path_length,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
goal_sampling_mode=variant['evaluation_goal_sampling_mode'],
)
expl_path_collector = VAEWrappedEnvPathCollector(
env,
expl_policy,
max_path_length,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
goal_sampling_mode=variant['exploration_goal_sampling_mode'],
)
algorithm = OnlineVaeOffpolicyAlgorithm(
trainer=trainer,
exploration_env=env,
evaluation_env=env,
exploration_data_collector=expl_path_collector,
evaluation_data_collector=eval_path_collector,
replay_buffer=replay_buffer,
vae=vae,
vae_trainer=vae_trainer,
uniform_dataset=uniform_dataset,
max_path_length=max_path_length,
**variant['algo_kwargs'])
if variant.get("save_video", True):
video_func = VideoSaveFunction(
env,
variant,
)
algorithm.post_train_funcs.append(video_func)
if variant['custom_goal_sampler'] == 'replay_buffer':
env.custom_goal_sampler = replay_buffer.sample_buffer_goals
algorithm.to(ptu.device)
vae.to(ptu.device)
algorithm.pretrain()
algorithm.train()
def her_td3_experiment(variant):
import gym
import multiworld.envs.mujoco
import multiworld.envs.pygame
import railrl.samplers.rollout_functions as rf
import railrl.torch.pytorch_util as ptu
from railrl.exploration_strategies.base import (
PolicyWrappedWithExplorationStrategy)
from railrl.exploration_strategies.epsilon_greedy import EpsilonGreedy
from railrl.exploration_strategies.gaussian_strategy import GaussianStrategy
from railrl.exploration_strategies.ou_strategy import OUStrategy
from railrl.torch.grill.launcher import get_video_save_func
from railrl.demos.her_td3bc import HerTD3BC
from railrl.torch.networks import FlattenMlp, TanhMlpPolicy
from railrl.data_management.obs_dict_replay_buffer import (
ObsDictRelabelingBuffer)
if 'env_id' in variant:
env = gym.make(variant['env_id'])
else:
env = variant['env_class'](**variant['env_kwargs'])
observation_key = variant['observation_key']
desired_goal_key = variant['desired_goal_key']
variant['algo_kwargs']['her_kwargs']['observation_key'] = observation_key
variant['algo_kwargs']['her_kwargs']['desired_goal_key'] = desired_goal_key
if variant.get('normalize', False):
raise NotImplementedError()
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
replay_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs'])
obs_dim = env.observation_space.spaces['observation'].low.size
action_dim = env.action_space.low.size
goal_dim = env.observation_space.spaces['desired_goal'].low.size
exploration_type = variant['exploration_type']
if exploration_type == 'ou':
es = OUStrategy(action_space=env.action_space, **variant['es_kwargs'])
elif exploration_type == 'gaussian':
es = GaussianStrategy(
action_space=env.action_space,
**variant['es_kwargs'],
)
elif exploration_type == 'epsilon':
es = EpsilonGreedy(
action_space=env.action_space,
**variant['es_kwargs'],
)
else:
raise Exception("Invalid type: " + exploration_type)
qf1 = FlattenMlp(input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs'])
qf2 = FlattenMlp(input_size=obs_dim + action_dim + goal_dim,
output_size=1,
**variant['qf_kwargs'])
policy = TanhMlpPolicy(input_size=obs_dim + goal_dim,
output_size=action_dim,
**variant['policy_kwargs'])
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
algorithm = HerTD3BC(env,
qf1=qf1,
qf2=qf2,
policy=policy,
exploration_policy=exploration_policy,
replay_buffer=replay_buffer,
demo_path=variant["demo_path"],
**variant['algo_kwargs'])
if variant.get("save_video", False):
rollout_function = rf.create_rollout_function(
rf.multitask_rollout,
max_path_length=algorithm.max_path_length,
observation_key=algorithm.observation_key,
desired_goal_key=algorithm.desired_goal_key,
)
video_func = get_video_save_func(
rollout_function,
env,
policy,
variant,
)
algorithm.post_epoch_funcs.append(video_func)
algorithm.to(ptu.device)
algorithm.train()
def experiment(variant):
env_params = ENV_PARAMS[variant['env']]
variant.update(env_params)
expl_env = NormalizedBoxEnv(variant['env_class']())
eval_env = NormalizedBoxEnv(variant['env_class']())
obs_dim = expl_env.observation_space.low.size
action_dim = eval_env.action_space.low.size
M = variant['layer_size']
es = GaussianAndEpislonStrategy(
action_space=expl_env.action_space,
max_sigma=.2,
min_sigma=.2, # constant sigma
epsilon=.3,
)
qf1 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=[M, M],
)
qf2 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=[M, M],
)
target_qf1 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=[M, M],
)
target_qf2 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=[M, M],
)
policy = TanhMlpPolicy(
input_size=obs_dim,
output_size=action_dim,
hidden_sizes=[M, M],
)
target_policy = TanhMlpPolicy(
input_size=obs_dim,
output_size=action_dim,
hidden_sizes=[M, M],
)
expl_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
eval_path_collector = MdpPathCollector(
eval_env,
policy,
)
replay_buffer = EnvReplayBuffer(
variant['replay_buffer_size'],
expl_env,
)
trainer = TD3(
policy=policy,
qf1=qf1,
qf2=qf2,
target_qf1=target_qf1,
target_qf2=target_qf2,
target_policy=target_policy,
**variant['trainer_kwargs']
)
if variant['collection_mode'] == 'online':
expl_path_collector = MdpStepCollector(
expl_env,
expl_policy,
)
algorithm = TorchOnlineRLAlgorithm(
trainer=trainer,
exploration_env=expl_env,
evaluation_env=eval_env,
exploration_data_collector=expl_path_collector,
evaluation_data_collector=eval_path_collector,
replay_buffer=replay_buffer,
max_path_length=variant['max_path_length'],
batch_size=variant['batch_size'],
num_epochs=variant['num_epochs'],
num_eval_steps_per_epoch=variant['num_eval_steps_per_epoch'],
num_expl_steps_per_train_loop=variant['num_expl_steps_per_train_loop'],
num_trains_per_train_loop=variant['num_trains_per_train_loop'],
min_num_steps_before_training=variant['min_num_steps_before_training'],
)
else:
expl_path_collector = MdpPathCollector(
expl_env,
expl_policy,
)
algorithm = TorchBatchRLAlgorithm(
trainer=trainer,
exploration_env=expl_env,
evaluation_env=eval_env,
exploration_data_collector=expl_path_collector,
evaluation_data_collector=eval_path_collector,
replay_buffer=replay_buffer,
max_path_length=variant['max_path_length'],
batch_size=variant['batch_size'],
num_epochs=variant['num_epochs'],
num_eval_steps_per_epoch=variant['num_eval_steps_per_epoch'],
num_expl_steps_per_train_loop=variant['num_expl_steps_per_train_loop'],
num_trains_per_train_loop=variant['num_trains_per_train_loop'],
min_num_steps_before_training=variant['min_num_steps_before_training'],
)
algorithm.to(ptu.device)
algorithm.train()
def tdm_td3_experiment(variant):
import railrl.samplers.rollout_functions as rf
import railrl.torch.pytorch_util as ptu
from railrl.data_management.obs_dict_replay_buffer import \
ObsDictRelabelingBuffer
from railrl.exploration_strategies.base import (
PolicyWrappedWithExplorationStrategy
)
from railrl.state_distance.tdm_networks import TdmQf, TdmPolicy
from railrl.state_distance.tdm_td3 import TdmTd3
from railrl.state_distance.subgoal_planner import SubgoalPlanner
from railrl.misc.asset_loader import local_path_from_s3_or_local_path
import joblib
preprocess_rl_variant(variant)
env = get_envs(variant)
es = get_exploration_strategy(variant, env)
observation_key = variant.get('observation_key', 'latent_observation')
desired_goal_key = variant.get('desired_goal_key', 'latent_desired_goal')
achieved_goal_key = desired_goal_key.replace("desired", "achieved")
vectorized = 'vectorized' in env.reward_type
variant['algo_kwargs']['tdm_kwargs']['vectorized'] = vectorized
variant['replay_buffer_kwargs']['vectorized'] = vectorized
if 'ckpt' in variant:
if 'ckpt_epoch' in variant:
epoch = variant['ckpt_epoch']
filename = local_path_from_s3_or_local_path(osp.join(variant['ckpt'], 'itr_%d.pkl' % epoch))
else:
filename = local_path_from_s3_or_local_path(osp.join(variant['ckpt'], 'params.pkl'))
print("Loading ckpt from", filename)
data = joblib.load(filename)
qf1 = data['qf1']
qf2 = data['qf2']
policy = data['policy']
variant['algo_kwargs']['base_kwargs']['reward_scale'] = policy.reward_scale
else:
obs_dim = (
env.observation_space.spaces[observation_key].low.size
)
goal_dim = (
env.observation_space.spaces[desired_goal_key].low.size
)
action_dim = env.action_space.low.size
variant['qf_kwargs']['vectorized'] = vectorized
norm_order = env.norm_order
variant['qf_kwargs']['norm_order'] = norm_order
env.reset()
_, rew, _, _ = env.step(env.action_space.sample())
if hasattr(rew, "__len__"):
variant['qf_kwargs']['output_dim'] = len(rew)
qf1 = TdmQf(
env=env,
observation_dim=obs_dim,
goal_dim=goal_dim,
action_dim=action_dim,
**variant['qf_kwargs']
)
qf2 = TdmQf(
env=env,
observation_dim=obs_dim,
goal_dim=goal_dim,
action_dim=action_dim,
**variant['qf_kwargs']
)
policy = TdmPolicy(
env=env,
observation_dim=obs_dim,
goal_dim=goal_dim,
action_dim=action_dim,
reward_scale=variant['algo_kwargs']['base_kwargs'].get('reward_scale', 1.0),
**variant['policy_kwargs']
)
eval_policy = None
if variant.get('eval_policy', None) == 'SubgoalPlanner':
eval_policy = SubgoalPlanner(
env,
qf1,
policy,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
state_based=variant.get("do_state_exp", False),
max_tau=variant['algo_kwargs']['tdm_kwargs']['max_tau'],
reward_scale=variant['algo_kwargs']['base_kwargs'].get('reward_scale', 1.0),
**variant['SubgoalPlanner_kwargs']
)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
replay_buffer = ObsDictRelabelingBuffer(
env=env,
observation_key=observation_key,
desired_goal_key=desired_goal_key,
achieved_goal_key=achieved_goal_key,
**variant['replay_buffer_kwargs']
)
algo_kwargs = variant['algo_kwargs']
algo_kwargs['replay_buffer'] = replay_buffer
base_kwargs = algo_kwargs['base_kwargs']
base_kwargs['training_env'] = env
base_kwargs['render'] = variant.get("render", False)
base_kwargs['render_during_eval'] = variant.get("render_during_eval", False)
tdm_kwargs = algo_kwargs['tdm_kwargs']
tdm_kwargs['observation_key'] = observation_key
tdm_kwargs['desired_goal_key'] = desired_goal_key
algorithm = TdmTd3(
env,
qf1=qf1,
qf2=qf2,
policy=policy,
exploration_policy=exploration_policy,
eval_policy=eval_policy,
**variant['algo_kwargs']
)
if variant.get("test_ckpt", False):
algorithm.post_epoch_funcs.append(get_update_networks_func(variant))
vis_variant = variant.get('vis_kwargs', {})
vis_list = vis_variant.get('vis_list', [])
if vis_variant.get("save_video", True):
rollout_function = rf.create_rollout_function(
rf.tdm_rollout,
init_tau=algorithm._sample_max_tau_for_rollout(),
decrement_tau=algorithm.cycle_taus_for_rollout,
cycle_tau=algorithm.cycle_taus_for_rollout,
max_path_length=algorithm.max_path_length,
observation_key=algorithm.observation_key,
desired_goal_key=algorithm.desired_goal_key,
vis_list=vis_list,
dont_terminate=True,
)
video_func = get_video_save_func(
rollout_function,
env,
variant,
)
algorithm.post_epoch_funcs.append(video_func)
if ptu.gpu_enabled():
print("using GPU")
algorithm.cuda()
if not variant.get("do_state_exp", False):
env.vae.cuda()
env.reset()
if not variant.get("do_state_exp", False):
env.dump_samples(epoch=None)
env.dump_reconstructions(epoch=None)
env.dump_latent_plots(epoch=None)
algorithm.train()
def experiment(variant):
rdim = variant["rdim"]
use_env_goals = variant["use_env_goals"]
vae_path = variant["vae_paths"][str(rdim)]
render = variant["render"]
wrap_mujoco_env = variant.get("wrap_mujoco_env", False)
# vae = torch.load(vae_path)
# print("loaded", vae_path)
from railrl.envs.wrappers import ImageMujocoEnv, NormalizedBoxEnv
from railrl.images.camera import sawyer_init_camera
env = variant["env"](**variant['env_kwargs'])
env = NormalizedBoxEnv(ImageMujocoEnv(
env,
imsize=84,
keep_prev=0,
init_camera=sawyer_init_camera,
))
if wrap_mujoco_env:
env = ImageMujocoEnv(env, 84, camera_name="topview", transpose=True, normalize=True)
if use_env_goals:
track_qpos_goal = variant.get("track_qpos_goal", 0)
env = VAEWrappedImageGoalEnv(env, vae_path, use_vae_obs=True,
use_vae_reward=True, use_vae_goals=True,
render_goals=render, render_rollouts=render, track_qpos_goal=track_qpos_goal)
else:
env = VAEWrappedEnv(env, vae_path, use_vae_obs=True,
use_vae_reward=True, use_vae_goals=True,
render_goals=render, render_rollouts=render)
env = MultitaskToFlatEnv(env)
if variant['normalize']:
env = NormalizedBoxEnv(env)
exploration_type = variant['exploration_type']
if exploration_type == 'ou':
es = OUStrategy(action_space=env.action_space)
elif exploration_type == 'gaussian':
es = GaussianStrategy(
action_space=env.action_space,
max_sigma=0.1,
min_sigma=0.1, # Constant sigma
)
elif exploration_type == 'epsilon':
es = EpsilonGreedy(
action_space=env.action_space,
prob_random_action=0.1,
)
else:
raise Exception("Invalid type: " + exploration_type)
obs_dim = env.observation_space.low.size
action_dim = env.action_space.low.size
qf1 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=[400, 300],
)
qf2 = FlattenMlp(
input_size=obs_dim + action_dim,
output_size=1,
hidden_sizes=[400, 300],
)
policy = TanhMlpPolicy(
input_size=obs_dim,
output_size=action_dim,
hidden_sizes=[400, 300],
)
exploration_policy = PolicyWrappedWithExplorationStrategy(
exploration_strategy=es,
policy=policy,
)
algorithm = TD3(
env,
training_env=env,
qf1=qf1,
qf2=qf2,
policy=policy,
exploration_policy=exploration_policy,
**variant['algo_kwargs']
)
algorithm.to(ptu.device)
env._wrapped_env.vae.to(ptu.device)
