lerobot/lerobot/common/policies/sac/modeling_sac.py

693 lines
27 KiB
Python

#!/usr/bin/env python
# Copyright 2024 The HuggingFace Inc. team.
# All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# TODO: (1) better device management
from collections import deque
from copy import deepcopy
import math
from typing import Callable, Optional, Sequence, Tuple
import einops
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F # noqa: N812
from huggingface_hub import PyTorchModelHubMixin
from torch import Tensor
from lerobot.common.policies.normalize import Normalize, Unnormalize
from lerobot.common.policies.sac.configuration_sac import SACConfig
class SACPolicy(
nn.Module,
PyTorchModelHubMixin,
library_name="lerobot",
repo_url="https://github.com/huggingface/lerobot",
tags=["robotics", "RL", "SAC"],
):
name = "sac"
def __init__(
self, config: SACConfig | None = None, dataset_stats: dict[str, dict[str, Tensor]] | None = None
):
super().__init__()
if config is None:
config = SACConfig()
self.config = config
if config.input_normalization_modes is not None:
self.normalize_inputs = Normalize(
config.input_shapes, config.input_normalization_modes, dataset_stats
)
else:
self.normalize_inputs = nn.Identity()
self.normalize_targets = Normalize(
config.output_shapes, config.output_normalization_modes, dataset_stats
)
self.unnormalize_outputs = Unnormalize(
config.output_shapes, config.output_normalization_modes, dataset_stats
)
encoder = SACObservationEncoder(config)
# Define networks
critic_nets = []
for _ in range(config.num_critics):
critic_net = Critic(encoder=encoder, network=MLP(**config.critic_network_kwargs))
critic_nets.append(critic_net)
self.critic_ensemble = create_critic_ensemble(critic_nets, config.num_critics)
self.critic_target = deepcopy(self.critic_ensemble)
self.actor = Policy(
encoder=encoder,
network=MLP(**config.actor_network_kwargs),
action_dim=config.output_shapes["action"][0],
**config.policy_kwargs,
)
if config.target_entropy is None:
config.target_entropy = -np.prod(config.output_shapes["action"][0]) # (-dim(A))
self.temperature = LagrangeMultiplier(init_value=config.temperature_init)
def reset(self):
"""
Clear observation and action queues. Should be called on `env.reset()`
queues are populated during rollout of the policy, they contain the n latest observations and actions
"""
self._queues = {
"observation.state": deque(maxlen=1),
"action": deque(maxlen=1),
}
if "observation.image" in self.config.input_shapes:
self._queues["observation.image"] = deque(maxlen=1)
if "observation.environment_state" in self.config.input_shapes:
self._queues["observation.environment_state"] = deque(maxlen=1)
@torch.no_grad()
def select_action(self, batch: dict[str, Tensor]) -> Tensor:
"""Select action for inference/evaluation"""
distribution = self.actor(batch)
# Sample from the distribution and return just the actions
actions = distribution.mode() # or distribution.sample() for stochastic actions
actions = self.unnormalize_outputs({"action": actions})["action"]
return actions
def forward(self, batch: dict[str, Tensor]) -> dict[str, Tensor | float]:
"""Run the batch through the model and compute the loss.
Returns a dictionary with loss as a tensor, and other information as native floats.
"""
batch = self.normalize_inputs(batch)
# batch shape is (b, 2, ...) where index 1 returns the current observation and
# the next observation for caluculating the right td index.
actions = batch["action"][:, 0]
rewards = batch["next.reward"][:, 0]
observations = {}
next_observations = {}
for k in batch:
if k.startswith("observation."):
observations[k] = batch[k][:, 0]
next_observations[k] = batch[k][:, 1]
# perform image augmentation
# reward bias from HIL-SERL code base
# add_or_replace={"rewards": batch["rewards"] + self.config["reward_bias"]} in reward_batch
# calculate critics loss
# 1- compute actions from policy
distribution = self.actor(observations)
action_preds = distribution.sample()
log_probs = distribution.log_prob(action_preds)
action_preds = torch.clamp(action_preds, -1, +1)
# 2- compute q targets
q_targets = self.target_qs(next_observations, action_preds)
# subsample critics to prevent overfitting if use high UTD (update to date)
if self.config.num_subsample_critics is not None:
indices = torch.randperm(self.config.num_critics)
indices = indices[:self.config.num_subsample_critics]
q_targets = q_targets[indices]
# critics subsample size
min_q = q_targets.min(dim=0)
# compute td target
td_target = rewards + self.discount * min_q
# 3- compute predicted qs
q_preds = self.critic_ensemble(observations, actions)
# 4- Calculate loss
# Compute state-action value loss (TD loss) for all of the Q functions in the ensemble.
#critics_loss = (
# (
# F.mse_loss(
# q_preds,
# einops.repeat(td_target, "t b -> e t b", e=q_preds.shape[0]),
# reduction="none",
# ).sum(0) # sum over ensemble
# # `q_preds_ensemble` depends on the first observation and the actions.
# * ~batch["observation.state_is_pad"][0]
# * ~batch["action_is_pad"]
# # q_targets depends on the reward and the next observations.
# * ~batch["next.reward_is_pad"]
# * ~batch["observation.state_is_pad"][1:]
# )
# .sum(0)
# .mean()
#)
# 4- Calculate loss
# Compute state-action value loss (TD loss) for all of the Q functions in the ensemble.
critics_loss = F.mse_loss(
q_preds, # shape: [num_critics, batch_size]
einops.repeat(td_target, "b -> e b", e=q_preds.shape[0]), # expand td_target to match q_preds shape
reduction="none"
).sum(0).mean()
# breakpoint()
# calculate actors loss
# 1- temperature
temperature = self.temperature()
# 2- get actions (batch_size, action_dim) and log probs (batch_size,)
<<<<<<< HEAD
actions, log_probs = self.actor(observations) \
=======
distribution = self.actor(observations)
actions = distribution.sample()
log_probs = distribution.log_prob(actions)
actions = torch.clamp(actions, -1, +1)
>>>>>>> d3c62b92 (Refactor SACPolicy for improved action sampling and standard deviation handling)
# 3- get q-value predictions
with torch.no_grad():
q_preds = self.critic_ensemble(observations, actions, return_type="mean")
actor_loss = (
-(q_preds - temperature * log_probs).mean()
* ~batch["observation.state_is_pad"][0]
* ~batch["action_is_pad"]
).mean()
# calculate temperature loss
# 1- calculate entropy
entropy = -log_probs.mean()
temperature_loss = self.temp(lhs=entropy, rhs=self.config.target_entropy)
loss = critics_loss + actor_loss + temperature_loss
return {
"critics_loss": critics_loss.item(),
"actor_loss": actor_loss.item(),
"temperature_loss": temperature_loss.item(),
"temperature": temperature.item(),
"entropy": entropy.item(),
"loss": loss,
}
def update(self):
self.critic_target.lerp_(self.critic_ensemble, self.config.critic_target_update_weight)
# TODO: implement UTD update
# First update only critics for utd_ratio-1 times
# for critic_step in range(self.config.utd_ratio - 1):
# only update critic and critic target
# Then update critic, critic target, actor and temperature
# for target_param, param in zip(self.critic_target.parameters(), self.critic_ensemble.parameters()):
# target_param.data.copy_(target_param.data * (1.0 - self.config.critic_target_update_weight) + param.data * self.critic_target_update_weight)
class MLP(nn.Module):
def __init__(
self,
hidden_dims: list[int],
activations: Callable[[torch.Tensor], torch.Tensor] | str = nn.SiLU(),
activate_final: bool = False,
dropout_rate: Optional[float] = None,
):
super().__init__()
self.activate_final = activate_final
layers = []
for i, size in enumerate(hidden_dims):
layers.append(nn.Linear(hidden_dims[i-1] if i > 0 else hidden_dims[0], size))
if i + 1 < len(hidden_dims) or activate_final:
if dropout_rate is not None and dropout_rate > 0:
layers.append(nn.Dropout(p=dropout_rate))
layers.append(nn.LayerNorm(size))
layers.append(
activations if isinstance(activations, nn.Module) else getattr(nn, activations)()
)
self.net = nn.Sequential(*layers)
def forward(self, x: torch.Tensor, train: bool = False) -> torch.Tensor:
# in training mode or not. TODO: find better way to do this
self.train(train)
return self.net(x)
class Critic(nn.Module):
def __init__(
self,
encoder: Optional[nn.Module],
network: nn.Module,
init_final: Optional[float] = None,
activate_final: bool = False,
device: str = "cuda",
):
super().__init__()
self.device = torch.device(device)
self.encoder = encoder
self.network = network
self.init_final = init_final
self.activate_final = activate_final
# Find the last Linear layer's output dimension
for layer in reversed(network.net):
if isinstance(layer, nn.Linear):
out_features = layer.out_features
break
# Output layer
if init_final is not None:
self.output_layer = nn.Linear(out_features, 1)
nn.init.uniform_(self.output_layer.weight, -init_final, init_final)
nn.init.uniform_(self.output_layer.bias, -init_final, init_final)
else:
self.output_layer = nn.Linear(out_features, 1)
orthogonal_init()(self.output_layer.weight)
self.to(self.device)
def forward(self, observations: torch.Tensor, actions: torch.Tensor, train: bool = False) -> torch.Tensor:
self.train(train)
observations = observations.to(self.device)
actions = actions.to(self.device)
obs_enc = observations if self.encoder is None else self.encoder(observations)
inputs = torch.cat([obs_enc, actions], dim=-1)
x = self.network(inputs)
value = self.output_layer(x)
return value.squeeze(-1)
def q_value_ensemble(
self, observations: torch.Tensor, actions: torch.Tensor, train: bool = False
) -> torch.Tensor:
observations = observations.to(self.device)
actions = actions.to(self.device)
if len(actions.shape) == 3: # [batch_size, num_actions, action_dim]
batch_size, num_actions = actions.shape[:2]
obs_expanded = observations.unsqueeze(1).expand(-1, num_actions, -1)
obs_flat = obs_expanded.reshape(-1, observations.shape[-1])
actions_flat = actions.reshape(-1, actions.shape[-1])
q_values = self(obs_flat, actions_flat, train)
return q_values.reshape(batch_size, num_actions)
else:
return self(observations, actions, train)
class Policy(nn.Module):
def __init__(
self,
encoder: Optional[nn.Module],
network: nn.Module,
action_dim: int,
std_parameterization: str = "exp",
std_min: float = 0.05,
std_max: float = 2.0,
tanh_squash_distribution: bool = False,
fixed_std: Optional[torch.Tensor] = None,
init_final: Optional[float] = None,
activate_final: bool = False,
device: str = "cuda",
):
super().__init__()
self.device = torch.device(device)
self.encoder = encoder
self.network = network
self.action_dim = action_dim
self.std_parameterization = std_parameterization
self.std_min = std_min
self.std_max = std_max
self.tanh_squash_distribution = tanh_squash_distribution
self.fixed_std = fixed_std.to(self.device) if fixed_std is not None else None
self.activate_final = activate_final
# Find the last Linear layer's output dimension
for layer in reversed(network.net):
if isinstance(layer, nn.Linear):
out_features = layer.out_features
break
# Mean layer
self.mean_layer = nn.Linear(out_features, action_dim)
if init_final is not None:
nn.init.uniform_(self.mean_layer.weight, -init_final, init_final)
nn.init.uniform_(self.mean_layer.bias, -init_final, init_final)
else:
orthogonal_init()(self.mean_layer.weight)
# Standard deviation layer or parameter
if fixed_std is None:
if std_parameterization == "uniform":
self.log_stds = nn.Parameter(torch.zeros(action_dim, device=self.device))
else:
self.std_layer = nn.Linear(out_features, action_dim)
if init_final is not None:
nn.init.uniform_(self.std_layer.weight, -init_final, init_final)
nn.init.uniform_(self.std_layer.bias, -init_final, init_final)
else:
orthogonal_init()(self.std_layer.weight)
self.to(self.device)
def forward(
self,
observations: torch.Tensor,
temperature: float = 1.0,
train: bool = False,
non_squash_distribution: bool = False,
) -> torch.distributions.Distribution:
self.train(train)
# Encode observations if encoder exists
if self.encoder is not None:
with torch.set_grad_enabled(train):
obs_enc = self.encoder(observations, train=train)
else:
obs_enc = observations
# Get network outputs
outputs = self.network(obs_enc)
means = self.mean_layer(outputs)
# Compute standard deviations
if self.fixed_std is None:
if self.std_parameterization == "exp":
log_stds = self.std_layer(outputs)
# Clamp log_stds to prevent too large or small values
log_stds = torch.clamp(log_stds, math.log(self.std_min), math.log(self.std_max))
stds = torch.exp(log_stds)
elif self.std_parameterization == "softplus":
stds = torch.nn.functional.softplus(self.std_layer(outputs))
stds = torch.clamp(stds, self.std_min, self.std_max)
elif self.std_parameterization == "uniform":
log_stds = torch.clamp(self.log_stds, math.log(self.std_min), math.log(self.std_max))
stds = torch.exp(log_stds).expand_as(means)
else:
raise ValueError(f"Invalid std_parameterization: {self.std_parameterization}")
else:
assert self.std_parameterization == "fixed"
stds = self.fixed_std.expand_as(means)
# Scale with temperature
temperature = torch.tensor(temperature, device=self.device)
stds = torch.clamp(stds, self.std_min, self.std_max) * torch.sqrt(temperature)
# Create distribution
if self.tanh_squash_distribution and not non_squash_distribution:
distribution = TanhMultivariateNormalDiag(
loc=means,
scale_diag=stds,
)
else:
distribution = torch.distributions.Normal(
loc=means,
scale=stds,
)
return distribution
def get_features(self, observations: torch.Tensor) -> torch.Tensor:
"""Get encoded features from observations"""
observations = observations.to(self.device)
if self.encoder is not None:
with torch.no_grad():
return self.encoder(observations, train=False)
return observations
class SACObservationEncoder(nn.Module):
"""Encode image and/or state vector observations.
TODO(ke-wang): The original work allows for (1) stacking multiple history frames and (2) using pretrained resnet encoders.
"""
def __init__(self, config: SACConfig):
"""
Creates encoders for pixel and/or state modalities.
"""
super().__init__()
self.config = config
if "observation.image" in config.input_shapes:
self.image_enc_layers = nn.Sequential(
nn.Conv2d(
config.input_shapes["observation.image"][0], config.image_encoder_hidden_dim, 7, stride=2
),
nn.ReLU(),
nn.Conv2d(config.image_encoder_hidden_dim, config.image_encoder_hidden_dim, 5, stride=2),
nn.ReLU(),
nn.Conv2d(config.image_encoder_hidden_dim, config.image_encoder_hidden_dim, 3, stride=2),
nn.ReLU(),
nn.Conv2d(config.image_encoder_hidden_dim, config.image_encoder_hidden_dim, 3, stride=2),
nn.ReLU(),
)
dummy_batch = torch.zeros(1, *config.input_shapes["observation.image"])
with torch.inference_mode():
out_shape = self.image_enc_layers(dummy_batch).shape[1:]
self.image_enc_layers.extend(
nn.Sequential(
nn.Flatten(),
nn.Linear(np.prod(out_shape), config.latent_dim),
nn.LayerNorm(config.latent_dim),
nn.Tanh(),
)
)
if "observation.state" in config.input_shapes:
self.state_enc_layers = nn.Sequential(
nn.Linear(config.input_shapes["observation.state"][0], config.state_encoder_hidden_dim),
nn.ELU(),
nn.Linear(config.state_encoder_hidden_dim, config.latent_dim),
nn.LayerNorm(config.latent_dim),
nn.Tanh(),
)
if "observation.environment_state" in config.input_shapes:
self.env_state_enc_layers = nn.Sequential(
nn.Linear(
config.input_shapes["observation.environment_state"][0], config.state_encoder_hidden_dim
),
nn.ELU(),
nn.Linear(config.state_encoder_hidden_dim, config.latent_dim),
nn.LayerNorm(config.latent_dim),
nn.Tanh(),
)
def forward(self, obs_dict: dict[str, Tensor]) -> Tensor:
"""Encode the image and/or state vector.
Each modality is encoded into a feature vector of size (latent_dim,) and then a uniform mean is taken
over all features.
"""
feat = []
# Concatenate all images along the channel dimension.
image_keys = [k for k in self.config.input_shapes if k.startswith("observation.image")]
for image_key in image_keys:
feat.append(flatten_forward_unflatten(self.image_enc_layers, obs_dict[image_key]))
if "observation.environment_state" in self.config.input_shapes:
feat.append(self.env_state_enc_layers(obs_dict["observation.environment_state"]))
if "observation.state" in self.config.input_shapes:
feat.append(self.state_enc_layers(obs_dict["observation.state"]))
return torch.stack(feat, dim=0).mean(0)
class LagrangeMultiplier(nn.Module):
def __init__(self, init_value: float = 1.0, constraint_shape: Sequence[int] = (), device: str = "cuda"):
super().__init__()
self.device = torch.device(device)
init_value = torch.log(torch.exp(torch.tensor(init_value, device=self.device)) - 1)
# Initialize the Lagrange multiplier as a parameter
self.lagrange = nn.Parameter(
torch.full(constraint_shape, init_value, dtype=torch.float32, device=self.device)
)
self.to(self.device)
def forward(self, lhs: Optional[torch.Tensor] = None, rhs: Optional[torch.Tensor] = None) -> torch.Tensor:
# Get the multiplier value based on parameterization
multiplier = torch.nn.functional.softplus(self.lagrange)
# Return the raw multiplier if no constraint values provided
if lhs is None:
return multiplier
# Move inputs to device
lhs = lhs.to(self.device)
if rhs is not None:
rhs = rhs.to(self.device)
# Use the multiplier to compute the Lagrange penalty
if rhs is None:
rhs = torch.zeros_like(lhs, device=self.device)
diff = lhs - rhs
assert diff.shape == multiplier.shape, f"Shape mismatch: {diff.shape} vs {multiplier.shape}"
return multiplier * diff
# The TanhMultivariateNormalDiag is a probability distribution that represents a transformed normal (Gaussian) distribution where:
# 1. The base distribution is a diagonal multivariate normal distribution
# 2. The samples from this normal distribution are transformed through a tanh function, which squashes the values to be between -1 and 1
# 3. Optionally, the values can be further transformed to fit within arbitrary bounds [low, high] using an affine transformation
# This type of distribution is commonly used in reinforcement learning, particularly for continuous action spaces
class TanhMultivariateNormalDiag(torch.distributions.TransformedDistribution):
DEFAULT_SAMPLE_SHAPE = torch.Size()
def __init__(
self,
loc: torch.Tensor,
scale_diag: torch.Tensor,
low: Optional[torch.Tensor] = None,
high: Optional[torch.Tensor] = None,
):
# Create base normal distribution
base_distribution = torch.distributions.Normal(loc=loc, scale=scale_diag)
# Create list of transforms
transforms = []
# Add tanh transform
transforms.append(torch.distributions.transforms.TanhTransform())
# Add rescaling transform if bounds are provided
if low is not None and high is not None:
transforms.append(
torch.distributions.transforms.AffineTransform(loc=(high + low) / 2, scale=(high - low) / 2)
)
# Initialize parent class
super().__init__(base_distribution=base_distribution, transforms=transforms)
# Store parameters
self.loc = loc
self.scale_diag = scale_diag
self.low = low
self.high = high
def mode(self) -> torch.Tensor:
"""Get the mode of the transformed distribution"""
# The mode of a normal distribution is its mean
mode = self.loc
# Apply transforms
for transform in self.transforms:
mode = transform(mode)
return mode
def rsample(self, sample_shape=DEFAULT_SAMPLE_SHAPE) -> torch.Tensor:
"""
Reparameterized sample from the distribution
"""
# Sample from base distribution
x = self.base_dist.rsample(sample_shape)
# Apply transforms
for transform in self.transforms:
x = transform(x)
return x
def log_prob(self, value: torch.Tensor) -> torch.Tensor:
"""
Compute log probability of a value
Includes the log det jacobian for the transforms
"""
# Initialize log prob
log_prob = torch.zeros_like(value[..., 0])
# Inverse transforms to get back to normal distribution
q = value
for transform in reversed(self.transforms):
q = transform.inv(q)
log_prob = log_prob - transform.log_abs_det_jacobian(q, transform(q))
# Add base distribution log prob
log_prob = log_prob + self.base_dist.log_prob(q).sum(-1)
return log_prob
def sample_and_log_prob(self, sample_shape=DEFAULT_SAMPLE_SHAPE) -> Tuple[torch.Tensor, torch.Tensor]:
"""
Sample from the distribution and compute log probability
"""
x = self.rsample(sample_shape)
log_prob = self.log_prob(x)
return x, log_prob
def entropy(self) -> torch.Tensor:
"""
Compute entropy of the distribution
"""
# Start with base distribution entropy
entropy = self.base_dist.entropy().sum(-1)
# Add log det jacobian for each transform
x = self.rsample()
for transform in self.transforms:
entropy = entropy + transform.log_abs_det_jacobian(x, transform(x))
x = transform(x)
return entropy
def create_critic_ensemble(critics: list[nn.Module], num_critics: int, device: str = "cuda") -> nn.ModuleList:
"""Creates an ensemble of critic networks"""
assert len(critics) == num_critics, f"Expected {num_critics} critics, got {len(critics)}"
return nn.ModuleList(critics).to(device)
def orthogonal_init():
return lambda x: torch.nn.init.orthogonal_(x, gain=1.0)
# borrowed from tdmpc
def flatten_forward_unflatten(fn: Callable[[Tensor], Tensor], image_tensor: Tensor) -> Tensor:
"""Helper to temporarily flatten extra dims at the start of the image tensor.
Args:
fn: Callable that the image tensor will be passed to. It should accept (B, C, H, W) and return
(B, *), where * is any number of dimensions.
image_tensor: An image tensor of shape (**, C, H, W), where ** is any number of dimensions and
can be more than 1 dimensions, generally different from *.
Returns:
A return value from the callable reshaped to (**, *).
"""
if image_tensor.ndim == 4:
return fn(image_tensor)
start_dims = image_tensor.shape[:-3]
inp = torch.flatten(image_tensor, end_dim=-4)
flat_out = fn(inp)
return torch.reshape(flat_out, (*start_dims, *flat_out.shape[1:]))