# Copyright (c) 2022-2026, The Isaac Lab Project Developers (https://github.com/isaac-sim/IsaacLab/blob/main/CONTRIBUTORS.md).
# All rights reserved.
#
# SPDX-License-Identifier: BSD-3-Clause
"""Sub-module with utilities for parsing and loading configurations."""
from __future__ import annotations
import collections
import importlib
import inspect
import os
import re
import warnings
from typing import TYPE_CHECKING
import gymnasium as gym
import yaml
from isaaclab_tasks.utils.hydra import _user_stacklevel, resolve_presets
if TYPE_CHECKING:
from isaaclab.envs import DirectRLEnvCfg, ManagerBasedRLEnvCfg
[docs]
def load_cfg_from_registry(task_name: str, entry_point_key: str) -> dict | object:
"""Load default configuration given its entry point from the gym registry.
This function loads the configuration object from the gym registry for the given task name.
It supports both YAML and Python configuration files.
It expects the configuration to be registered in the gym registry as:
.. code-block:: python
gym.register(
id="My-Awesome-Task-v0",
...
kwargs={"env_entry_point_cfg": "path.to.config:ConfigClass"},
)
The parsed configuration object for above example can be obtained as:
.. code-block:: python
from isaaclab_tasks.utils.parse_cfg import load_cfg_from_registry
cfg = load_cfg_from_registry("My-Awesome-Task-v0", "env_entry_point_cfg")
Args:
task_name: The name of the environment.
entry_point_key: The entry point key to resolve the configuration file.
Returns:
The parsed configuration object. If the entry point is a YAML file, it is parsed into a dictionary.
If the entry point is a Python class, it is instantiated and returned.
Raises:
ValueError: If the entry point key is not available in the gym registry for the task.
"""
spec = gym.spec(task_name.split(":")[-1])
# Emit a FutureWarning when loading the env cfg for a retired task
# registered as a deprecation alias. The migration metadata lives in the
# gym.register kwargs under the ``deprecated`` key as a dict whose
# ``alias`` field is the equivalent CLI command, e.g.
# ``"deprecated": {"alias": "--task=Isaac-Cartpole-Camera-Direct presets=rgb"}``.
# FutureWarning (vs DeprecationWarning) matches the existing convention
# for end-user-facing IsaacLab deprecations (cfg fields, preset-name
# aliases) and is shown by Python's default filter regardless of which
# frame the warning is attributed to.
if entry_point_key == "env_cfg_entry_point":
deprecation = spec.kwargs.get("deprecated") or {}
new_command = deprecation.get("alias")
if new_command:
warnings.warn(
f"Task '{spec.id}' is deprecated and will be removed in a future release. Use '{new_command}'.",
FutureWarning,
stacklevel=_user_stacklevel(),
)
# obtain the configuration entry point
cfg_entry_point = spec.kwargs.get(entry_point_key)
# check if entry point exists
if cfg_entry_point is None:
# get existing agents and algorithms
agents = collections.defaultdict(list)
for k in spec.kwargs:
if k.endswith("_cfg_entry_point") and k != "env_cfg_entry_point":
spec = (
k.replace("_cfg_entry_point", "")
.replace("rl_games", "rl-games")
.replace("rsl_rl", "rsl-rl")
.split("_")
)
agent = spec[0].replace("-", "_")
algorithms = [item.upper() for item in (spec[1:] if len(spec) > 1 else ["PPO"])]
agents[agent].extend(algorithms)
msg = "\nExisting RL library (and algorithms) config entry points: "
for agent, algorithms in agents.items():
msg += f"\n |-- {agent}: {', '.join(algorithms)}"
# raise error
raise ValueError(
f"Could not find configuration for the environment: '{task_name}'."
f"\nPlease check that the gym registry has the entry point: '{entry_point_key}'."
f"{msg if agents else ''}"
)
# parse the default config file
if isinstance(cfg_entry_point, str) and cfg_entry_point.endswith(".yaml"):
if os.path.exists(cfg_entry_point):
# absolute path for the config file
config_file = cfg_entry_point
else:
# resolve path to the module location
mod_name, file_name = cfg_entry_point.split(":")
mod_path = os.path.dirname(importlib.import_module(mod_name).__file__)
# obtain the configuration file path
config_file = os.path.join(mod_path, file_name)
# load the configuration
print(f"[INFO]: Parsing configuration from: {config_file}")
with open(config_file, encoding="utf-8") as f:
cfg = yaml.full_load(f)
else:
if callable(cfg_entry_point):
# resolve path to the module location
mod_path = inspect.getfile(cfg_entry_point)
# load the configuration
cfg_cls = cfg_entry_point()
elif isinstance(cfg_entry_point, str):
# resolve path to the module location
mod_name, attr_name = cfg_entry_point.split(":")
mod = importlib.import_module(mod_name)
cfg_cls = getattr(mod, attr_name)
else:
cfg_cls = cfg_entry_point
# load the configuration
print(f"[INFO]: Parsing configuration from: {cfg_entry_point}")
if callable(cfg_cls):
cfg = cfg_cls()
else:
cfg = cfg_cls
return cfg
[docs]
def parse_env_cfg(
task_name: str, device: str = "cuda:0", num_envs: int | None = None, use_fabric: bool | None = None
) -> ManagerBasedRLEnvCfg | DirectRLEnvCfg:
"""Parse configuration for an environment and override based on inputs.
Args:
task_name: The name of the environment.
device: The device to run the simulation on. Defaults to "cuda:0".
num_envs: Number of environments to create. Defaults to None, in which case it is left unchanged.
use_fabric: Whether to enable/disable fabric interface. If false, all read/write operations go through USD.
This slows down the simulation but allows seeing the changes in the USD through the USD stage.
Defaults to None, in which case it is left unchanged.
Returns:
The parsed configuration object.
Raises:
RuntimeError: If the configuration for the task is not a class. We assume users always use a class for the
environment configuration.
"""
# load the default configuration
cfg = load_cfg_from_registry(task_name.split(":")[-1], "env_cfg_entry_point")
# check that it is not a dict
# we assume users always use a class for the configuration
if isinstance(cfg, dict):
raise RuntimeError(f"Configuration for the task: '{task_name}' is not a class. Please provide a class.")
# Resolve any PresetCfg wrappers to their default preset so the config
# is usable without a Hydra CLI override (e.g. in tests).
# Must happen BEFORE attribute overrides, otherwise overrides on PresetCfg wrapper
# fields (e.g. cfg.scene when scene is a PresetCfg) get discarded when the wrapper
# is replaced by its .default.
cfg = resolve_presets(cfg)
# simulation device
cfg.sim.device = device
# disable fabric to read/write through USD
if use_fabric is not None:
cfg.sim.use_fabric = use_fabric
# number of environments
if num_envs is not None:
cfg.scene.num_envs = num_envs
return cfg
[docs]
def get_checkpoint_path(
log_path: str,
run_dir: str = ".*",
checkpoint: str = ".*",
other_dirs: list[str] = None,
sort_alpha: bool = True,
preferred_checkpoint: str | None = None,
) -> str:
"""Get path to the model checkpoint in input directory.
The checkpoint file is resolved as: ``<log_path>/<run_dir>/<*other_dirs>/<checkpoint>``, where the
:attr:`other_dirs` are intermediate folder names to concatenate. These cannot be regex expressions.
If :attr:`run_dir` and :attr:`checkpoint` are regex expressions then the most recent (highest alphabetical order)
run and checkpoint are selected. To disable this behavior, set the flag :attr:`sort_alpha` to False.
Args:
log_path: The log directory path to find models in.
run_dir: The regex expression for the name of the directory containing the run. Defaults to the most
recent directory created inside :attr:`log_path`.
other_dirs: The intermediate directories between the run directory and the checkpoint file. Defaults to
None, which implies that checkpoint file is directly under the run directory.
checkpoint: The regex expression for the model checkpoint file. Defaults to ``".*"``, which matches all
checkpoints and selects the most recent one.
sort_alpha: Whether to sort the runs by alphabetical order. Defaults to True.
If False, the folders in :attr:`run_dir` are sorted by the last modified time.
preferred_checkpoint: An optional regex expression that is matched before :attr:`checkpoint`. When it is
provided and matches at least one file, that match is used; otherwise resolution falls back to
:attr:`checkpoint`. This allows preferring a specific checkpoint (e.g. the best or final model) while
still resolving the latest available checkpoint when the preferred file has not been written yet (e.g.
for short runs). Defaults to None, which matches :attr:`checkpoint` directly.
Returns:
The path to the model checkpoint.
Raises:
ValueError: When no runs are found in the input directory.
ValueError: When no checkpoints are found in the input directory.
"""
# check if runs present in directory
try:
# find all runs in the directory that math the regex expression
runs = [
os.path.join(log_path, run) for run in os.scandir(log_path) if run.is_dir() and re.match(run_dir, run.name)
]
# sort matched runs by alphabetical order (latest run should be last)
if sort_alpha:
runs.sort()
else:
runs = sorted(runs, key=os.path.getmtime)
# create last run file path
if other_dirs is not None:
run_path = os.path.join(runs[-1], *other_dirs)
else:
run_path = runs[-1]
except IndexError:
raise ValueError(f"No runs present in the directory: '{log_path}' match: '{run_dir}'.")
# prefer ``preferred_checkpoint`` when given and it matches; otherwise fall back to the general
# ``checkpoint`` pattern (e.g. resolve the latest numbered checkpoint when the preferred best/final
# checkpoint file has not been written yet)
model_checkpoints = []
if preferred_checkpoint is not None:
model_checkpoints = [f for f in os.listdir(run_path) if re.match(preferred_checkpoint, f)]
if len(model_checkpoints) == 0:
model_checkpoints = [f for f in os.listdir(run_path) if re.match(checkpoint, f)]
# check if any checkpoints are present
if len(model_checkpoints) == 0:
patterns = f"'{checkpoint}'"
if preferred_checkpoint is not None:
patterns = f"'{preferred_checkpoint}' nor '{checkpoint}'"
raise ValueError(f"No checkpoints in the directory: '{run_path}' match {patterns}.")
# sort naturally so numbered checkpoints such as *_10 come after *_9 even with long filename prefixes
model_checkpoints.sort(key=lambda m: [int(token) if token.isdigit() else token for token in re.split(r"(\d+)", m)])
# get latest matched checkpoint file
checkpoint_file = model_checkpoints[-1]
return os.path.join(run_path, checkpoint_file)
