Skip to content

added failure if users try to determine export method for direct rl e… #7081

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
kellyguo11 merged 5 commits into from
Aug 20, 2026
Merged

added failure if users try to determine export method for direct rl e… #7081

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
f78d321
added failure if users try to determine export method for direct rl e…
frlai Aug 13, 2026
39d6e4f
Merge branch 'develop' into frlai/warn_direct_env_has_no_export_with
frlai Aug 14, 2026
54a62cf
Merge branch 'develop' into frlai/warn_direct_env_has_no_export_with
frlai Aug 18, 2026
3503b93
Merge branch 'develop' into frlai/warn_direct_env_has_no_export_with
frlai Aug 19, 2026
3f758b7
Merge branch 'develop' into frlai/warn_direct_env_has_no_export_with
frlai Aug 20, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions scripts/reinforcement_learning/leapp/export_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,9 +41,9 @@ def add_common_export_args(parser: argparse.ArgumentParser, *, agent_default: st
parser.add_argument(
"--export_method",
type=str,
default="onnx-dynamo",
choices=["onnx-dynamo", "onnx-torchscript", "jit-script", "jit-trace"],
help="Method to export the policy",
default=None,
choices=["onnx-dynamo", "onnx-torchscript", "jit-script", "jit-trace", "pt2"],

@isaaclab-review-bot isaaclab-review-bot Bot Aug 13, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🔵 Suggestion · Api — Unrelated pt2 choice added without documentation

Adding "pt2" widens the accepted values of the public --export_method flag and is forwarded verbatim to patch_env_for_export, but it is unrelated to this change's stated purpose and appears in neither the help text nor the changelog fragment. Either drop it from this change or document it in the help string and changelog alongside the supporting backend support.

help="Method to export the policy. Defaults to onnx-dynamo.",
)
parser.add_argument(
"--export_save_path",
Expand Down
8 changes: 7 additions & 1 deletion scripts/reinforcement_learning/leapp/rl_games/export.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -219,11 +219,17 @@ def export_rl_games_agent(
graph_name = args_cli.export_task_name if args_cli.export_task_name is not None else task_name

if isinstance(env.unwrapped, ManagerBasedRLEnv):
export_method = "onnx-dynamo" if args_cli.export_method is None else args_cli.export_method
patch_env_for_export(
env,
export_method=args_cli.export_method,
export_method=export_method,
required_obs_groups=_required_obs_groups(agent_cfg),
)
elif args_cli.export_method is not None:
raise ValueError(
"--export_method is only supported for manager-based environments. For direct environments, "
"set export_with directly in the annotate.output_tensors() call instead."
)

if isinstance(env.unwrapped.cfg, DirectMARLEnvCfg):
env = multi_agent_to_single_agent(env)
Expand Down
8 changes: 7 additions & 1 deletion scripts/reinforcement_learning/leapp/rsl_rl/export.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -263,6 +263,7 @@ def export_rsl_rl_agent(
graph_name = args_cli.export_task_name if args_cli.export_task_name is not None else task_name

if isinstance(env.unwrapped, ManagerBasedRLEnv):
export_method = "onnx-dynamo" if args_cli.export_method is None else args_cli.export_method
# Patch only the observation groups consumed by the actor policy.
# This filters out the critic and teacher observation groups.
obs_groups_cfg = getattr(agent_cfg, "obs_groups", None)
Expand All @@ -272,9 +273,14 @@ def export_rsl_rl_agent(
required_obs_groups = {"policy"}
patch_env_for_export(
env,
export_method=args_cli.export_method,
export_method=export_method,
required_obs_groups=required_obs_groups,
)
elif args_cli.export_method is not None:
raise ValueError(
"--export_method is only supported for manager-based environments. For direct environments, "
"set export_with directly in the annotate.output_tensors() call instead."
)

env = RslRlVecEnvWrapper(env, clip_actions=agent_cfg.clip_actions)

Expand Down
9 changes: 8 additions & 1 deletion scripts/reinforcement_learning/leapp/sb3/export.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -284,11 +284,18 @@ def export_sb3_agent(
try:
env = gym.make(args_cli.task, cfg=env_cfg, render_mode=None)
if not isinstance(env.unwrapped, ManagerBasedRLEnv):
if args_cli.export_method is not None:

@isaaclab-review-bot isaaclab-review-bot Bot Aug 13, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🟡 Warning · Implementation — SB3 direct-env error masks unsupported-environment message

SB3 LEAPP export does not support direct environments at all (the following NotImplementedError). Checking --export_method first replaces that accurate message with advice to "set export_with directly in the annotate.output_tensors() call", which cannot be followed here and implies direct export works for SB3. Raise the NotImplementedError first, or drop this extra check in the SB3 backend.

raise ValueError(
"--export_method is only supported for manager-based environments. For direct environments, "
"set export_with directly in the annotate.output_tensors() call instead."
)
raise NotImplementedError("SB3 LEAPP export currently supports manager-based environments only.")

export_method = "onnx-dynamo" if args_cli.export_method is None else args_cli.export_method

policy_node_name = ensure_env_spec_id(env)
graph_name = args_cli.export_task_name if args_cli.export_task_name is not None else task_name
patch_env_for_export(env, export_method=args_cli.export_method, required_obs_groups={"policy"})
patch_env_for_export(env, export_method=export_method, required_obs_groups={"policy"})

print(f"[INFO] Loading model checkpoint from: {checkpoint_path}")
agent = _load_agent(checkpoint_path, device=env.unwrapped.device)
Expand Down
8 changes: 7 additions & 1 deletion scripts/reinforcement_learning/leapp/skrl/export.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -225,7 +225,13 @@ def export_skrl_agent(
graph_name = args_cli.export_task_name if args_cli.export_task_name is not None else task_name

if isinstance(env.unwrapped, ManagerBasedRLEnv):
patch_env_for_export(env, export_method=args_cli.export_method, required_obs_groups={"policy"})
export_method = "onnx-dynamo" if args_cli.export_method is None else args_cli.export_method
patch_env_for_export(env, export_method=export_method, required_obs_groups={"policy"})
elif args_cli.export_method is not None:
raise ValueError(
"--export_method is only supported for manager-based environments. For direct environments, "
"set export_with directly in the annotate.output_tensors() call instead."
)

if isinstance(env.unwrapped.cfg, DirectMARLEnvCfg) and algorithm in ["ppo"]:
env = multi_agent_to_single_agent(env)
Expand Down
4 changes: 4 additions & 0 deletions source/isaaclab/changelog.d/frlai-warn-direct-env-export-method.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,4 @@
Fixed
^^^^^

* Raised an error when ``--export_method`` is used to export a direct RL environment with LEAPP.
Loading