Skip to content

onnx export: properly handle obs_keys #231

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
edbeeching merged 4 commits into from
May 6, 2025
Merged

onnx export: properly handle obs_keys #231

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
aec2f22
onnx export: properly handle obs_key
Nijco Apr 7, 2025
8be5175
Merge branch 'edbeeching:main' into main
Nijco Apr 22, 2025
c5a41bb
fix: onnx export verification respects any obs keys
Nijco Apr 22, 2025
7ae1aaa
return support for onnx verification in case of 'use_obs_array'
Nijco May 4, 2025
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
13 changes: 7 additions & 6 deletions godot_rl/wrappers/onnx/stable_baselines_export.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -81,10 +81,10 @@ def export_model_as_onnx(model, onnx_model_path: str, use_obs_array: bool = Fals
args=(dummy_input, torch.zeros(1).float()),
f=onnx_model_path,
opset_version=17,
input_names=["obs", "state_ins"],
input_names=obs_keys + ["state_ins"],
output_names=["output", "state_outs"],
dynamic_axes={
"obs": {0: "batch_size"},
**{sub_obs: {0: "batch_size"} for sub_obs in obs_keys},
"state_ins": {0: "batch_size"}, # variable length axes
"output": {0: "batch_size"},
"state_outs": {0: "batch_size"},
Expand Down Expand Up @@ -119,16 +119,17 @@ def verify_onnx_export(ppo: PPO, onnx_model_path: str, num_tests=10, use_obs_arr
if use_obs_array:
obs = np.expand_dims(ppo.observation_space.sample(), axis=0)
obs2 = torch.tensor(obs)
obs = {"obs": obs}
else:
obs = dict(ppo.observation_space.sample())
obs_dict = dict(ppo.observation_space.sample())
obs2 = {}
for k, v in obs.items():
for k, v in obs_dict.items():
obs2[k] = torch.from_numpy(v).unsqueeze(0)
obs = [v for v in obs.values()]
obs = {k: [v] for k, v in obs_dict.items()}

with torch.no_grad():
action_sb3, _, _ = sb3_model(obs2, deterministic=True)

action_onnx, state_outs = ort_sess.run(None, {"obs": obs, "state_ins": np.array([0.0], dtype=np.float32)})
action_onnx, state_outs = ort_sess.run(None, {**obs, "state_ins": np.array([0.0], dtype=np.float32)})
assert np.allclose(action_sb3, action_onnx, atol=1e-5), "Mismatch in action output"
assert np.allclose(state_outs, np.array([0.0]), atol=1e-5), "Mismatch in state_outs output"