Fixes RSL-RL ONNX exporter for empirical normalization (#78)

The current onnx exporter does not export the empirical normalization
layer. This MR adds the empirical normalization exporting to the JIT
and ONNX exporters for RSL-RL.

- Bug fix (non-breaking change which fixes an issue)

- [x] I have run the [`pre-commit` checks](https://pre-commit.com/) with
`./orbit.sh --format`
- [x] My changes generate no new warnings
- [ ] I have added tests that prove my fix is effective or that my
feature works
- [x] I have run all the tests with `./orbit.sh --test` and they pass
(some did timeout)
- [x] I have updated the changelog and the corresponding version in the
extension's `config/extension.toml` file
- [ ] I have added my name to the `CONTRIBUTORS.md` or my name already
exists there

---------

Co-authored-by: Mayank Mittal <[email protected]>

Author: Nemantor

source/extensions/omni.isaac.orbit_tasks/config/extension.toml

@@ -1,7 +1,7 @@
 [package]
 
 # Note: Semantic Versioning is used: https://semver.org/
-version = "0.6.1"
+version = "0.6.2"
 
 # Description
 title = "ORBIT Environments"

source/extensions/omni.isaac.orbit_tasks/docs/CHANGELOG.rst

@@ -1,6 +1,18 @@
 Changelog
 ---------
 
+0.6.2 (2024-05-31)
+~~~~~~~~~~~~~~~~~~
+
+Added
+^^^^^
+
+* Added exporting of empirical normalization layer to ONNX and JIT when exporting the model using
+  :meth:`omni.isaac.orbit.actuators.ActuatorNetMLP.export` method. Previously, the normalization layer
+  was not exported to the ONNX and JIT models. This caused the exported model to not work properly
+  when used for inference.
+
+
 0.6.1 (2024-04-16)
 ~~~~~~~~~~~~~~~~~~
 

source/extensions/omni.isaac.orbit_tasks/omni/isaac/orbit_tasks/utils/wrappers/rsl_rl/exporter.py

@@ -8,33 +8,37 @@
 import torch
 
 
-def export_policy_as_jit(actor_critic: object, path: str, filename="policy.pt"):
+def export_policy_as_jit(actor_critic: object, normalizer: object | None, path: str, filename="policy.pt"):
     """Export policy into a Torch JIT file.
 
     Args:
         actor_critic: The actor-critic torch module.
+        normalizer: The empirical normalizer module. If None, Identity is used.
         path: The path to the saving directory.
         filename: The name of exported JIT file. Defaults to "policy.pt".
 
     Reference:
         https://github.com/leggedrobotics/legged_gym/blob/master/legged_gym/utils/helpers.py#L180
     """
-    policy_exporter = _TorchPolicyExporter(actor_critic)
+    policy_exporter = _TorchPolicyExporter(actor_critic, normalizer)
     policy_exporter.export(path, filename)
 
 
-def export_policy_as_onnx(actor_critic: object, path: str, filename="policy.onnx", verbose=False):
+def export_policy_as_onnx(
+    actor_critic: object, normalizer: object | None, path: str, filename="policy.onnx", verbose=False
+):
     """Export policy into a Torch ONNX file.
 
     Args:
         actor_critic: The actor-critic torch module.
+        normalizer: The empirical normalizer module. If None, Identity is used.
         path: The path to the saving directory.
-        filename: The name of exported JIT file. Defaults to "policy.pt".
+        filename: The name of exported ONNX file. Defaults to "policy.onnx".
         verbose: Whether to print the model summary. Defaults to False.
     """
     if not os.path.exists(path):
         os.makedirs(path, exist_ok=True)
-    policy_exporter = _OnnxPolicyExporter(actor_critic, verbose)
+    policy_exporter = _OnnxPolicyExporter(actor_critic, normalizer, verbose)
     policy_exporter.export(path, filename)
 
 
@@ -50,7 +54,7 @@ class _TorchPolicyExporter(torch.nn.Module):
         https://github.com/leggedrobotics/legged_gym/blob/master/legged_gym/utils/helpers.py#L193
     """
 
-    def __init__(self, actor_critic):
+    def __init__(self, actor_critic, normalizer=None):
         super().__init__()
         self.actor = copy.deepcopy(actor_critic.actor)
         self.is_recurrent = actor_critic.is_recurrent
@@ -61,16 +65,22 @@ def __init__(self, actor_critic):
             self.register_buffer("cell_state", torch.zeros(self.rnn.num_layers, 1, self.rnn.hidden_size))
             self.forward = self.forward_lstm
             self.reset = self.reset_memory
+        # copy normalizer if exists
+        if normalizer:
+            self.normalizer = copy.deepcopy(normalizer)
+        else:
+            self.normalizer = torch.nn.Identity()
 
     def forward_lstm(self, x):
+        x = self.normalizer(x)
         x, (h, c) = self.rnn(x.unsqueeze(0), (self.hidden_state, self.cell_state))
         self.hidden_state[:] = h
         self.cell_state[:] = c
         x = x.squeeze(0)
         return self.actor(x)
 
     def forward(self, x):
-        return self.actor(x)
+        return self.actor(self.normalizer(x))
 
     @torch.jit.export
     def reset(self):
@@ -91,7 +101,7 @@ def export(self, path, filename):
 class _OnnxPolicyExporter(torch.nn.Module):
     """Exporter of actor-critic into ONNX file."""
 
-    def __init__(self, actor_critic, verbose=False):
+    def __init__(self, actor_critic, normalizer=None, verbose=False):
         super().__init__()
         self.verbose = verbose
         self.actor = copy.deepcopy(actor_critic.actor)
@@ -100,14 +110,20 @@ def __init__(self, actor_critic, verbose=False):
             self.rnn = copy.deepcopy(actor_critic.memory_a.rnn)
             self.rnn.cpu()
             self.forward = self.forward_lstm
+        # copy normalizer if exists
+        if normalizer:
+            self.normalizer = copy.deepcopy(normalizer)
+        else:
+            self.normalizer = torch.nn.Identity()
 
     def forward_lstm(self, x_in, h_in, c_in):
+        x_in = self.normalizer(x_in)
         x, (h, c) = self.rnn(x_in.unsqueeze(0), (h_in, c_in))
         x = x.squeeze(0)
         return self.actor(x), h, c
 
     def forward(self, x):
-        return self.actor(x)
+        return self.actor(self.normalizer(x))
 
     def export(self, path, filename):
         self.to("cpu")

source/standalone/workflows/rsl_rl/play.py

@@ -46,6 +46,7 @@
 from omni.isaac.orbit_tasks.utils.wrappers.rsl_rl import (
     RslRlOnPolicyRunnerCfg,
     RslRlVecEnvWrapper,
+    export_policy_as_jit,
     export_policy_as_onnx,
 )
 
@@ -78,9 +79,14 @@ def main():
     # obtain the trained policy for inference
     policy = ppo_runner.get_inference_policy(device=env.unwrapped.device)
 
-    # export policy to onnx
+    # export policy to onnx/jit
     export_model_dir = os.path.join(os.path.dirname(resume_path), "exported")
-    export_policy_as_onnx(ppo_runner.alg.actor_critic, export_model_dir, filename="policy.onnx")
+    export_policy_as_jit(
+        ppo_runner.alg.actor_critic, ppo_runner.obs_normalizer, path=export_model_dir, filename="policy.pt"
+    )
+    export_policy_as_onnx(
+        ppo_runner.alg.actor_critic, ppo_runner.obs_normalizer, path=export_model_dir, filename="policy.onnx"
+    )
 
     # reset environment
     obs, _ = env.get_observations()