Patch obs space and update trained agents

DLR-RM · Nov 5, 2024 · e55201a · e55201a
1 parent 2f7ac9a
commit e55201a
Show file tree

Hide file tree

Showing 4 changed files with 6 additions and 2 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -10,6 +10,7 @@
 ### New Features
 - Added `CrossQ` hyperparameters for SB3-contrib (@danielpalen)
 - Added Gymnasium v1.0 support
+- `--custom-objects` in `enjoy.py` now also patches obs space (when bounds are changed) to solve "Observation spaces do not match" errors
 
 ### Bug fixes
 - Replaced deprecated `huggingface_hub.Repository` when pushing to Hugging Face Hub by the recommended `HfApi` (see https://huggingface.co/docs/huggingface_hub/concepts/git_vs_http) (@cochaviz)

diff --git a/requirements.txt b/requirements.txt
@@ -1,7 +1,7 @@
 gym==0.26.2
 stable-baselines3[extra,tests,docs]>=2.4.0a11,<3.0
 box2d-py==2.3.8
-pybullet_envs_gymnasium>=0.4.0
+pybullet_envs_gymnasium>=0.5.0
 # minigrid
 cloudpickle>=2.2.1
 # optuna plots:

diff --git a/rl-trained-agents b/rl-trained-agents
diff --git a/rl_zoo3/enjoy.py b/rl_zoo3/enjoy.py
@@ -184,12 +184,15 @@ def enjoy() -> None:  # noqa: C901
             "learning_rate": 0.0,
             "lr_schedule": lambda _: 0.0,
             "clip_range": lambda _: 0.0,
+            "observation_space": env.observation_space, # load models with different obs bounds
         }
 
     if "HerReplayBuffer" in hyperparams.get("replay_buffer_class", ""):
         kwargs["env"] = env
 
     model = ALGOS[algo].load(model_path, custom_objects=custom_objects, device=args.device, **kwargs)
+    # Uncomment to save patched file (for instance gym -> gymnasium)
+    # model.save(model_path)
     obs = env.reset()
 
     # Deterministic by default except for atari games