Remove old hacks and update pybullet gymnasium

CHANGELOG.md

@@ -1,4 +1,4 @@
-## Release 2.2.0a11 (WIP)
+## Release 2.2.0a12 (WIP)
 
 ### Breaking Changes
 - Removed `gym` dependency, the package is still required for some pretrained agents.
@@ -10,6 +10,8 @@
 - Added `ppo_lstm` to hyperparams_opt.py (@technocrat13)
 
 ### Bug fixes
+- Upgraded to `pybullet_envs_gymnasium>=0.4.0`
+- Removed old hacks (for instance limiting offpolicy algorithms to one env at test time)
 
 ### Documentation
 

requirements.txt

@@ -3,7 +3,7 @@ stable-baselines3[extra_no_roms,tests,docs]>=2.2.0a11,<3.0
 sb3-contrib>=2.2.0a11,<3.0
 box2d-py==2.3.8
 pybullet
-pybullet_envs_gymnasium
+pybullet_envs_gymnasium>=0.4.0
 # minigrid
 # scikit-optimize
 optuna~=3.0

rl_zoo3/enjoy.py

@@ -126,9 +126,6 @@ def enjoy() -> None:  # noqa: C901
     # Off-policy algorithm only support one env for now
     off_policy_algos = ["qrdqn", "dqn", "ddpg", "sac", "her", "td3", "tqc"]
 
-    if algo in off_policy_algos:
-        args.n_envs = 1
-
     set_random_seed(args.seed)
 
     if args.num_threads > 0:

rl_zoo3/utils.py

@@ -211,9 +211,6 @@ def create_test_env(
     :param env_kwargs: Optional keyword argument to pass to the env constructor
     :return:
     """
-    # Avoid circular import
-    from rl_zoo3.exp_manager import ExperimentManager
-
     # Create the environment and wrap it if necessary
     assert hyperparams is not None
     env_wrapper = get_wrapper_class(hyperparams)
@@ -224,12 +221,8 @@ def create_test_env(
         del hyperparams["env_wrapper"]
 
     vec_env_kwargs: Dict[str, Any] = {}
-    vec_env_cls = DummyVecEnv
-    if n_envs > 1 or (ExperimentManager.is_bullet(env_id) and should_render):
-        # HACK: force SubprocVecEnv for Bullet env
-        # as Pybullet envs does not follow gym.render() interface
-        vec_env_cls = SubprocVecEnv  # type: ignore[assignment]
-        # start_method = 'spawn' for thread safe
+    # Avoid potential shared memory issue
+    vec_env_cls = SubprocVecEnv if n_envs > 1 else DummyVecEnv
 
     # Fix for gym 0.26, to keep old behavior
     env_kwargs = env_kwargs or {}
@@ -252,7 +245,7 @@ def make_env(**kwargs) -> gym.Env:
         seed=seed,
         wrapper_class=env_wrapper,
         env_kwargs=env_kwargs,
-        vec_env_cls=vec_env_cls,
+        vec_env_cls=vec_env_cls,  # type: ignore[arg-type]
         vec_env_kwargs=vec_env_kwargs,
     )
 

rl_zoo3/version.txt

@@ -1 +1 @@
-2.2.0a11
+2.2.0a12