Fixes memory leak in video recorder

scripts/benchmarks/benchmark_non_rl.py

@@ -81,13 +81,18 @@
 import os
 import torch
 from datetime import datetime
+from gymnasium.wrappers import RecordVideo
 
 from isaaclab.envs import DirectMARLEnvCfg, DirectRLEnvCfg, ManagerBasedRLEnvCfg
 from isaaclab.utils.dict import print_dict
+from isaaclab.utils.recorder import stop_recording
 
 import isaaclab_tasks  # noqa: F401
 from isaaclab_tasks.utils.hydra import hydra_task_config
 
+# monkey-patch the stop_recording function to fix memory leak error (see https://github.com/Farama-Foundation/Gymnasium/pull/1444)
+RecordVideo.stop_recording = stop_recording
+
 imports_time_end = time.perf_counter_ns()
 
 
@@ -138,7 +143,7 @@ def main(env_cfg: ManagerBasedRLEnvCfg | DirectRLEnvCfg | DirectMARLEnvCfg, agen
         }
         print("[INFO] Recording videos during training.")
         print_dict(video_kwargs, nesting=4)
-        env = gym.wrappers.RecordVideo(env, **video_kwargs)
+        env = RecordVideo(env, **video_kwargs)
 
     task_startup_time_end = time.perf_counter_ns()
 

scripts/benchmarks/benchmark_rlgames.py

@@ -68,6 +68,7 @@
 import random
 import torch
 from datetime import datetime
+from gymnasium.wrappers import RecordVideo
 
 from rl_games.common import env_configurations, vecenv
 from rl_games.common.algo_observer import IsaacAlgoObserver
@@ -76,6 +77,7 @@
 from isaaclab.envs import DirectMARLEnvCfg, DirectRLEnvCfg, ManagerBasedRLEnvCfg
 from isaaclab.utils.dict import print_dict
 from isaaclab.utils.io import dump_pickle, dump_yaml
+from isaaclab.utils.recorder import stop_recording
 
 from isaaclab_rl.rl_games import RlGamesGpuEnv, RlGamesVecEnvWrapper
 
@@ -86,6 +88,7 @@
 
 sys.path.append(os.path.join(os.path.dirname(os.path.abspath(__file__)), "../.."))
 
+
 from isaaclab.utils.timer import Timer
 from scripts.benchmarks.utils import (
     log_app_start_time,
@@ -100,6 +103,9 @@
     parse_tf_logs,
 )
 
+# monkey-patch the stop_recording function to fix memory leak error (see https://github.com/Farama-Foundation/Gymnasium/pull/1444)
+RecordVideo.stop_recording = stop_recording
+
 torch.backends.cuda.matmul.allow_tf32 = True
 torch.backends.cudnn.allow_tf32 = True
 torch.backends.cudnn.deterministic = False
@@ -190,7 +196,7 @@ def main(env_cfg: ManagerBasedRLEnvCfg | DirectRLEnvCfg | DirectMARLEnvCfg, agen
         }
         print("[INFO] Recording videos during training.")
         print_dict(video_kwargs, nesting=4)
-        env = gym.wrappers.RecordVideo(env, **video_kwargs)
+        env = RecordVideo(env, **video_kwargs)
 
     # wrap around environment for rl-games
     env = RlGamesVecEnvWrapper(env, rl_device, clip_obs, clip_actions)

scripts/benchmarks/benchmark_rsl_rl.py

@@ -70,12 +70,14 @@
 import numpy as np
 import torch
 from datetime import datetime
+from gymnasium.wrappers import RecordVideo
 
 from rsl_rl.runners import OnPolicyRunner
 
 from isaaclab.envs import DirectMARLEnvCfg, DirectRLEnvCfg, ManagerBasedRLEnvCfg
 from isaaclab.utils.dict import print_dict
 from isaaclab.utils.io import dump_pickle, dump_yaml
+from isaaclab.utils.recorder import stop_recording
 
 from isaaclab_rl.rsl_rl import RslRlOnPolicyRunnerCfg, RslRlVecEnvWrapper
 
@@ -88,6 +90,7 @@
 from isaacsim.core.utils.extensions import enable_extension
 
 enable_extension("isaacsim.benchmark.services")
+
 from isaacsim.benchmark.services import BaseIsaacBenchmark
 
 from isaaclab.utils.timer import Timer
@@ -104,6 +107,9 @@
     parse_tf_logs,
 )
 
+# monkey-patch the stop_recording function to fix memory leak error (see https://github.com/Farama-Foundation/Gymnasium/pull/1444)
+RecordVideo.stop_recording = stop_recording
+
 torch.backends.cuda.matmul.allow_tf32 = True
 torch.backends.cudnn.allow_tf32 = True
 torch.backends.cudnn.deterministic = False
@@ -183,7 +189,7 @@ def main(env_cfg: ManagerBasedRLEnvCfg | DirectRLEnvCfg | DirectMARLEnvCfg, agen
         }
         print("[INFO] Recording videos during training.")
         print_dict(video_kwargs, nesting=4)
-        env = gym.wrappers.RecordVideo(env, **video_kwargs)
+        env = RecordVideo(env, **video_kwargs)
     # wrap around environment for rsl-rl
     env = RslRlVecEnvWrapper(env)
 

scripts/reinforcement_learning/rl_games/play.py

@@ -60,6 +60,7 @@
 import random
 import time
 import torch
+from gymnasium.wrappers import RecordVideo
 
 from rl_games.common import env_configurations, vecenv
 from rl_games.common.player import BasePlayer
@@ -75,13 +76,17 @@
 from isaaclab.utils.assets import retrieve_file_path
 from isaaclab.utils.dict import print_dict
 from isaaclab.utils.pretrained_checkpoint import get_published_pretrained_checkpoint
+from isaaclab.utils.recorder import stop_recording
 
 from isaaclab_rl.rl_games import RlGamesGpuEnv, RlGamesVecEnvWrapper
 
 import isaaclab_tasks  # noqa: F401
 from isaaclab_tasks.utils import get_checkpoint_path
 from isaaclab_tasks.utils.hydra import hydra_task_config
 
+# monkey-patch the stop_recording function to fix memory leak error (see https://github.com/Farama-Foundation/Gymnasium/pull/1444)
+RecordVideo.stop_recording = stop_recording
+
 # PLACEHOLDER: Extension template (do not remove this comment)
 
 
@@ -154,7 +159,7 @@ def main(env_cfg: ManagerBasedRLEnvCfg | DirectRLEnvCfg | DirectMARLEnvCfg, agen
         }
         print("[INFO] Recording videos during training.")
         print_dict(video_kwargs, nesting=4)
-        env = gym.wrappers.RecordVideo(env, **video_kwargs)
+        env = RecordVideo(env, **video_kwargs)
 
     # wrap around environment for rl-games
     env = RlGamesVecEnvWrapper(env, rl_device, clip_obs, clip_actions, obs_groups, concate_obs_groups)

scripts/reinforcement_learning/rl_games/train.py

@@ -64,6 +64,7 @@
 import os
 import random
 from datetime import datetime
+from gymnasium.wrappers import RecordVideo
 
 import omni
 from rl_games.common import env_configurations, vecenv
@@ -80,12 +81,16 @@
 from isaaclab.utils.assets import retrieve_file_path
 from isaaclab.utils.dict import print_dict
 from isaaclab.utils.io import dump_pickle, dump_yaml
+from isaaclab.utils.recorder import stop_recording
 
 from isaaclab_rl.rl_games import RlGamesGpuEnv, RlGamesVecEnvWrapper
 
 import isaaclab_tasks  # noqa: F401
 from isaaclab_tasks.utils.hydra import hydra_task_config
 
+# monkey-patch the stop_recording function to fix memory leak error (see https://github.com/Farama-Foundation/Gymnasium/pull/1444)
+RecordVideo.stop_recording = stop_recording
+
 # PLACEHOLDER: Extension template (do not remove this comment)
 
 
@@ -177,7 +182,7 @@ def main(env_cfg: ManagerBasedRLEnvCfg | DirectRLEnvCfg | DirectMARLEnvCfg, agen
         }
         print("[INFO] Recording videos during training.")
         print_dict(video_kwargs, nesting=4)
-        env = gym.wrappers.RecordVideo(env, **video_kwargs)
+        env = RecordVideo(env, **video_kwargs)
 
     # wrap around environment for rl-games
     env = RlGamesVecEnvWrapper(env, rl_device, clip_obs, clip_actions, obs_groups, concate_obs_groups)

scripts/reinforcement_learning/rsl_rl/play.py

@@ -57,6 +57,7 @@
 import os
 import time
 import torch
+from gymnasium.wrappers import RecordVideo
 
 from rsl_rl.runners import DistillationRunner, OnPolicyRunner
 
@@ -70,13 +71,17 @@
 from isaaclab.utils.assets import retrieve_file_path
 from isaaclab.utils.dict import print_dict
 from isaaclab.utils.pretrained_checkpoint import get_published_pretrained_checkpoint
+from isaaclab.utils.recorder import stop_recording
 
 from isaaclab_rl.rsl_rl import RslRlBaseRunnerCfg, RslRlVecEnvWrapper, export_policy_as_jit, export_policy_as_onnx
 
 import isaaclab_tasks  # noqa: F401
 from isaaclab_tasks.utils import get_checkpoint_path
 from isaaclab_tasks.utils.hydra import hydra_task_config
 
+# monkey-patch the stop_recording function to fix memory leak error (see https://github.com/Farama-Foundation/Gymnasium/pull/1444)
+RecordVideo.stop_recording = stop_recording
+
 # PLACEHOLDER: Extension template (do not remove this comment)
 
 
@@ -129,7 +134,7 @@ def main(env_cfg: ManagerBasedRLEnvCfg | DirectRLEnvCfg | DirectMARLEnvCfg, agen
         }
         print("[INFO] Recording videos during training.")
         print_dict(video_kwargs, nesting=4)
-        env = gym.wrappers.RecordVideo(env, **video_kwargs)
+        env = RecordVideo(env, **video_kwargs)
 
     # wrap around environment for rsl-rl
     env = RslRlVecEnvWrapper(env, clip_actions=agent_cfg.clip_actions)

scripts/reinforcement_learning/rsl_rl/train.py

@@ -76,6 +76,7 @@
 import os
 import torch
 from datetime import datetime
+from gymnasium.wrappers import RecordVideo
 
 import omni
 from rsl_rl.runners import DistillationRunner, OnPolicyRunner
@@ -89,13 +90,17 @@
 )
 from isaaclab.utils.dict import print_dict
 from isaaclab.utils.io import dump_pickle, dump_yaml
+from isaaclab.utils.recorder import stop_recording
 
 from isaaclab_rl.rsl_rl import RslRlBaseRunnerCfg, RslRlVecEnvWrapper
 
 import isaaclab_tasks  # noqa: F401
 from isaaclab_tasks.utils import get_checkpoint_path
 from isaaclab_tasks.utils.hydra import hydra_task_config
 
+# monkey-patch the stop_recording function to fix memory leak error (see https://github.com/Farama-Foundation/Gymnasium/pull/1444)
+RecordVideo.stop_recording = stop_recording
+
 # PLACEHOLDER: Extension template (do not remove this comment)
 
 torch.backends.cuda.matmul.allow_tf32 = True
@@ -171,7 +176,7 @@ def main(env_cfg: ManagerBasedRLEnvCfg | DirectRLEnvCfg | DirectMARLEnvCfg, agen
         }
         print("[INFO] Recording videos during training.")
         print_dict(video_kwargs, nesting=4)
-        env = gym.wrappers.RecordVideo(env, **video_kwargs)
+        env = RecordVideo(env, **video_kwargs)
 
     # wrap around environment for rsl-rl
     env = RslRlVecEnvWrapper(env, clip_actions=agent_cfg.clip_actions)

scripts/reinforcement_learning/sb3/play.py

@@ -66,6 +66,7 @@
 import random
 import time
 import torch
+from gymnasium.wrappers import RecordVideo
 
 from stable_baselines3 import PPO
 from stable_baselines3.common.vec_env import VecNormalize
@@ -79,13 +80,17 @@
 )
 from isaaclab.utils.dict import print_dict
 from isaaclab.utils.pretrained_checkpoint import get_published_pretrained_checkpoint
+from isaaclab.utils.recorder import stop_recording
 
 from isaaclab_rl.sb3 import Sb3VecEnvWrapper, process_sb3_cfg
 
 import isaaclab_tasks  # noqa: F401
 from isaaclab_tasks.utils.hydra import hydra_task_config
 from isaaclab_tasks.utils.parse_cfg import get_checkpoint_path
 
+# monkey-patch the stop_recording function to fix memory leak error (see https://github.com/Farama-Foundation/Gymnasium/pull/1444)
+RecordVideo.stop_recording = stop_recording
+
 # PLACEHOLDER: Extension template (do not remove this comment)
 
 
@@ -147,7 +152,7 @@ def main(env_cfg: ManagerBasedRLEnvCfg | DirectRLEnvCfg | DirectMARLEnvCfg, agen
         }
         print("[INFO] Recording videos during training.")
         print_dict(video_kwargs, nesting=4)
-        env = gym.wrappers.RecordVideo(env, **video_kwargs)
+        env = RecordVideo(env, **video_kwargs)
     # wrap around environment for stable baselines
     env = Sb3VecEnvWrapper(env, fast_variant=not args_cli.keep_all_info)
 

scripts/reinforcement_learning/sb3/train.py

@@ -77,6 +77,7 @@ def cleanup_pbar(*args):
 import os
 import random
 from datetime import datetime
+from gymnasium.wrappers import RecordVideo
 
 import omni
 from stable_baselines3 import PPO
@@ -92,12 +93,16 @@ def cleanup_pbar(*args):
 )
 from isaaclab.utils.dict import print_dict
 from isaaclab.utils.io import dump_pickle, dump_yaml
+from isaaclab.utils.recorder import stop_recording
 
 from isaaclab_rl.sb3 import Sb3VecEnvWrapper, process_sb3_cfg
 
 import isaaclab_tasks  # noqa: F401
 from isaaclab_tasks.utils.hydra import hydra_task_config
 
+# monkey-patch the stop_recording function to fix memory leak error (see https://github.com/Farama-Foundation/Gymnasium/pull/1444)
+RecordVideo.stop_recording = stop_recording
+
 # PLACEHOLDER: Extension template (do not remove this comment)
 
 
@@ -169,7 +174,7 @@ def main(env_cfg: ManagerBasedRLEnvCfg | DirectRLEnvCfg | DirectMARLEnvCfg, agen
         }
         print("[INFO] Recording videos during training.")
         print_dict(video_kwargs, nesting=4)
-        env = gym.wrappers.RecordVideo(env, **video_kwargs)
+        env = RecordVideo(env, **video_kwargs)
 
     # wrap around environment for stable baselines
     env = Sb3VecEnvWrapper(env, fast_variant=not args_cli.keep_all_info)

scripts/reinforcement_learning/skrl/play.py

@@ -79,6 +79,7 @@
 import random
 import time
 import torch
+from gymnasium.wrappers import RecordVideo
 
 import skrl
 from packaging import version
@@ -97,6 +98,7 @@
 elif args_cli.ml_framework.startswith("jax"):
     from skrl.utils.runner.jax import Runner
 
+
 from isaaclab.envs import (
     DirectMARLEnv,
     DirectMARLEnvCfg,
@@ -106,13 +108,17 @@
 )
 from isaaclab.utils.dict import print_dict
 from isaaclab.utils.pretrained_checkpoint import get_published_pretrained_checkpoint
+from isaaclab.utils.recorder import stop_recording
 
 from isaaclab_rl.skrl import SkrlVecEnvWrapper
 
 import isaaclab_tasks  # noqa: F401
 from isaaclab_tasks.utils import get_checkpoint_path
 from isaaclab_tasks.utils.hydra import hydra_task_config
 
+# monkey-patch the stop_recording function to fix memory leak error (see https://github.com/Farama-Foundation/Gymnasium/pull/1444)
+RecordVideo.stop_recording = stop_recording
+
 # PLACEHOLDER: Extension template (do not remove this comment)
 
 # config shortcuts
@@ -188,7 +194,7 @@ def main(env_cfg: ManagerBasedRLEnvCfg | DirectRLEnvCfg | DirectMARLEnvCfg, expe
         }
         print("[INFO] Recording videos during training.")
         print_dict(video_kwargs, nesting=4)
-        env = gym.wrappers.RecordVideo(env, **video_kwargs)
+        env = RecordVideo(env, **video_kwargs)
 
     # wrap around environment for skrl
     env = SkrlVecEnvWrapper(env, ml_framework=args_cli.ml_framework)  # same as: `wrap_env(env, wrapper="auto")`