Enable Video Recording During Custom Training with SKRL in IsaacLab

[RodsCoimbra]

Hello,
I'm using SKRL in IsaacLab with a custom network that can't be defined using the usual YAML configuration files used for train.py provided by IsaacLab. I followed the NVIDIA Isaac Lab examples from SKRL to implement this custom setup.
Now, I would like to record videos during training, similar to what the default train.py implementation in IsaacLab uses with flags like --video --video_length 100 --video_interval 10000.
However, when trying to implement video recording similar to that file, I encountered the following error:

Traceback (most recent call last):
  ...
RuntimeError: Cannot render 'rgb_array' when the simulation render mode is 'NO_GUI_OR_RENDERING'. Please set the simulation render mode to: 'PARTIAL_RENDERING' or 'FULL_RENDERING'. If running headless, make sure --enable_cameras is set.
...
UserWarning: WARN: Ignored saving a video as there were zero frames to save.
  logger.warn("Ignored saving a video as there were zero frames to save.")

The code I used was the following:

from skrl.agents.torch.ppo import PPO_RNN as PPO
from skrl.envs.loaders.torch import load_isaaclab_env
from skrl.envs.wrappers.torch import wrap_env
from skrl.memories.torch import RandomMemory
from skrl.trainers.torch import SequentialTrainer
from skrl.utils import set_seed
import os
from datetime import datetime
from cfg_file import agent_cfg
from network import Policy, Value
import sys
import gymnasium as gym

video_flag = True

set_seed(42)  
env = load_isaaclab_env(task_name="Isaac-Tiago-Direct-V0")
datetime = datetime.now().strftime("%d-%m_%H-%M-%S")

if video_flag:
    video_kwargs = {
        "video_folder": os.path.join(datetime, "videos", "train"),
        "step_trigger": lambda step: step % 10000 == 0,
        "video_length": 100,
        "disable_logger": True,
    }
    print("[INFO] Recording videos during training.")
    env = gym.wrappers.RecordVideo(env, **video_kwargs)
    
env = wrap_env(env)

device = env.device
memory = RandomMemory(memory_size=128, num_envs=env.num_envs, device=device)

models = {}
models["policy"] = Policy(env.observation_space, env.action_space, device, num_envs=env.num_envs)
models["value"] = Value(env.observation_space, env.action_space, device, num_envs=env.num_envs)

timesteps = 100_000
cfg = agent_cfg(env, device, timesteps)
log_dir = datetime + f"_PPO_{cfg['experiment']['experiment_name']}"
experiment_dir = os.path.join(cfg["experiment"]["directory"], log_dir)
cfg['experiment']['experiment_name'] = experiment_dir
print(f"Exact experiment name requested from command line {log_dir}")

agent = PPO(models=models,
            memory=memory,
            cfg=cfg,
            observation_space=env.observation_space,
            action_space=env.action_space,
            device=device)

# configure and instantiate the RL trainer
cfg_trainer = {"timesteps": timesteps, "headless": True, "environment_info": "log"}
trainer = SequentialTrainer(cfg=cfg_trainer, env=env, agents=agent)

# start training
trainer.train()

What could be causing this issue? Is there a working example of using video recording with a custom training loop?
Thanks in advance!