Training with an RL Agent

In the previous tutorials, we covered how to define an RL task environment, register it into the gym registry, and interact with it using a random agent. We now move on to the next step: training an RL agent to solve the task.

Although the envs.ManagerBasedRLEnv conforms to the gymnasium.Env interface, it is not exactly a gym environment. The input and outputs of the environment are not numpy arrays, but rather based on torch tensors with the first dimension being the number of environment instances.

Additionally, most RL libraries expect their own variation of an environment interface. For example, Stable-Baselines3 expects the environment to conform to its VecEnv API which expects a list of numpy arrays instead of a single tensor. Similarly, RSL-RL, RL-Games and SKRL expect a different interface. Since there is no one-size-fits-all solution, we do not base the envs.ManagerBasedRLEnv on any particular learning library. Instead, we implement wrappers to convert the environment into the expected interface. These are specified in the isaaclab_rl module.

In this tutorial, we will use Stable-Baselines3 to train an RL agent to solve the cartpole balancing task.

Caution

Wrapping the environment with the respective learning framework’s wrapper should happen in the end, i.e. after all other wrappers have been applied. This is because the learning framework’s wrapper modifies the interpretation of environment’s APIs which may no longer be compatible with gymnasium.Env.

The Code

For this tutorial, we use the training implementation from Stable-Baselines3 workflow in the isaaclab_rl.entrypoints.backends.train_sb3 module.

Code for train_sb3.py

# Copyright (c) 2022-2026, The Isaac Lab Project Developers (https://github.com/isaac-sim/IsaacLab/blob/main/CONTRIBUTORS.md).
# All rights reserved.
#
# SPDX-License-Identifier: BSD-3-Clause

"""Stable-Baselines3 training backend of the unified reinforcement learning entrypoint."""

from __future__ import annotations

import argparse
import contextlib
import gc
import logging
import os
import signal
import sys
import time
from datetime import datetime
from pathlib import Path

import numpy as np
from stable_baselines3 import PPO
from stable_baselines3.common.callbacks import CheckpointCallback, LogEveryNTimesteps
from stable_baselines3.common.vec_env import VecNormalize

from isaaclab.app import add_launcher_args, launch_simulation, report_activity
from isaaclab.envs import DirectMARLEnvCfg
from isaaclab.utils.seed import configure_seed

import isaaclab_tasks  # noqa: F401
from isaaclab_tasks.utils import resolve_task_config, setup_preset_cli

from ...sb3 import Sb3VecEnvWrapper, process_sb3_cfg
from ..common import (
    CHECKPOINT_SELECTORS,
    add_common_train_args,
    apply_env_overrides,
    apply_video_recording,
    close_env,
    create_isaaclab_env,
    dump_train_configs,
    enable_cameras_for_video,
    pre_launch_video_config,
    resolve_checkpoint_selector,
    set_hydra_args,
    show_run_summary,
    startup_screen,
    wrap_sensor_capture,
    write_run_manifest,
)
from . import cli_args_sb3 as cli_args

logger = logging.getLogger(__name__)

# PLACEHOLDER: Extension template (do not remove this comment)
with contextlib.suppress(ImportError):
    import isaaclab_tasks_experimental  # noqa: F401


def _cleanup_pbar(*args) -> None:
    """Stop training and close the rich progress bars on Ctrl+C."""
    for obj in gc.get_objects():
        if "tqdm_rich" in type(obj).__name__:
            obj.close()
    raise KeyboardInterrupt


def _parse_args(argv: list[str]) -> argparse.Namespace:
    """Parse Stable-Baselines3 training arguments."""
    parser = argparse.ArgumentParser(description="Train an RL agent with Stable-Baselines3.")
    add_common_train_args(
        parser,
        agent_default="sb3_cfg_entry_point",
        agent_help="Name of the RL agent configuration entry point.",
        include_distributed=False,
    )
    parser.add_argument("--log_interval", type=int, default=100_000, help="Log data every n timesteps.")
    parser.add_argument("--checkpoint", type=str, default=None, help="Checkpoint path, or latest/best.")
    parser.add_argument(
        "--keep_all_info",
        action="store_true",
        default=False,
        help="Use a slower SB3 wrapper but keep all the extra training info.",
    )
    add_launcher_args(parser)
    args_cli, hydra_args = setup_preset_cli(parser, argv)
    enable_cameras_for_video(args_cli)
    set_hydra_args(hydra_args)
    return args_cli


def _resolve_checkpoint(args_cli: argparse.Namespace, log_root_path: str) -> str | None:
    """Resolve the checkpoint to resume from, or None when training starts from scratch."""
    if args_cli.checkpoint in CHECKPOINT_SELECTORS:
        return resolve_checkpoint_selector(
            log_root_path,
            args_cli.checkpoint,
            library="sb3",
            task=args_cli.task,
            checkpoint_pattern=r"model(?:_.*)?\.zip",
            preferred_checkpoint_pattern=r"model\.zip",
            metadata={"agent": args_cli.agent},
        )
    return args_cli.checkpoint


def run(argv: list[str]) -> None:
    """Train a Stable-Baselines3 agent."""
    signal.signal(signal.SIGINT, _cleanup_pbar)
    args_cli = _parse_args(argv)
    with startup_screen(args_cli, num_stages=3) as screen:
        env_cfg, agent_cfg = resolve_task_config(args_cli.task, args_cli.agent)
        pre_launch_video_config(env_cfg, args_cli)
        screen.stage("Launching simulation")
        with launch_simulation(env_cfg, args_cli), contextlib.ExitStack() as cleanup:
            show_run_summary(screen, args_cli, env_cfg, library="sb3", action="train")
            apply_env_overrides(args_cli, env_cfg)
            agent_cfg = cli_args.update_sb3_cfg(agent_cfg, args_cli)
            if args_cli.max_iterations is not None:
                agent_cfg["n_timesteps"] = args_cli.max_iterations * agent_cfg["n_steps"] * env_cfg.scene.num_envs
            env_cfg.seed = agent_cfg["seed"]

            log_root_path = os.path.abspath(os.path.join("logs", "sb3", args_cli.task))
            logger.info(f"Logging experiment in directory: {log_root_path}")
            run_name = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
            logger.info(f"Exact experiment name requested from command line: {run_name}")
            log_dir = os.path.join(log_root_path, run_name)
            write_run_manifest(log_dir, library="sb3", task=args_cli.task, metadata={"agent": args_cli.agent})
            dump_train_configs(log_dir, env_cfg, agent_cfg)
            (Path(log_dir) / "command.txt").write_text(" ".join(sys.orig_argv))

            checkpoint_path = _resolve_checkpoint(args_cli, log_root_path)
            agent_cfg = process_sb3_cfg(agent_cfg, env_cfg.scene.num_envs)
            policy_arch = agent_cfg.pop("policy")
            n_timesteps = agent_cfg.pop("n_timesteps")
            norm_args = {}
            for key in ("normalize_input", "normalize_value", "clip_obs"):
                if key in agent_cfg:
                    norm_args[key] = agent_cfg.pop(key)

            env_cfg.log_dir = log_dir
            apply_video_recording(env_cfg, log_dir, args_cli)

            screen.stage("Creating environment")
            env = create_isaaclab_env(
                args_cli.task,
                env_cfg,
                args_cli,
                convert_marl_to_single_agent=isinstance(env_cfg, DirectMARLEnvCfg),
            )
            cleanup.callback(lambda: close_env(env))
            env = wrap_sensor_capture(env, log_dir, args_cli)

            screen.stage("Preparing agent")
            start_time = time.time()
            report_activity("Wrapping environment")
            env = Sb3VecEnvWrapper(env, fast_variant=not args_cli.keep_all_info)
            report_activity(None)
            if norm_args.get("normalize_input"):
                print(f"Normalizing input, {norm_args=}")
                env = VecNormalize(
                    env,
                    training=True,
                    norm_obs=norm_args["normalize_input"],
                    norm_reward=norm_args.get("normalize_value", False),
                    clip_obs=norm_args.get("clip_obs", 100.0),
                    gamma=agent_cfg["gamma"],
                    clip_reward=np.inf,
                )

            report_activity("Building policy")
            agent = PPO(policy_arch, env, verbose=1, tensorboard_log=log_dir, **agent_cfg)
            report_activity(None)
            if checkpoint_path is not None:
                agent = agent.load(checkpoint_path, env, print_system_info=True)

            # configure_seed must run after PPO construction and loading so torch determinism does not disturb
            # SB3's initialization
            if args_cli.deterministic:
                configure_seed(env_cfg.seed, torch_deterministic=True)

            callbacks = [
                CheckpointCallback(save_freq=1000, save_path=log_dir, name_prefix="model", verbose=2),
                LogEveryNTimesteps(n_steps=args_cli.log_interval),
            ]

            screen.close()
            with contextlib.suppress(KeyboardInterrupt):
                agent.learn(total_timesteps=n_timesteps, callback=callbacks, progress_bar=True, log_interval=None)

            agent.save(os.path.join(log_dir, "model"))
            print(f"Saving to:\n{os.path.join(log_dir, 'model.zip')}")
            if isinstance(env, VecNormalize):
                print("Saving normalization")
                env.save(os.path.join(log_dir, "model_vecnormalize.pkl"))
            print(f"Training time: {round(time.time() - start_time, 2)} seconds")

The Code Explained

Most of the code above is boilerplate code to create logging directories, saving the parsed configurations, and setting up different Stable-Baselines3 components. For this tutorial, the important part is creating the environment and wrapping it with the Stable-Baselines3 wrapper.

There are three wrappers used in the code above:

1. gymnasium.wrappers.RecordVideo: This wrapper records a video of the environment and saves it to the specified directory. This is useful for visualizing the agent’s behavior during training.

2. wrappers.sb3.Sb3VecEnvWrapper: This wrapper converts the environment into a Stable-Baselines3 compatible environment.

3. stable_baselines3.common.vec_env.VecNormalize: This wrapper normalizes the environment’s observations and rewards.

Each of these wrappers wrap around the previous wrapper by following env = wrapper(env, *args, **kwargs) repeatedly. The final environment is then used to train the agent. For more information on how these wrappers work, please refer to the Wrapping environments documentation.

The Code Execution

We train a PPO agent from Stable-Baselines3 to solve the cartpole balancing task.

Training the agent

There are three main ways to train the agent. Each of them has their own advantages and disadvantages. It is up to you to decide which one you prefer based on your use case.

Headless execution

When no visualizer is requested, no interactive visualizer window is opened during training. This is useful when training on a remote server or when you do not need live visual feedback, which can add some compute cost. Rendering can still be active for sensor/camera data capture when enabled by the workflow.

uv (Recommended)

uv run --extra sb3 isaaclab train --rl_library sb3 --task Isaac-Cartpole --num_envs 64

Headless execution with off-screen render

Since the above command does not open an interactive visualizer, it is not possible to monitor behavior live in a viewport window. To capture visual output during training, enable camera/sensor rendering in the workflow and pass --video to record the agent behavior.

uv (Recommended)

uv run --extra sb3,video isaaclab train --rl_library sb3 --task Isaac-Cartpole --num_envs 64 --video

The videos are saved to the logs/sb3/Isaac-Cartpole/<run-dir>/videos/train directory. You can open these videos using any video player.

For tasks with on-scene cameras, you can also save the sensor image outputs directly during training with --capture_env_sensors. See Capturing sensor frames during training for the available options and output formats.

Interactive execution

While the above two methods are useful for training the agent, they don’t allow you to interact with the simulation to see what is happening. In this case, run the training command as follows:

uv (Recommended)

uv run --extra sb3 isaaclab train --rl_library sb3 --task Isaac-Cartpole --num_envs 64 --viz kit

This will open the Kit visualizer window and you can see the agent training in the environment. However, this can slow down the training process because interactive visual feedback is enabled. As a workaround, you can switch between different render modes in the "Isaac Lab" window that is docked on the bottom-right corner of the screen. To learn more about these render modes, please check the sim.SimulationContext.RenderMode class.

Viewing the logs

On a separate terminal, you can monitor the training progress by executing the following command:

uv (Recommended)

# execute from the root directory of the repository
uv run python -m tensorboard.main --logdir logs/sb3/Isaac-Cartpole

Playing the trained agent

Once the training is complete, you can visualize the trained agent by executing the following command:

uv (Recommended)

# execute from the root directory of the repository
uv run --extra sb3 isaaclab play --rl_library sb3 --task Isaac-Cartpole --num_envs 32 --viz kit

The above command will load the latest checkpoint from the logs/sb3/Isaac-Cartpole directory. You can also specify a specific checkpoint by passing the --checkpoint flag.