Training with an RL Agent

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 script from Stable-Baselines3 workflow in the scripts/reinforcement_learning/sb3 directory.

Code for train.py
  1# Copyright (c) 2022-2025, The Isaac Lab Project Developers (https://github.com/isaac-sim/IsaacLab/blob/main/CONTRIBUTORS.md).
  2# All rights reserved.
  3#
  4# SPDX-License-Identifier: BSD-3-Clause
  5
  6# Copyright (c) 2022-2025, The Isaac Lab Project Developers.
  7# All rights reserved.
  8#
  9# SPDX-License-Identifier: BSD-3-Clause
 10
 11"""Script to train RL agent with Stable Baselines3.
 12
 13Since Stable-Baselines3 does not support buffers living on GPU directly,
 14we recommend using smaller number of environments. Otherwise,
 15there will be significant overhead in GPU->CPU transfer.
 16"""
 17
 18"""Launch Isaac Sim Simulator first."""
 19
 20import argparse
 21import sys
 22
 23from isaaclab.app import AppLauncher
 24
 25# add argparse arguments
 26parser = argparse.ArgumentParser(description="Train an RL agent with Stable-Baselines3.")
 27parser.add_argument("--video", action="store_true", default=False, help="Record videos during training.")
 28parser.add_argument("--video_length", type=int, default=200, help="Length of the recorded video (in steps).")
 29parser.add_argument("--video_interval", type=int, default=2000, help="Interval between video recordings (in steps).")
 30parser.add_argument("--num_envs", type=int, default=None, help="Number of environments to simulate.")
 31parser.add_argument("--task", type=str, default=None, help="Name of the task.")
 32parser.add_argument("--seed", type=int, default=None, help="Seed used for the environment")
 33parser.add_argument("--max_iterations", type=int, default=None, help="RL Policy training iterations.")
 34# append AppLauncher cli args
 35AppLauncher.add_app_launcher_args(parser)
 36# parse the arguments
 37args_cli, hydra_args = parser.parse_known_args()
 38# always enable cameras to record video
 39if args_cli.video:
 40    args_cli.enable_cameras = True
 41
 42# clear out sys.argv for Hydra
 43sys.argv = [sys.argv[0]] + hydra_args
 44
 45# launch omniverse app
 46app_launcher = AppLauncher(args_cli)
 47simulation_app = app_launcher.app
 48
 49"""Rest everything follows."""
 50
 51import gymnasium as gym
 52import numpy as np
 53import os
 54import random
 55from datetime import datetime
 56
 57from stable_baselines3 import PPO
 58from stable_baselines3.common.callbacks import CheckpointCallback
 59from stable_baselines3.common.logger import configure
 60from stable_baselines3.common.vec_env import VecNormalize
 61
 62from isaaclab.envs import (
 63    DirectMARLEnv,
 64    DirectMARLEnvCfg,
 65    DirectRLEnvCfg,
 66    ManagerBasedRLEnvCfg,
 67    multi_agent_to_single_agent,
 68)
 69from isaaclab.utils.dict import print_dict
 70from isaaclab.utils.io import dump_pickle, dump_yaml
 71
 72from isaaclab_rl.sb3 import Sb3VecEnvWrapper, process_sb3_cfg
 73
 74import isaaclab_tasks  # noqa: F401
 75from isaaclab_tasks.utils.hydra import hydra_task_config
 76
 77# PLACEHOLDER: Extension template (do not remove this comment)
 78
 79
 80@hydra_task_config(args_cli.task, "sb3_cfg_entry_point")
 81def main(env_cfg: ManagerBasedRLEnvCfg | DirectRLEnvCfg | DirectMARLEnvCfg, agent_cfg: dict):
 82    """Train with stable-baselines agent."""
 83    # randomly sample a seed if seed = -1
 84    if args_cli.seed == -1:
 85        args_cli.seed = random.randint(0, 10000)
 86
 87    # override configurations with non-hydra CLI arguments
 88    env_cfg.scene.num_envs = args_cli.num_envs if args_cli.num_envs is not None else env_cfg.scene.num_envs
 89    agent_cfg["seed"] = args_cli.seed if args_cli.seed is not None else agent_cfg["seed"]
 90    # max iterations for training
 91    if args_cli.max_iterations is not None:
 92        agent_cfg["n_timesteps"] = args_cli.max_iterations * agent_cfg["n_steps"] * env_cfg.scene.num_envs
 93
 94    # set the environment seed
 95    # note: certain randomizations occur in the environment initialization so we set the seed here
 96    env_cfg.seed = agent_cfg["seed"]
 97    env_cfg.sim.device = args_cli.device if args_cli.device is not None else env_cfg.sim.device
 98
 99    # directory for logging into
100    run_info = datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
101    log_root_path = os.path.abspath(os.path.join("logs", "sb3", args_cli.task))
102    print(f"[INFO] Logging experiment in directory: {log_root_path}")
103    # The Ray Tune workflow extracts experiment name using the logging line below, hence, do not change it (see PR #2346, comment-2819298849)
104    print(f"Exact experiment name requested from command line: {run_info}")
105    log_dir = os.path.join(log_root_path, run_info)
106    # dump the configuration into log-directory
107    dump_yaml(os.path.join(log_dir, "params", "env.yaml"), env_cfg)
108    dump_yaml(os.path.join(log_dir, "params", "agent.yaml"), agent_cfg)
109    dump_pickle(os.path.join(log_dir, "params", "env.pkl"), env_cfg)
110    dump_pickle(os.path.join(log_dir, "params", "agent.pkl"), agent_cfg)
111
112    # post-process agent configuration
113    agent_cfg = process_sb3_cfg(agent_cfg)
114    # read configurations about the agent-training
115    policy_arch = agent_cfg.pop("policy")
116    n_timesteps = agent_cfg.pop("n_timesteps")
117
118    # create isaac environment
119    env = gym.make(args_cli.task, cfg=env_cfg, render_mode="rgb_array" if args_cli.video else None)
120
121    # convert to single-agent instance if required by the RL algorithm
122    if isinstance(env.unwrapped, DirectMARLEnv):
123        env = multi_agent_to_single_agent(env)
124
125    # wrap for video recording
126    if args_cli.video:
127        video_kwargs = {
128            "video_folder": os.path.join(log_dir, "videos", "train"),
129            "step_trigger": lambda step: step % args_cli.video_interval == 0,
130            "video_length": args_cli.video_length,
131            "disable_logger": True,
132        }
133        print("[INFO] Recording videos during training.")
134        print_dict(video_kwargs, nesting=4)
135        env = gym.wrappers.RecordVideo(env, **video_kwargs)
136
137    # wrap around environment for stable baselines
138    env = Sb3VecEnvWrapper(env)
139
140    if "normalize_input" in agent_cfg:
141        env = VecNormalize(
142            env,
143            training=True,
144            norm_obs="normalize_input" in agent_cfg and agent_cfg.pop("normalize_input"),
145            norm_reward="normalize_value" in agent_cfg and agent_cfg.pop("normalize_value"),
146            clip_obs="clip_obs" in agent_cfg and agent_cfg.pop("clip_obs"),
147            gamma=agent_cfg["gamma"],
148            clip_reward=np.inf,
149        )
150
151    # create agent from stable baselines
152    agent = PPO(policy_arch, env, verbose=1, **agent_cfg)
153    # configure the logger
154    new_logger = configure(log_dir, ["stdout", "tensorboard"])
155    agent.set_logger(new_logger)
156
157    # callbacks for agent
158    checkpoint_callback = CheckpointCallback(save_freq=1000, save_path=log_dir, name_prefix="model", verbose=2)
159    # train the agent
160    agent.learn(total_timesteps=n_timesteps, callback=checkpoint_callback)
161    # save the final model
162    agent.save(os.path.join(log_dir, "model"))
163
164    # close the simulator
165    env.close()
166
167
168if __name__ == "__main__":
169    # run the main function
170    main()
171    # close sim app
172    simulation_app.close()

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#

If the --headless flag is set, the simulation is not rendered during training. This is useful when training on a remote server or when you do not want to see the simulation. Typically, it speeds up the training process since only physics simulation step is performed.

./isaaclab.sh -p scripts/reinforcement_learning/sb3/train.py --task Isaac-Cartpole-v0 --num_envs 64 --headless

Headless execution with off-screen render#

Since the above command does not render the simulation, it is not possible to visualize the agent’s behavior during training. To visualize the agent’s behavior, we pass the --enable_cameras which enables off-screen rendering. Additionally, we pass the flag --video which records a video of the agent’s behavior during training.

./isaaclab.sh -p scripts/reinforcement_learning/sb3/train.py --task Isaac-Cartpole-v0 --num_envs 64 --headless --video

The videos are saved to the logs/sb3/Isaac-Cartpole-v0/<run-dir>/videos/train directory. You can open these videos using any video player.

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, you can ignore the --headless flag and run the training script as follows:

./isaaclab.sh -p scripts/reinforcement_learning/sb3/train.py --task Isaac-Cartpole-v0 --num_envs 64

This will open the Isaac Sim window and you can see the agent training in the environment. However, this will slow down the training process since the simulation is rendered on the screen. 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:

# execute from the root directory of the repository
./isaaclab.sh -p -m tensorboard.main --logdir logs/sb3/Isaac-Cartpole-v0

Playing the trained agent#

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

# execute from the root directory of the repository
./isaaclab.sh -p scripts/reinforcement_learning/sb3/play.py --task Isaac-Cartpole-v0 --num_envs 32 --use_last_checkpoint

The above command will load the latest checkpoint from the logs/sb3/Isaac-Cartpole-v0 directory. You can also specify a specific checkpoint by passing the --checkpoint flag.