FlmainGO-Light Training Guide

Last updated: 2025-01-04

This is a Linux-focused, copy/paste-friendly guide for running FlaminGO Light reinforcement learning workflows on Isaac Lab + rsl_rl, based on:

• Repository: https://github.com/jaykorea/cocelo-rl-isaaclab

---

1. Installation options

Choose one of the following paths.

• Option A (recommended): Isaac Sim via pip + Isaac Lab from source

  • Pros: simplest workflow; works well with Python virtual environments

  • Cons: on Linux, requires GLIBC 2.35+ (Ubuntu 22.04 is typically OK; Ubuntu 20.04 often is not)

• Option B: Isaac Sim standalone binaries + Isaac Lab from source

  • Recommended if you encounter GLIBC or system compatibility issues with the pip package.

• Option C (advanced): Isaac Lab via pip

  • Useful when you want Isaac Lab as a library dependency in your own project. You may not get the full Isaac Lab repository scripts/examples experience.

• Option D: Docker

  • Good for servers and reproducible environments.

---

2. Prerequisites

2.1 System requirements (high-level)

• Linux workstation with NVIDIA GPU (recommended) and a modern NVIDIA driver

• Sufficient disk space (Isaac Sim + caches can be large)

• Stable internet connection (first launch may download extensions/caches)

Note: Exact driver/CUDA compatibility can vary by Isaac Sim version. If you hit runtime/graphics issues, confirm your driver meets the requirements for your chosen Isaac Sim release.

2.2 Tools

• git

• cmake, build-essential (for Isaac Lab native dependencies)

• Python 3.11 environment (Isaac Sim 5.x uses Python 3.11)

Install build tools (Ubuntu/Debian):

Install build tools (Ubuntu/Debian)

sudo apt update
sudo apt install -y git cmake build-essential

---

3.1 Option A (recommended): Isaac Sim (pip) + Isaac Lab (source)

Use this option if your system has GLIBC 2.35+ and you prefer a clean Python virtualenv-based workflow.

1

Verify GLIBC version

ldd --version

If GLIBC is below 2.35, use Option B (standalone binaries).

2

Create and activate a Python 3.11 environment

Option 1: Conda

conda create -n env_isaaclab python=3.11
conda activate env_isaaclab

Option 2: uv

uv venv --python 3.11 --seed env_isaaclab
source env_isaaclab/bin/activate

3

Upgrade pip

pip install --upgrade pip

4

Install Isaac Sim (pip)

Example: pin to 5.1.0 (adjust as needed)

pip install "isaacsim[all,extscache]==5.1.0" --extra-index-url https://pypi.nvidia.com

5

Install PyTorch (CUDA build)

Example for x86_64 CUDA wheels (adjust to match your CUDA/driver constraints):

pip install -U torch==2.7.0 torchvision==0.22.0 --index-url https://download.pytorch.org/whl/cu128

6

Verify Isaac Sim launches

First launch may download extensions and take time.

GUI launch:

isaacsim

Optional (launch a specific experience):

isaacsim isaacsim.exp.full.kit

7

Install Isaac Lab (from source)

Clone the repo:

git clone https://github.com/isaac-sim/IsaacLab.git
cd IsaacLab

Install native deps:

sudo apt update
sudo apt install -y cmake build-essential

Run the installer script:

./isaaclab.sh --install   # same as ./isaaclab.sh -i

Install only a specific RL backend (example: rl_games):

./isaaclab.sh --install rl_games

8

Verify Isaac Lab runs

Recommended (use the helper script):

./isaaclab.sh -p scripts/tutorials/00_sim/create_empty.py

Alternative (run directly with Python from your active venv):

python scripts/tutorials/00_sim/create_empty.py

Expected result:

• A viewport window opens (often empty/black for an empty scene); exit with Ctrl+C.

9

(Optional) Run a sample training job (headless)

./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py --task=Isaac-Ant-v0 --headless

---

3.2 Option B: Isaac Sim (standalone binaries) + Isaac Lab (source)

Use this path if GLIBC is too old for the pip package or you prefer a self-contained Isaac Sim installation.

1

Install Isaac Sim (standalone zip)

Example for Linux x86_64 zip. Adjust the filename/path for the package you downloaded.

mkdir -p ~/isaacsim
cd ~/Downloads
unzip "isaac-sim-standalone-5.1.0-linux-x86_64.zip" -d ~/isaacsim
cd ~/isaacsim
./post_install.sh
./isaac-sim.selector.sh

2

Configure environment variables and verify

export ISAACSIM_PATH="${HOME}/isaacsim"
export ISAACSIM_PYTHON_EXE="${ISAACSIM_PATH}/python.sh"

# GUI launch
${ISAACSIM_PATH}/isaac-sim.sh

# Python sanity checks
${ISAACSIM_PYTHON_EXE} -c "print('Isaac Sim configuration is now complete.')"
${ISAACSIM_PYTHON_EXE} ${ISAACSIM_PATH}/standalone_examples/api/isaacsim.core.api/add_cubes.py

If you suspect cache/config conflicts from older installs:

${ISAACSIM_PATH}/isaac-sim.sh --reset-user

3

Install Isaac Lab (source) and link to Isaac Sim

Clone Isaac Lab:

git clone https://github.com/isaac-sim/IsaacLab.git
cd IsaacLab

Link Isaac Sim into the repo (_isaac_sim):

ln -s ${ISAACSIM_PATH} _isaac_sim

Install + verify:

sudo apt update
sudo apt install -y cmake build-essential

./isaaclab.sh --install
./isaaclab.sh -p scripts/tutorials/00_sim/create_empty.py

---

3.3 Option C (advanced): Isaac Lab via pip

Use this if you want Isaac Lab as a dependency in your own codebase rather than running the full repo scripts.

pip install "isaaclab[isaacsim,all]==2.3.0" --extra-index-url https://pypi.nvidia.com

Note: Version constraints for isaaclab and isaacsim should be aligned. If you encounter dependency conflicts, prefer Option A/B or consult the official docs for the supported version matrix.

---

3.4 Option D: Docker

3.4.1 Use the Isaac Lab repository Docker workflow

git clone https://github.com/isaac-sim/IsaacLab.git
cd IsaacLab

# Start container (detached)
./docker/container.py start

# Enter container
./docker/container.py enter base

Tip: Many setups work more smoothly if the Isaac Lab repo is located under /home to avoid permission and volume-mount edge cases.

✅ Important Notice

To ensure everything works correctly, please install rsl-rl-lib version 2.2.4 using pip.

Command:

pip install rsl-rl-lib==2.2.4

⚠️ Note: Other versions may cause unexpected issues—2.2.4 is required.

---

4. Workflow

1

Clone this repository

Run:

git clone https://github.com/jaykorea/cocelo-rl-isaaclab.git
cd cocelo-rl-isaaclab

2

Activate your Isaac Lab Python environment

Use the same Python environment you used for Isaac Lab. (Example below uses conda.)

conda activate env_isaaclab

3

Install this repo in editable mode

Run from the repo root:

pip install -e .

4

(Important) Unzip assets (.usd)

Some .usd assets may not be included as raw files through Git. If the repo provides zipped assets, make sure the .usd files actually exist on disk after unzipping.

• Example path mentioned in the repo:

  • lab/flamingo/assets/data/Robots/Flamingo/flamingo_rev01_4_1/

✅ Quick check: confirm you see .usd files under the expected directory before training.

5

(Optional) Sanity check: list environments

If you have the Isaac Lab launcher available, you can list environments and filter by keyword:

./isaaclab.sh -p scripts/environments/list_envs.py --keyword Flamingo

If you don’t have isaaclab.sh in your current setup, skip this step.

6

Train (rsl_rl)

Training script:

• scripts/reinforcement_learning/rsl_rl/train.py

Basic command:

python scripts/reinforcement_learning/rsl_rl/train.py \
  --task {TASK_NAME} \
  --algo ppo \
  --num_envs 4096 \
  --headless \
  --num_policy_stacks {POLICY_STACKS} \
  --num_critic_stacks {CRITIC_STACKS}

Example (Track velocity):

python scripts/reinforcement_learning/rsl_rl/train.py \
  --task Isaac-Velocity-Flat-Flamingo-v1-ppo \
  --algo ppo \
  --num_envs 4096 \
  --headless \
  --num_policy_stacks 2 \
  --num_critic_stacks 2

7

Play (run a trained policy)

Play script:

• scripts/reinforcement_learning/rsl_rl/play.py

Basic command:

python scripts/reinforcement_learning/rsl_rl/play.py \
  --task {TASK_NAME} \
  --algo ppo \
  --num_envs 64 \
  --num_policy_stacks {POLICY_STACKS} \
  --num_critic_stacks {CRITIC_STACKS} \
  --load_run {RUN_FOLDER_NAME} \
  --plot False

Example (Track velocity):

python scripts/reinforcement_learning/rsl_rl/play.py \
  --task Isaac-Velocity-Flat-Flamingo-Play-v1-ppo \
  --algo ppo \
  --num_envs 64 \
  --num_policy_stacks 2 \
  --num_critic_stacks 2 \
  --load_run 2025-03-16_17-09-35 \
  --plot False

8

Where results are saved (logs)

Isaac Lab RL scripts typically write runs under:

• logs/<library>/<task>/<date-time>/

For rsl_rl, that usually becomes:

• logs/rsl_rl/<task>/<date-time>/

You can point --load_run to the <date-time> folder when playing.

---

---

Troubleshooting

Installing IsaacLab

1) GLIBC errors (pip install / launch fails)

Symptom: GLIBC_2.35 not found (or similar)

Resolution: Use Option B (standalone binaries) or upgrade your OS to a GLIBC 2.35+ distribution.

2) First launch is slow / downloads many assets

The first isaacsim run may download extensions and compile caches. Subsequent launches are typically faster.

3) Black window / rendering issues

• Confirm your NVIDIA driver is installed and functional.

• Try resetting user config:

For pip-installed Isaac Sim:

isaacsim --reset-user

For standalone:

${ISAACSIM_PATH}/isaac-sim.sh --reset-user

4) Cache/config conflicts after reinstall

Reset user settings as above, and consider clearing Omniverse cache directories if needed.

Installing FlaminGO Light

1) No module named ... or scripts import errors

• Ensure you run commands from the repo root (cocelo-rl-isaaclab/).

• Confirm the environment is active and pip install -e . completed without errors.

2) Robot/scene fails to load or asset path errors

• Confirm .usd files exist under lab/flamingo/assets/....

• If needed, unzip the assets into the directory structure expected by the repo.

3) Isaac Sim/Isaac Lab mismatch

• Ensure the Isaac Lab version you installed matches the Isaac Sim version required by your workflow.

• If in doubt, follow the official Isaac Lab installation + quickstart docs again.

---

References

• cocelo-rl-isaaclab: https://github.com/jaykorea/cocelo-rl-isaaclab

• Isaac Lab binaries install: https://isaac-sim.github.io/IsaacLab/main/source/setup/installation/binaries_installation.html

• Isaac Lab quickstart: https://isaac-sim.github.io/IsaacLab/main/source/setup/quickstart.html

• Isaac Lab RL scripts overview: https://isaac-sim.github.io/IsaacLab/main/source/overview/reinforcement-learning/rl_existing_scripts.html

• Isaac Lab Docker guide: https://isaac-sim.github.io/IsaacLab/main/source/deployment/docker.html

• Isaac Sim docs: https://docs.isaacsim.omniverse.nvidia.com/