robotis_lab

ISAAC_FFW_RL_reach_play.mp4

Overview

robotis_lab is a research-oriented repository based on Isaac Lab, designed to enable reinforcement learning (RL) and imitation learning (IL) experiments using Robotis robots in simulation. This project provides simulation environments, configuration tools, and task definitions tailored for Robotis hardware, leveraging NVIDIA Isaac Sim’s powerful GPU-accelerated physics engine and Isaac Lab’s modular RL pipeline.

Important

This repository currently depends on IsaacLab v2.2.0 or higher.

Installation (Docker)

Docker installation provides a consistent environment with all dependencies pre-installed.

Prerequisites:

• Docker and Docker Compose installed
• NVIDIA Container Toolkit installed
• NVIDIA GPU with appropriate drivers

Steps:

1. Clone robotis_lab repository with submodules:

   git clone --recurse-submodules https://github.com/ROBOTIS-GIT/robotis_lab.git
   cd robotis_lab

   If you already cloned without submodules, initialize them:

   git submodule update --init --recursive

2. Build and start the Docker container:

   ./docker/container.sh start

3. Enter the container:

   ./docker/container.sh enter

Docker Commands:

• ./docker/container.sh start - Build and start the container
• ./docker/container.sh enter - Enter the running container
• ./docker/container.sh stop - Stop the container
• ./docker/container.sh logs - View container logs
• ./docker/container.sh clean - Remove container and image

What's included in the Docker image:

• Isaac Sim 5.1.0
• Isaac Lab v2.3.0 (from third_party submodule)
• CycloneDDS 0.10.2 (from third_party submodule)
• robotis_dds_python (from third_party submodule)
• LeRobot 0.3.3 (in separate virtual environment at ~/lerobot_env)
• All required dependencies and configurations

Try examples

Sim2Sim

Reinforcement learning

OMY Reach Task

# Train
python scripts/reinforcement_learning/rsl_rl/train.py --task RobotisLab-Reach-OMY-v0 --num_envs=512 --headless

# Play
python scripts/reinforcement_learning/rsl_rl/play.py --task RobotisLab-Reach-OMY-v0 --num_envs=16

OMY Lift Task

# Train
python scripts/reinforcement_learning/rsl_rl/train.py --task RobotisLab-Lift-Cube-OMY-v0 --num_envs=512 --headless

# Play
python scripts/reinforcement_learning/rsl_rl/play.py --task RobotisLab-Lift-Cube-OMY-v0 --num_envs=16

OMY Open drawer Task

# Train
python scripts/reinforcement_learning/rsl_rl/train.py --task RobotisLab-Open-Drawer-OMY-v0 --num_envs=512 --headless

# Play
python scripts/reinforcement_learning/rsl_rl/play.py --task RobotisLab-Open-Drawer-OMY-v0 --num_envs=16

FFW-BG2 reach Task

# Train
python scripts/reinforcement_learning/rsl_rl/train.py --task RobotisLab-Reach-FFW-BG2-v0 --num_envs=512 --headless

# Play
python scripts/reinforcement_learning/rsl_rl/play.py --task RobotisLab-Reach-FFW-BG2-v0 --num_envs=16

Imitation learning

If you want to control a SINGLE ROBOT with the keyboard during playback, add --keyboard at the end of the play script.

Key bindings:
=========================== =========================
Command                     Key
=========================== =========================
Toggle gripper (open/close) K      
Move arm along x-axis       W / S   
Move arm along y-axis       A / D
Move arm along z-axis       Q / E
Rotate arm along x-axis     Z / X
Rotate arm along y-axis     T / G
Rotate arm along z-axis     C / V
=========================== =========================

OMY Stack Task (Stack the blocks in the following order: blue → red → green.)

# Teleop and record
python scripts/imitation_learning/isaaclab_recorder/record_demos.py --task RobotisLab-Stack-Cube-OMY-IK-Rel-v0 --teleop_device keyboard --dataset_file ./datasets/dataset.hdf5 --num_demos 10

# Annotate
python scripts/imitation_learning/isaaclab_mimic/annotate_demos.py --device cuda --task RobotisLab-Stack-Cube-OMY-IK-Rel-Mimic-v0 --auto --input_file ./datasets/dataset.hdf5 --output_file ./datasets/annotated_dataset.hdf5 --headless

# Mimic data
python scripts/imitation_learning/isaaclab_mimic/generate_dataset.py \
--device cuda --num_envs 100 --generation_num_trials 1000 \
--input_file ./datasets/annotated_dataset.hdf5 --output_file ./datasets/generated_dataset.hdf5 --headless

# Train
python scripts/imitation_learning/robomimic/train.py \
--task RobotisLab-Stack-Cube-OMY-IK-Rel-v0 --algo bc \
--dataset ./datasets/generated_dataset.hdf5

# Play
python scripts/imitation_learning/robomimic/play.py \
--device cuda --task RobotisLab-Stack-Cube-OMY-IK-Rel-v0 --num_rollouts 50 \
--checkpoint /PATH/TO/desired_model_checkpoint.pth

FFW-BG2 Pick and Place Task (Move the red stick into the basket.)

# Teleop and record
python scripts/imitation_learning/isaaclab_recorder/record_demos.py --task RobotisLab-PickPlace-FFW-BG2-IK-Rel-v0 --teleop_device keyboard --dataset_file ./datasets/dataset.hdf5 --num_demos 10 --enable_cameras

# Annotate
python scripts/imitation_learning/isaaclab_mimic/annotate_demos.py --device cuda --task RobotisLab-PickPlace-FFW-BG2-Mimic-v0 --input_file ./datasets/dataset.hdf5 --output_file ./datasets/annotated_dataset.hdf5 --enable_cameras

# Mimic data
python scripts/imitation_learning/isaaclab_mimic/generate_dataset.py \
--device cuda --num_envs 20 --generation_num_trials 300 \
--input_file ./datasets/annotated_dataset.hdf5 --output_file ./datasets/generated_dataset.hdf5 --enable_cameras --headless

# Train
python scripts/imitation_learning/robomimic/train.py \
--task RobotisLab-PickPlace-FFW-BG2-IK-Rel-v0 --algo bc \
--dataset ./datasets/generated_dataset.hdf5

# Play
python scripts/imitation_learning/robomimic/play.py \
--device cuda --task RobotisLab-PickPlace-FFW-BG2-IK-Rel-v0  --num_rollouts 50 \
--checkpoint /PATH/TO/desired_model_checkpoint.pth --enable_cameras

Sim2Real

Important

OMY Hardware Setup: To run Sim2Real with the real OMY robot, you need to bring up the robot.

This can be done using ROBOTIS’s open_manipulator repository.

AI WORKER Hardware Setup: To run Sim2Real with the real AI WORKER robot, you need to bring up the robot.

This can be done using ROBOTIS’s ai_worker repository.

The training and inference of the collected dataset should be carried out using physical_ai_tools. This can be done using ROBOTIS’s physical_ai_tools

Reinforcement learning

OMY Reach Task Introduction YouTube

sim2real.mp4

Run Sim2Real Reach Policy on OMY

# Train
python scripts/reinforcement_learning/rsl_rl/train.py --task RobotisLab-Reach-OMY-v0 --num_envs=512 --headless

# Play (You must run rsl_rl play in order to generate the policy file.)
python scripts/reinforcement_learning/rsl_rl/play.py --task RobotisLab-Reach-OMY-v0 --num_envs=16

# Sim2Real
python scripts/sim2real/reinforcement_learning/inference/OMY/reach/run_omy_reach.py --model_dir=<2025-07-10_08-47-09>

Replace <2025-07-10_08-47-09> with the actual timestamp folder name under:

logs/rsl_rl/reach_omy/

Imitation learning

OMY Pick and Place Task

Sim2Sim

506326011-26108031-55b0-4809-8d56-1faa495920c1.mp4

Sim2Real

506326055-02cd503e-0f24-4bec-aea2-79010c21f874.mp4

# Teleop and record demos
python scripts/sim2real/imitation_learning/recorder/record_demos.py --task=RobotisLab-Real-Pick-Place-Bottle-OMY-v0 --robot_type OMY --dataset_file ./datasets/omy_pick_place_task.hdf5 --num_demos 10 --enable_cameras

[Option] Mimic generate dataset

# Data convert ee_pose action from joint action
python scripts/sim2real/imitation_learning/mimic/action_data_converter.py --input_file ./datasets/omy_pick_place_task.hdf5 --output_file ./datasets/processed_omy_pick_place_task.hdf5 --action_type ik

# Annotate dataset
python scripts/sim2real/imitation_learning/mimic/annotate_demos.py --task RobotisLab-Real-Mimic-Pick-Place-Bottle-OMY-v0 --auto --input_file ./datasets/processed_omy_pick_place_task.hdf5 --output_file ./datasets/annotated_dataset.hdf5 --enable_cameras --headless

# Generate dataset
python scripts/sim2real/imitation_learning/mimic/generate_dataset.py --device cuda --num_envs 10 --task RobotisLab-Real-Mimic-Pick-Place-Bottle-OMY-v0 --generation_num_trials 500 --input_file ./datasets/annotated_dataset.hdf5 --output_file ./datasets/generated_dataset.hdf5 --enable_cameras --headless

# Data convert joint action from ee_pose action
python scripts/sim2real/imitation_learning/mimic/action_data_converter.py --input_file ./datasets/generated_dataset.hdf5 --output_file ./datasets/processed_generated_dataset.hdf5 --action_type joint

# Data convert lerobot dataset from IsaacLab hdf dataset
lerobot-python scripts/sim2real/imitation_learning/data_converter/OMY/isaaclab2lerobot.py \
    --task=RobotisLab-Real-Pick-Place-Bottle-OMY-v0 \
    --robot_type OMY \
    --dataset_file ./datasets/<processed_omy_pick_place_task.hdf5> or ./datasets/<processed_generated_dataset.hdf5>

# Inference in simulation
python scripts/sim2real/imitation_learning/inference/inference_demos.py --task RobotisLab-Real-Pick-Place-Bottle-OMY-v0 --robot_type OMY --enable_cameras

License

This repository is licensed under the Apache 2.0 License. See LICENSE for details.

Third-party components

• Isaac Lab: BSD-3-Clause License, see LICENSE-IsaacLab