Introducing LuckyWorld

Featured robots in this release

So100

Open-Source Manipulator.

Stretch

Mobile manipulator robot designed for home environments.

DJI Matrice 300 RTK

Commercial drone platform for precision mapping and industrial applications.

Here’s how it works.

→

Installation & Setup

→

Recording Episodes

→

Upload & HuggingFace

→

Training Policy

→

Inference

Installation & Setup

Recording Episodes

Upload & HuggingFace

# Install the Hugging Face CLI
pip install -U "huggingface_hub[cli]"

# Login with your Hugging Face credentials
huggingface-cli login

# Push your dataset files
huggingface-cli upload luckyrobots/so100_Dataset_250_V0.1 . --repo-type=dataset

# Setting Up Tag
python lerobot\examples\0_setup_dataset_tag.py

Training Policy

# Train your policy with diffusion model
# This command will start training using the SO100 dataset

python lerobot/scripts/train.py \
    --dataset.repo_id=luckyrobots/so100_Dataset_250_V0.1 \
    --policy.type=diffusion \
    --dataset.video_backend=torchcodec \
    2>&1 | tee training_log.txt

# Parameters explained:
# --dataset.repo_id: HuggingFace dataset repository
# --policy.type: Policy type (e.g. Diffusion, pi0, ACT)
# --dataset.video_backend: Video processing backend
# 2>&1 | tee: Save training logs to training_log.txt

Inference

# Run evaluation with trained policy
python lerobot/scripts/eval.py \
    --policy.path=outputs/train/diffusion/checkpoints/020000/pretrained_model \
    --env.type=luckyworld \
    --eval.use_async_envs=false \
    --eval.n_episodes=5 \
    --eval.batch_size=1 \
    --policy.use_amp=false \
    --policy.device=cuda

Get your robots ready for the real world.

Whether you’re building cutting-edge AI, refining autonomous systems, or accelerating robotics research, Lucky gives you the synthetic data and simulation tools to bridge the gap between the lab and the real world.

Get Started

Visit Our Github

Join Us On Discord

Find Us On X

Visit Our Github

Join Us On Discord

Find Us On X