Fine-tune GR00T N1.5 for LeRobot SO-101 Arm and Deploy on Jetson Thor

Execution model

Run one phase at a time. After each phase:

• Relay all command output to the user.
• If output contains [STOP] → stop immediately, consult the failure decision tree below.
• If output ends with [OK] → tell the user "Phase N complete" and proceed to the next phase.

Prerequisites

Phase 1 — Flash Thor system image (~20 min)

Download the ISO image from https://developer.nvidia.com/embedded/jetpack/downloads (select Thor 38.2).

Create bootable USB with Balena Etcher (https://etcher.balena.io/):

1. Select the ISO image
2. Select the USB drive as target
3. Click Flash and wait for completion

Insert USB, keyboard, display cable, and power into Thor. Boot and select:

• Boot Manager → select USB drive → Esc → Continue
• Jetson Thor options → Flash Jetson AGX Thor Developer Kit on NVMe

Wait ~15 minutes for flashing, then wait for Update Progress to reach 100%.

[OK] when Ubuntu 24.04 initial setup screen appears. [STOP] if screen goes black — see FAQ in source.

Phase 2 — Install development dependencies (~10–20 min)

Install JetPack SDK:

sudo apt update
sudo apt install nvidia-jetpack

Install additional tools:

sudo apt install firefox python3 python3-pip
sudo pip3 install -U pip
sudo pip3 install jetson-stats

Install Miniconda:

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-aarch64.sh
chmod +x Miniconda3-latest-Linux-aarch64.sh
./Miniconda3-latest-Linux-aarch64.sh
source ~/.bashrc
conda --version

For GPU PyTorch on Thor, download pre-compiled wheels (Python 3.10 + CUDA 13):

• PyTorch 2.9 and TorchVision 0.24 wheels from Seeed SharePoint links in references/source.body.md
• Additional wheels at https://pypi.jetson-ai-lab.io/sbsa/cu130

[OK] when conda --version and python3 -c "import torch; print(torch.cuda.is_available())" both succeed.

Phase 3 — Data collection with SO-ARM

Follow the complete SO-ARM tutorial at: https://wiki.seeedstudio.com/lerobot_so100m_new/

Key steps:

1. Configure motors before assembly
2. Assemble leader and follower arms
3. Calibrate both arms (do NOT connect cameras during calibration)
4. Add cameras (wrist + front)
5. Record dataset episodes

On Thor, two USB cameras must be on different USB hub controllers. Use one USB-A port and one external USB-C hub.

[OK] when dataset is collected and saved locally.

Phase 4 — Train GR00T N1.5 on cloud (~varies)

Set up NVIDIA Brev (https://login.brev.nvidia.com/signin) or equivalent cloud GPU.

On the cloud server, install Conda and clone Isaac-GR00T:

mkdir -p ~/miniconda3 && cd ~/miniconda3
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh
bash miniconda.sh -b -u -p ~/miniconda3
source ~/miniconda3/bin/activate
conda init --all

git clone https://github.com/NVIDIA/Isaac-GR00T
cd Isaac-GR00T
conda create -n gr00t python=3.10
conda activate gr00t
pip install --upgrade setuptools
pip install -e .[base]
pip install --no-build-isolation flash-attn==2.7.1.post4

Fine-tune:

python scripts/gr00t_finetune.py \
   --dataset-path ./demo_data/so101-table-cleanup/ \
   --num-gpus 1 \
   --output-dir ./so101-checkpoints \
   --max-steps 10000 \
   --data-config so100_dualcam \
   --video-backend torchvision_av

Default fine-tuning requires ~25GB VRAM. Add --no-tune_diffusion_model to reduce VRAM usage.

Download the trained checkpoint from the cloud server.

[OK] when checkpoint files are available locally on Thor.

Phase 5 — Run GR00T N1.5 inference on Thor

Pull the reference Docker image (or use your own environment):

docker pull johnnync/isaac-gr00t:r38.2.arm64-sbsa-cu130-24.04

Start the container:

sudo docker run --rm -it \
  --network=host \
  -e NVIDIA_DRIVER_CAPABILITIES=compute,utility,video,graphics \
  --runtime nvidia \
  --privileged \
  -v /tmp/.X11-unix:/tmp/.X11-unix \
  -v /etc/X11:/etc/X11 \
  --device /dev/nvhost-vic \
  -v /dev:/dev \
  johnnync/isaac-gr00t:r38.2.arm64-sbsa-cu130-24.04

Inside the container, install GR00T:

git clone https://github.com/NVIDIA/Isaac-GR00T.git
cd Isaac-GR00T
pip install --upgrade setuptools
pip install -e .[thor]

Terminal 1 — Start inference server:

python scripts/inference_service.py --server \
    --model_path ./so101-checkpoints \
    --embodiment-tag new_embodiment \
    --data-config so100_dualcam \
    --denoising-steps 4

Terminal 2 — Start inference client (in same container via docker exec):

python examples/SO-100/eval_lerobot.py \
    --robot.type=so100_follower \
    --robot.port=/dev/ttyACM0 \
    --robot.id=my_awesome_follower_arm \
    --robot.cameras="{ wrist: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30}, front: {type: opencv, index_or_path: 2, width: 640, height: 480, fps: 30}}" \
    --policy_host=0.0.0.0 \
    --lang_instruction="Grab pens and place into pen holder."

Replace index_or_path with your camera indices (find with ls /dev/video*).

[OK] when the robot arm executes the learned behavior.

Failure decision tree

Reference files

• references/source.body.md — full original Seeed tutorial with Thor flashing screenshots, Docker image pip list, Brev platform walkthrough, and FAQ (reference only)