01 · Overview

Previous: None — this is the first page Next: 02-Environment-Setup

This page spells it out: what the complete XLeRobot workflow is, what hardware you need, and whether your machine is supported.

---

Workflow

The complete XLeRobot process has four steps:

① Calibrate robot  →  ② Teleoperate and record  →  ③ Train model  →  ④ Deploy for autonomy
   calibrate            record                       train             eval / rollout

The Docker image has the software environment for all four steps already set up — Python, PyTorch, LeRobot, the XLeRobot robot code, and the servo and camera drivers. You no longer have to fight version conflicts.

The only things you need to provide yourself: a robot (or get the software working first), a computer, and one USB cable.

---

Hardware support

Platform	Support
Jetson Orin series (JetPack 7.2)	✅ Currently the officially supported platform, with GPU acceleration
Jetson Nano (2019 original)	❌ Not supported — too little compute; see below
Raspberry Pi / regular Ubuntu / x86 PC	⏳ Not supported yet — see below
Windows / macOS	❌ Not supported (serial and camera passthrough only works smoothly on Linux)

💡 This image currently targets Jetson Orin devices only.

• Jetson Orin Nano / Orin NX / AGX Orin: requires JetPack 7.2 (L4T 39.2 / CUDA 13.2). AGX Orin 64GB is only fully supported on 7.2 as well.
• Jetson Nano (2019 original): ❌ never. It's Maxwell architecture (compute capability 5.3), and CUDA 13 dropped Maxwell support entirely, while torch requires at least sm_80.
• Raspberry Pi / x86 PC: supported by the image's design (the generic CPU edition is based on python:3.12-slim, which upstream ships in both amd64 and arm64 versions), but the published CPU edition isn't multi-arch yet — a pull may get you amd64-only or arm64-only. That opens up once the multi-arch build is done.

4 GB of memory or more is recommended. The Jetson Orin Nano's 3.4 GB can run teleoperation, but training will struggle.

System requirements

• Ubuntu 20.04 / 22.04 / 24.04, or Debian 12+
• Raspberry Pi: requires a 64-bit OS (uname -m shows aarch64)
• Kernel support for USB serial (present on virtually all systems)

Check your architecture

uname -m

• aarch64 → Raspberry Pi, Jetson (use the arm64 image)
• x86_64 → regular PC (use the amd64 image)

---

The folders you need to read

#	Folder	Contents	When to read it
1	deployment/ (right here)	Pull the image, run the image	First time — read 01 → 02 → 03 in order
2	xlerobot/	Running the whole XLeRobot	When you have the whole robot (dual arms + base)
3	lerobot/	Running the original LeRobot	When you only have a single SO-101 / SO-100 arm or a LeKiwi
4	troubleshooting/	Common problems and quick reference	When something goes wrong

⚠️ Folders 2 and 3 are alternatives — pick one based on your hardware:

• Whole XLeRobot (dual arms + base) → go to xlerobot/
• A single SO-101 / SO-100 arm, LeKiwi → go to lerobot/

The image is one shared environment — you can switch between the two paths at any time without reinstalling anything.

---

Shortest path (just get it running)

On a Jetson (JetPack 7.2), a few commands to see the robot move first — then come back for the details:

# 1. Install Docker (skip if it's already installed)
curl -fsSL https://get.docker.com | sudo sh && sudo usermod -aG docker $USER

# 2. Give the container GPU access (Jetson-specific, only needs doing once)
sudo nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml --disable-hook enable-cuda-compat
sudo systemctl restart docker

# 3. Pull the image (~3.6 GB)
docker pull chenyang310/xlerobot:jetson

# 4. Start the container
docker run -it --rm --device nvidia.com/gpu=all \
  --privileged --network host -v /dev:/dev \
  -v ~/.cache/huggingface:/home/jetson/.cache/huggingface \
  -e HF_ENDPOINT=https://hf-mirror.com \
  chenyang310/xlerobot:jetson

# 5. Run teleoperation in the container (the first run enters the calibration guide automatically)
python examples/4_xlerobot_teleop_keyboard.py

⚠️ The quick path skips two things you'll have to do sooner or later. If you get stuck, go back to the matching section:

• Registry mirrors for mainland China (step 3 may hang) → 02-Environment-Setup §2
• --disable-hook enable-cuda-compat can't be omitted (the hook is buggy on JetPack 7.2 and stops the container from starting) → 03-Deployment §4

Step 5 above runs the whole XLeRobot example. If you only have a single SO-101 arm, use the lerobot-* command line instead — see lerobot/.

---

Next: 02-Environment-Setup — install Docker, set up China-region mirrors, connect the hardware