Unit 0 — Orientation

DK1 Quickstart: Before You Begin

Read this first. This page tells you exactly what hardware you need, what software to install, how long setup takes, and what you will be able to do when you finish the full bimanual path.

Time Commitment

How Long Does This Take?

From box to bimanual data collection. The DK1's two-arm setup takes longer than a single-arm kit — plan accordingly.

Unboxing & assembly

~45m

Verify contents, inspect servos on both arms, mount tabletop hardware

Software install

~30m

Clone trlc-dk1 repo, install as LeRobot plugin with uv

Port detection

~20m

Run lerobot-find-port, assign ttyACM* for both arms

Calibration

~20m

Calibrate leader arm then follower arm separately

First teleoperation

~30m

Run lerobot-teleoperate, verify leader-follower tracking

First dataset

~1h

Bimanual recording, quality review, upload to HuggingFace

Total first-day time: roughly 3–4 hours for a complete bimanual first session. Comfortable to spread across two days for beginners.

What You Need

Hardware Checklist

Everything that needs to be on your desk before you start. Items marked "in the box" ship with the DK1 kit.

✓

Two DK1 arm units (leader + follower) In the box — follower arm: 7 DOF + gripper (DM4340 + DM4310 servos). Leader arm: 7 DOF + gripper (Dynamixel XL330 servos). Both arms must be present.
✓

Two USB-C cables (one per arm) In the box — connects each arm to the host PC's USB ports. Use separate USB controllers if possible to avoid bandwidth contention.
✓

Power supply for follower arm In the box — the follower arm (DM4340 motors) requires a dedicated power brick. The leader arm is bus-powered over USB.
✓

Tabletop mounting hardware In the box — M5 bolts, base plates for both arms. Stable mounting is critical: the follower arm produces torques that will shift an unsecured arm during teleop.
✓

Host PC running Ubuntu 22.04 Not included — native Ubuntu install required. Minimum: 8 GB RAM, 4 USB 3.0 ports (2 arms + 2 cameras). Download Ubuntu →
✓

Two cameras (wrist-view + overhead) Not included but required for data collection — USB webcam or RealSense D435i. Mount one on each follower arm end-effector, one overhead for workspace view.
✓

1.5 m × 1 m clear workspace Required — the bimanual setup requires more table space than a single arm. Both arms should reach the shared workspace center. See the Safety page →

⚠

Bimanual-specific note: The DK1 uses a leader/follower configuration. The leader arm (Dynamixel XL330) is moved by the operator's hand. The follower arm (DM4340) replicates those movements in real time. Both arms must be calibrated and connected before starting any recording session. Never power on just one arm.

Software Prerequisites

What to Install Before You Start

The Setup Guide walks through each step in detail. This is the summary so you can prepare in advance.

Operating System

Ubuntu 22.04 LTS (Jammy). Ubuntu 20.04 also works but 22.04 is recommended. macOS and Windows are not supported for real hardware.

Python

Python 3.10+ (bundled with Ubuntu 22.04). The DK1 plugin targets Python 3.10+. Install pip and uv:

sudo apt update
sudo apt install python3-pip python3-venv -y
pip install uv

ROS2 Humble

ROS2 Humble Hawksbill is the recommended distribution. The DK1 supports ROS2 for higher-level bimanual control (optional — not required for the LeRobot-only data collection path):

sudo apt install ros-humble-desktop ros-humble-ros2-control \
  ros-humble-ros2-controllers -y

LeRobot

LeRobot (HuggingFace) is the primary framework for DK1 data collection. Install via pip:

pip install lerobot

DK1 LeRobot Plugin

The TRLC-DK1 repository provides a LeRobot plugin that registers the dk1_leader, dk1_follower, bi_dk1_leader, and bi_dk1_follower device types:

git clone https://github.com/TRLC-AI/trlc-dk1.git
cd trlc-dk1
uv pip install -e .

Simulation (optional — no hardware needed)

✓ Simulation available

You can test the DK1 software stack without hardware using MuJoCo with the TRLC-DK1 bimanual model:

pip install mujoco
# Clone the DK1 MuJoCo model
git clone https://github.com/TRLC-AI/trlc-dk1-mujoco.git
python trlc-dk1-mujoco/examples/bimanual_sim.py

The MuJoCo model includes both leader and follower arm physics, correct workspace geometry, and gripper dynamics. See the Software → Simulation section for full setup.

End Goal

What You Can Do After the Full Path

After completing all 5 steps — unboxing through first bimanual dataset — you will be able to:

✓ Teleop both arms simultaneously using the leader arm for natural, human-demonstration-quality bimanual control

✓ Record synchronized bimanual episodes: 14 joint states, 2 grippers, multi-camera video, all timestamped

✓ Export datasets in LeRobot / HuggingFace format with bi_dk1_leader and bi_dk1_follower device schema

✓ Train ACT (Action Chunking Transformer) on bimanual pick-and-place tasks with your recorded data

✓ Run bimanual policy inference on the DK1 follower arm pair with a trained model

✓ Validate policies in MuJoCo bimanual simulation before deploying to real hardware

Support

How to Get Help

You won't be going it alone. Here is where to ask questions at each stage.

💬

Ready? Start the Setup Guide.

Once you have both arms and Ubuntu installed, the Setup Guide walks you through every step.

Step 1: Unboxing & Setup → Read Safety First DK1 Forum