DK1 Bimanual: From Setup to Two-Handed Robot Learning
A structured, sequential path that takes you from hardware setup to running a trained bimanual imitation learning policy on both DK1 arms simultaneously. Designed for learners who have completed the OpenArm path or have equivalent single-arm robotics experience.
Before You Start — Check These Prerequisites
- Completed the OpenArm 101 path or have single-arm teleoperation and data collection experience
- Comfortable with ROS 2 basics (launch files, topic inspection, joint state monitoring)
- Python 3.10 or higher installed; familiar with virtual environments and pip
- Access to a DK1 kit (both follower arms + leader controller) or the MuJoCo bimanual sim
- Ubuntu 22.04 or 24.04 — macOS and Windows are not supported for hardware operation
- About 16 hours of total time, ideally spread across 4–5 sessions
Not sure if you're ready? Read Unit 0 — it maps the DK1 hardware to what you already know from OpenArm and tells you exactly what's different.
Your Path at a Glance
Complete the units in order. Each unit has a clear completion check — bimanual systems have tighter coupling between units, so don't skip ahead.
Before You Begin: Bimanual Orientation Required
Why two arms? Hardware and software checklists. Leader/follower concept explained. Time estimates and where to get help.
~30 min · No codeMount and Wire Two Arms Required
Physical mounting for a bimanual workspace, leader and follower arm setup, camera placement for full workspace coverage.
~3 hoursConfigure Leader/Follower Control Required
SDK install for bimanual, pairing config, Python connection test for both arms, sync verification.
~2 hoursTeleoperate Your First Bimanual Task Required
Understand bimanual teleoperation, calibrate workspace for two arms, practice a cube handoff, and reach the 5-minute continuous milestone.
~2 hoursRecord Synchronized Bimanual Demonstrations Required
Why bimanual data collection is harder. LeRobot bimanual dataset format. Recording workflow and quality checklist. Target: 100 clean demos.
~3 hoursTrain a Bimanual Policy with ACT Required
Why ACT excels at bimanual coordination. Training command, reading bimanual training curves, hyperparameters. Target: >60% success.
~4 hoursRun and Improve Your Bimanual Policy Required
Inference setup for two arms, bimanual evaluation protocol, failure modes unique to bimanual, and the data flywheel for improvement.
~2 hoursTime Breakdown
| Unit | Topic | Time |
|---|---|---|
| Unit 0 | Bimanual Orientation | ~30 min |
| Unit 1 | Hardware Setup | ~3 h |
| Unit 2 | Leader/Follower Config | ~2 h |
| Unit 3 | First Bimanual Teleoperation | ~2 h |
| Unit 4 | Data Collection (100 demos) | ~3 h |
| Unit 5 | ACT Policy Training | ~4 h |
| Unit 6 | Deploy & Improve | ~2 h |
| Total | ~16 h 30 min | |
Takes about 30 minutes. Covers the bimanual-specific concepts and checklists before any hardware work begins.