OpenArm 101
The most complete OpenArm resource on the internet. From unboxing to recording your first AI training dataset — every step is here. 8 pages covering hardware setup, software, data collection, tutorials, safety, and community.
Your Setup Journey
Follow these steps after unboxing your OpenArm. Takes about 6–7 hours total. New to OpenArm? Read the Quickstart first →
Unboxing & Safety Check
Verify contents, inspect joints, set up workspace
Software Environment
Install Ubuntu dependencies, SocketCAN, ROS2
First Motion
Run fake hardware mode, then real hardware
Calibration & Homing
Set joint zeros, verify workspace
Teleoperation
Connect operator device, record first session
Data Collection
RLDS/LeRobot format, quality checks
AI Model Training
Fine-tune ACT/Diffusion Policy, deploy
Hardware at a Glance
Compatible AI Models
OpenArm 101 is purpose-built for policy learning. These models work out of the box with your recorded data.
ACT
Action Chunking Transformer — best for pick-and-place tasks. Predicts action sequences from camera observations.
View model →Diffusion Policy
Best for contact-rich manipulation. Generates smooth trajectories using denoising diffusion over action space.
View model →OpenVLA
Best for language-conditioned tasks. Combines vision-language understanding with robot action prediction.
View model →Technical Guides
Detailed research articles covering every layer of the OpenArm stack.