Chapter 2

OpenArm platform

OpenArm is built for contact-rich manipulation and learning-ready data: reproducible hardware, clear interfaces, and tight integration with modern robot-learning stacks.

OpenArm manipulator and data-centric control

OpenArm resources View in store

Learning outcomes

Contrast teaching arms with a logging-first manipulation platform.
Describe CAN setup concerns (termination, IDs) that affect dataset validity.
Plan calibration and tool-change checks before policy experiments.

Learn

Data loop, drives, calibration, and SVRC ops context.

Practice

Sketch left/right URDF mapping; list one episode field you will log.

Challenge

Outline a minimal eval harness; ask for feedback on the Forum.

Facilitation: Split research vs classroom groups — research teams read learning-ready data first.

Self-check

When move from SO-101 to OpenArm?

When you need sustained logging, CAN bandwidth, and evaluation throughput beyond teach-repeat demos.

What must stay consistent across arms?

Kinematic naming, handedness, and time base for multi-modal recordings.

STEM alignment: systems integration, testable interfaces, responsible iteration on hardware.

1. Why a separate “platform” chapter

OpenArm sits at the intersection of how SVRC thinks about the data loop and real hardware you can source in Palo Alto. Unlike small teaching arms, the emphasis is on sustained logging, calibration stability, and throughput suitable for imitation learning and evaluation.

2. CAN, drives, and left/right arms

Production arms typically expose multi-axis drives over CAN. You will configure the interface (bitrate, termination), assign controller IDs, and map left/right arms consistently to your URDF or MJCF so policies and datasets stay unambiguous.

3. Calibration & tool changes

Run vendor calibration after mechanical changes. If you swap end-effectors or add mass, update payload estimates or conservatively reduce speed/acceleration until re-identified.

4. Software ecosystem

Teams often bridge OpenArm to recording tools (e.g. LeRobot-style datasets), ROS 2 for integration, or HTTP/WebSocket bridges for quick UI prototypes — see Communication & architecture. Pair with what makes data learning-ready when designing episodes.

5. Operations at SVRC

Same-day pickup on core SKUs, local support, and alignment with our Data Services when you want SVRC to operate collection for you. Specs and pricing are always confirmed on the store and product pages.

← Dexterous hands · Next: Unitree G1 →