An Open-Source Robot Arm for Human Manipulation Data Collection
Abstract
As technology evolves, robots designated for niche tasks are becoming obsolete, and success relies on robots that are capable of adapting to human environments — resulting in a surge in development of humanoid robotics. However, current robotic arm solutions present a challenge: high-performance systems are often prohibitively expensive or proprietary, while accessible options lack the necessary technical capabilities for advanced research. To address this, OpenArm was created as an open-source platform, providing a robust and affordable entry point into humanoid robotics. It’s scalable and modifiable design supports various research applications, including teleoperation, reinforcement learning, and imitation learning, with dedicated software and simulation examples in Gazebo, MoveIt, and Genesis. The Reazon HI Lab team aims to foster community-driven development and incorporate user feedback to advance humanoid robotics research through OpenArm's open-source nature.
Project Overview
OpenArm is a fully open-source humanoid robot arm designed to support AI/ML researchers, humanoid developers, and enthusiasts looking for an affordable, capable system.
Hardware Design:
OpenArm’s hardware is designed to maximize affordability and capability, there are several key features that directly contribute to this criteria:
- Manufacturing: Uses sheet metal & CNC machining for durability, overcoming the limitations of 3D-printed components — entirely self-purchasable online.
- Modular & Scalable: Users can modify parts to fit different motors, end effectors, and additional components via provided STEP files.
- High-precision DAMIAO BLDC motors with integrated planetary gearboxes to reduce backlash and optimize control.
- Currently supports CAN communication for motor control at speeds up to 1Mbps.
- Capable of control frequency of 1kHz with CANFD integration (planned for future versions of OpenArm)
Software & Simulation:
To support research and development efforts, OpenArm is designed to work with various simulation software and provide the necessary software tools for rapid testing and iteration:
- Fully integrated with ROS2
- Unilateral and bilateral force feedback implementation.
- Supports physics simulation in:
- Genesis – A highly optimized physics engine offering high-speed performance with user-friendly functionality.
- MuJoCo – Designed for physics-based reinforcement learning.
- MoveIt – Built for motion planning applications.
To learn more about the features built into OpenArm, or to discover some examples of applications and studies that have been conducted, please take a look at the GitHub.
What’s Next?
The HI Lab team is committed to continuously improving OpenArm and integrating more features to support the growing base of researchers.
Upcoming Changes
5 Finger End Effector
- Examples of OpenArm currently use a 1 degree of freedom prismatic end effector, limiting the humanoid resemblance.
- In the future, a 5 fingered end effector is planning to be adapted and integrated with OpenArm with corresponding software control and simulation examples.
Leader Control Arm for Bilateral Feedback
- OpenArm currently can be configured for use in teleoperation by setting up an additional robot arm, use tracking hardware for input, or be mapped with a third party controller.
- In the future, a dedicated leader arm is being designed to act as a control arm allowing for greater support into bilateral feedback research endeavors.
CANFD Support
- OpenArm currently operates using CAN communication at 1Mbps.
- In the future, it will feature full CAN FD support, increasing maximum bandwidth to up to 5 Mbps allowing for a control frequency of 1kHz
Stay tuned for more features and items to be implemented beyond the near future!
Contact
Have a question or want to collaborate? Reach out to the Reazon HI Lab team!
Email: hi_public@reazon.jp
