Openarm
Overview
OpenArm is a fully open-source, 7-degree-of-freedom (7-DOF) humanoid robotic arm platform designed for physical AI research, contact-rich manipulation, and real-world data-driven robotics.
Originally designed by Enactic and officially supported by CEREBOTO as a manufacturing partner, OpenArm exposes its entire hardware and software stack, enabling researchers and developers to build, modify, deploy, and scale real robotic systems without proprietary constraints.
OpenArm is purpose-built for teams working at the intersection of hardware, control, learning, and data, where access to the full system—mechanical, electrical, and software—is essential.
Design Philosophy
OpenArm is designed around four core principles:
Accessibility
Delivering research-grade humanoid manipulation capabilities at a fraction of the cost of traditional humanoid robotic systems.
Openness
Complete open access to CAD files, firmware, control software, simulation assets, and manufacturing data—no black boxes.
Practicality
Optimized for real-world tasks, repeated experimentation, and long-term deployment in research and development environments.
Safety
Human-centered design with compliance, backdrivability, mechanical limits, and controlled interaction as first-class considerations.
The long-term vision of OpenArm is to lower the barrier to advanced robotics research, enabling more teams to work with real robots—not just simulations.
Mechanical Architecture
Human-Scale Proportions
OpenArm adopts anthropomorphic proportions comparable to a 160–165 cm human arm, offering:
A practical workspace and reach (633 mm)
Controlled inertia for safe interaction
Natural motion patterns suitable for manipulation research
Stable dynamics for contact-rich tasks
Each arm weighs approximately 5.5 kg, balancing rigidity and compliance.
Degrees of Freedom & Structure
7 DOF per arm, enabling human-like dexterity
Modular aluminum structural frame with stainless steel base plate
Uniform M6 mounting grid for flexible installation
Secure desktop or fixture mounting
Every joint is equipped with mechanical hard stops to prevent overextension.
Actuation & Load Capacity
Motor System
OpenArm uses DAMIAO-series integrated joint motors, selected per joint based on torque, speed, and compliance requirements:
Shoulder joints: DM-J8009P-2EC (high torque, load-bearing)
Elbow joints: DM4340 series (balanced torque and compactness)
Wrist & gripper joints: DM-J4310-2EC (high-speed response and precision)
The design combines QDD and high-performance geared joints to strike a balance between backdrivability, rigidity, and compact form factor.
Payload Definitions
Rated payload: 4.1 kg
Sustained load at full extension for 1 minute
Peak payload: 6.0 kg
Dynamic lift-and-hold test over a 3-second motion
⚠️ Payload values include the end effector mass.
For example, with a 1.5 kg gripper installed, usable payload is reduced accordingly.
End Effector & Customization
Gripper System
Maximum opening width: 88 mm
Direct-drive motor actuation
Smooth linear guidance using bearings and sliders
Suitable for force feedback and bidirectional control
The gripper is designed for precise grasping, contact interaction, and learning-based manipulation.
Custom End Effectors
OpenArm supports easy replacement of the end-effector interface component (J8_B), allowing:
Custom grippers
Sensorized tools
Experimental end-effector designs
Design guidelines emphasize mechanical alignment, cable routing, and manufacturability (3D printing or CNC).
Electrical & Control Interface
Control bus: CAN / CAN-FD
Control frequency: up to 1 kHz
Operating voltage: 24V DC (select joints support up to 48V)
The high-frequency control loop enables low-latency torque, velocity, and position control, critical for learning-based and force-controlled applications.
Software Ecosystem
OpenArm provides a complete software toolchain connecting hardware to modern robotics and AI workflows.
Robot Description
Modular XACRO-based design
URDF generation for simulation and control
Compatible with major robotics frameworks
Low-Level Control API
Direct joint-level access
Position, velocity, and torque control
Real-time response at up to 1 kHz
ROS2 Integration
Native ROS2 nodes and message interfaces
RViz visualization support
Compatibility with MoveIt and motion planners
Simulation Support
OpenArm is fully supported in:
MuJoCo — ideal for reinforcement learning and contact modeling
Isaac Sim — high-fidelity physics and rendering
Simulation assets are calibrated to match physical dynamics, supporting sim-to-real transfer and large-scale parallel training.
Control & Teleoperation
Real-time gravity compensation
Impedance and force control
Bilateral teleoperation with force feedback
Smooth, human-like operation for demonstration collection
Data-Centric Design
OpenArm is designed not just as a robot, but as a data generation platform:
Structured recording of real-world interactions
Support for imitation learning datasets
Repeatable, high-quality human demonstrations
Alignment between simulation data and real-world trajectories
This makes OpenArm particularly suitable for physical AI, embodied intelligence, and robotics dataset creation.
Application Domains
Research
Imitation learning and skill acquisition
Reinforcement learning and sim-to-real validation
Human–robot interaction studies
Contact-rich manipulation
Education
University robotics labs
Embodied AI training programs
Robotics competitions and coursework
Industrial Exploration
Light assembly tasks
Vision-guided inspection
Human–robot collaboration research
Safety Considerations
OpenArm includes:
Mechanical joint limits
Backdrivable joint design
Emergency stop support (hardware and software)
Clear installation and operational guidelines
It is intended for controlled environments operated by trained users.
Availability & Manufacturing
OpenArm is available as:
DIY Kit — full component set for self-assembly
Fully Assembled System — calibrated and ready for deployment
Open and Evolving
OpenArm is an active, evolving project.
Contributions from researchers, developers, and industry partners are encouraged.
The goal is simple but ambitious:
Make high-quality humanoid robotic manipulation accessible, modifiable, and real.