We will explore the construction of the Robotics Collective’s Aachen Open-Source Humanoid and its machine learning pipeline that enable it to perform precise object manipulation, using state-of-the-art machine learning frameworks for robotics.
Building and deploying AI systems for robotic tasks involves complex engineering, large datasets, and intensive training procedures. This complexity makes it difficult for individual researchers or small teams to experiment with robotics in a meaningful way. The objective of this project is to addresses the high barrier to entry in robotics. We aims to lower these barriers using open source projects like LeRobot and So100 arms. The https://github.com/huggingface/lerobot platform by Hugging Face provides a suite of models, datasets, and tools built on the PyTorch framework for imitation and reinforcement learning.
Fig 1: Aachen Open-Source Humanoid (we forgot to add legs)
This article covers:
The following table lists the components used in the construction of the robot.
| Item | Specification / Link | Quantity |
|---|---|---|
| Hose Clamp | 50–70 mm Link | 3 |
| Flathead Screws | M4 x 40 mm Link | 20 |
| Self Tapping Screws | M4 x 40 | 20 |
| Hex Nuts | M4 Link | 20 |
| Pipe | DN40 x 250 mm Link | 2 |
| Pipe | DN40 x 500 mm | 1 |
| So100 Arms | Approximate quantity | 2 |
| 3-Way Connector | DN40 | 1 |
| 3D Printed Parts | Custom, unspecified | Several |
| Wooden Plank | Size unspecified | 1 |
The robot itself is primarily made of two SO-100 arms, which is fully open-source:
GitHub - TheRobotStudio/SO-ARM100: Standard Open Arm 100
SO100: A five axis arm with a gripper attached. The joint motors are Feetech STS3215 servos with magnetic encoders. The motor control board handles the communication with the motors. Feetech SDK offers python bindings to read and write joint angles to individual servos. The motors have built in PID control loop.
The structural members are built using DN40 pipes along, connectors and some 3D printed parts (connector and base).
STL files: