The idea
The principle of a myoelectric prosthesis. When a muscle contracts, it gives off a small electrical signal that sensors on the skin can read. A model turns that signal into an intention, open, close, how hard, and the gripper of a robot arm follows. In a second phase, motion sensors on the arm and forearm drive the shoulder and the elbow.
Why this one
It is the field I want to work in: neurotechnology, signals from the body turned into control. Prostheses, exoskeletons, human-machine interfaces. This project is my first real step there.
Who does what
We are two. Raphaël takes the robotics and the hardware: the gripper, the mechanics, turning commands into motor movements. I own the whole software chain: acquisition, filtering, decoding, real-time command, safety, and a dashboard that replays recorded sessions.
How it is built
- A wireless band made by the team: four MyoWare 2.0 EMG sensors on an ESP32-S3, streaming over Wi-Fi to a small PC under Ubuntu.
- That PC runs the ONNX model and a ROS node that drives an AgileX PiPER arm over its CAN bus.
- A simple baseline first, LDA or SVM on hand-made features. Deep learning only if it does better.
- Speed limits and a mandatory emergency stop. A short calibration at each session, because EMG drifts from day to day.
Calendar
- October 2026: scoping done, sensors and model chosen together.
- End of November: stable band, clean signal.
- End of January: gripper and first model.
- End of February: real-time control.
- March: tests on the real arm. Mid-April 2027: defence.