Questions about using ESP32-C6 and a single antenna for gesture recognition (CSI-based ML)

[01BreadGuy10]

Hello, I'm a learner currently researching CSI (Channel State Information). I would like to use an ESP32-C6 chip and one antenna to perform machine learning for gesture recognition. Specifically, I aim to recognize three pairs of gestures: push/pull, left/right, and up/down.

My questions are:

Can my existing ESP32-C6 board and a single antenna meet the requirements for this task?

Can I use the code in this repository to perform training for these gestures?

If the training is successful, how can I upload the dataset I recorded during training to this repository?

Thank you for your help!

[francescopace]

Hello @01BreadGuy10, I’m going to convert this issue into a GitHub Discussion (Q&A) so the conversation can stay open and easier to find for other users with similar setup/training questions.

Quick pointers in the meantime:

• Support and setup: https://github.com/francescopace/espectre/blob/main/SETUP.md
• ML data collection guide: https://github.com/francescopace/espectre/blob/main/micro-espectre/ML_DATA_COLLECTION.md
• Data contribution workflow: https://github.com/francescopace/espectre/blob/main/CONTRIBUTING.md#data-contributions

Note: current ML training pipeline is focused on baseline vs movement; gesture recognition is planned for 3.x.

Continuing in Discussion 👇

[francescopace]

@01BreadGuy10 If your goal is CSI-based gesture recognition, please start from the branch:

• https://github.com/francescopace/espectre/tree/feature/gesture_detection

That branch already contains the experimental end-to-end gesture pipeline (Python + C++):

• Gesture training pipeline: https://github.com/francescopace/espectre/blob/feature/gesture_detection/micro-espectre/tools/11_train_gesture_model.py
• Gesture streaming benchmark (offline + live): https://github.com/francescopace/espectre/blob/feature/gesture_detection/micro-espectre/tools/13_test_gesture_stream.py
• Runtime gesture detector (MicroPython): https://github.com/francescopace/espectre/blob/feature/gesture_detection/micro-espectre/src/gesture_detector.py
• Runtime gesture detector (ESPHome/C++): https://github.com/francescopace/espectre/blob/feature/gesture_detection/components/espectre/gesture_detector.cpp
• Tooling overview: https://github.com/francescopace/espectre/blob/feature/gesture_detection/micro-espectre/tools/README.md

For ESPHome, that branch also adds:

• gesture_detection_enabled
• gesture_sensor (text sensor output)

Current exported model on this branch is still experimental and currently trained on classes like:

• wave, circle_cw, no_gesture

The training script is already structured to expand to more labels (including push, pull, swipe_left, swipe_right, etc.) as soon as enough labeled data is collected.

So yes: for your use case (ESP32-C6 + single antenna + gesture ML), this is the right branch to start from.