Espikify

Brew your PyTorch SNN models into ESP32-S3-ready C headers.

Features • Quick Start • CLI Usage • Generated Output • Examples • ESP32 Tests • Citation

✨ Features

• Seamless Transformation: Export trained PyTorch SNN attitude and control checkpoints into ESP32-S3-ready C/header files.
• ESP32-S3 Ready: Generate controller headers and runtime files for ESP32-S3 firmware projects.
• Different Runtime Backends: Choose between optimized and event-driven runtime modes.
• Flexible Model Inputs: Use direct .pt files, folders containing model.pt, or Weights & Biases run names.
• Controller Header Generation: Generate the main controller configuration header and controller component headers.
• Batch Processing: Convert the attitude and control networks with a single command.

🚀 Quick Start

Installation

git clone <repo-url>
cd espikify
python -m pip install -e .

Package requirements:

Requirement	Version
Python	>=3.10
numpy	>=1.24
torch	>=2.0
wandb	>=0.16

Check the CLI:

espikify --help

🧑🏻‍💻 CLI usage

espikify \
  --att-model ATT_MODEL \
  --control-model CONTROL_MODEL \
  --mode {optimized,event-driven} \
  [--output-dir DIR] \
  [--wandb-project PROJECT] \
  [--models-dir DIR] \
  [--att-mask CSV] \
  [--control-mask CSV] \
  [--no-copy-runtime]

✅ Generated output

An export writes:

<output-dir>/
├── test_controller_fwr_conf.h
├── controller/
│   └── test_controller_*_file.h
└── runtime/              # omitted with --no-copy-runtime

📝 Examples

Export in optimized mode

espikify \
  --att-model ./models/attitude/model.pt \
  --control-model ./models/control \
  --mode optimized \
  --output-dir ./build/optimized_export

Export in event-driven mode

espikify \
  --att-model ./models/attitude \
  --control-model ./models/control \
  --mode event-driven \
  --output-dir ./build/event_export

👷🏻 ESP32 test examples

The repository includes two Arduino sketches for testing generated controller exports:

Sketch	Use with
optimized-snn.ino	--mode optimized
snn-event-driven.ino	--mode event-driven

Use the sketch that matches the selected export mode. Both sketches load test_controller_fwr_conf.h, build a NetworkController, run forward_network, and stream controller output over serial. The event-driven sketch also uses the event-driven runtime path and reports spike-count statistics.

🎉 Acknowledgements

Espikify stands on the shoulders of giants. We are grateful to these projects and communities:

• TinySNN — MIT Licensed. This work inspired the project.
• Espressif Systems — for the ESP32-S3 platform and ecosystem.

Contributions and issue reports are welcome.

🙌 Citation

Rim El Filali, Chenrui Feng, Chao Gao, and Weibin Gu (2026). Spiking Neural Network Control of a Flapping-Wing Robot on Resource-Constrained Hardware. Proceedings of The 10th Conference on Robot Learning, in Proceedings of Machine Learning Research.