Why this exists Architecture Reference kit Roadmap Contribute
Open source · Apache 2.0 · Built on EmbedIQ

Animals sense
what we cannot.

An open-source, field-deployable sensor pipeline framework for animal-augmented environmental and biological sensing. Hardware reference designs, firmware, and ML models — fork it, deploy it, extend it.

View on GitHub Contribute
Built by Ritzy Lab· Sibling of SensorPipe· Powered by EmbedIQ· Apache 2.0
animalsensepipe — field deployment
$ animalsensepipe run --config deployment.yaml [L1] firmware: animal_pack OK · site_pack OK [L2] streams: imu · audio · gps · seismic · env (time-aligned) [L3] baseline: animal motion profile learned (T+72h) [L4] event: behavioral anomaly detected 2026-05-11 03:14 [L5] correlate: USGS M2.3 · 18 km · +47 min [ logged · not predicted ]

The platform layer for animal-augmented sensing is missing.

We mapped the landscape before writing a line of code. Three buckets — closed product companies, satellite tracking, lab-only behavior tools — and a gap in the middle that nobody is building.

🔒

Closed players race for proprietary disease screeners

Dognosis (patented dog-BCI cancer detection), SpotitEarly (trained-dog DTC screening), Owlstone (synthetic e-nose) — all closed, all vertical. None expose a substrate layer the rest of the field can build on.

May 2026 — verified competitive landscape
🛰

Academic tracking sees from space, not the ground

ICARUS at Max Planck tracks 100,000+ animals via satellite with GPS + accelerometer tags. Useful, but single-modality and altitude-limited. The ground-level multi-modal pipeline — chemical, acoustic, seismic, behavioral — does not exist.

ICARUS 2.0 launched Nov 2025 · complementary, not competitive
🧪

Open-source behavior tools live in the rodent lab

DeepLabCut, AlphaTracker, MARS, LabGym — all computer-vision pose estimation, all built for inside-the-lab neuroscience. None are field-deployable. None frame the animal as a sensor for external phenomena.

10+ open tools surveyed · no field-deployable multi-modal pipeline among them

Four layers, from sensor silicon to the model.

AnimalSensePipe is software and reference hardware that sits between raw multi-modal animal and environmental sensor data and a trained ML model. EmbedIQ on the device, Python on the host. Every layer is pluggable. Bring your own sensors, your own animals, your own model.

// LAYER 1

Firmware (on-device)

EmbedIQ-based sensor drivers, hardware time-sync across modalities, on-device calibration, ring-buffer storage, MQTT/HTTP egress with backpressure.

// LAYER 2

Ingestion service

Streams from animal and site packs land in a time-series database (InfluxDB or TimescaleDB), tagged by deployment session and animal ID.

// LAYER 3

Calibration & fusion

Python SDK with per-animal baseline learning, per-site environmental baseline, time-alignment, and a multi-modal dataframe builder.

// LAYER 4

ML interface

Plug-in model interface with one reference model — anomaly detection on multi-modal time series. Swap in your own model without touching the pipeline.

FIRST DEPLOYMENT — ANIMAL PACK + SITE PACK
$ animalsensepipe init --kit collar-v0
→ firmware flashed · time-sync verified  (animal_pack, site_pack)
$ animalsensepipe deploy --location home --animal max
→ streaming · imu · audio · seismic · env  (session_20260511.aspipe)

Off-the-shelf parts. Reproducible by anyone.

Every part is commodity, every supply is known, every line of firmware is open. No custom silicon, no exotic sensors. The hardware story is "EmbedIQ plus commodity parts" — exactly the showcase the community can fork.

Animal-worn pack

target unit cost · $40 – $70
  • MCU — ESP32-S3 (EmbedIQ-supported)
  • IMU — LSM6DSV 6-axis motion / orientation
  • Mic — ICS-43434 digital MEMS for vocalization & ambient
  • GPS — u-blox NEO-M9N (optional)
  • Power — 1000–2000 mAh LiPo + battery management IC
  • Connectivity — Wi-Fi (primary), BLE (config), LoRa (optional)
  • Enclosure — 3D-printed collar mount, IP54-rated

Site pack

target unit cost · $80 – $150
  • Gateway — ESP32-S3 or RPi Zero 2W
  • Seismic — SM-24 geophone + ADS1256 ADC
  • Environment — BME280 (P / T / humidity)
  • Magnetometer — MMC5983MA (EM-precursor research)
  • Audio — measurement-grade MEMS mic
  • Video — UVC or RPi Camera (optional)
  • Power — wall + battery backup, optional solar

Five ways to plug in.

Whether you build PCBs, write firmware, train models, run a deployment, or want to help without writing code yet — there's a clear entry point and a "good first issue" tag for each.

Build hardware

You like PCBs, 3D printing, sensor selection. Help refine the BOM, prototype the enclosure, test off-the-shelf alternatives, document build photos.

good-first-issue · hardware

Write firmware

You work in embedded C / Rust. Help build sensor drivers on EmbedIQ, hardware time-sync, ring-buffer storage, MQTT egress. The EmbedIQ docs are public.

good-first-issue · firmware

Train models

You're an ML student or researcher. Improve the anomaly-detection reference model, add new fusion primitives, contribute notebooks for new use cases.

good-first-issue · ml

Run a deployment

You have access to a pet, farm, or field site. Deploy a kit, collect data, publish your dataset. Your real-world deployment is the most valuable contribution.

good-first-issue · field

Document & translate

You want to help but aren't writing code yet. Write build tutorials, translate the docs, improve the website copy, run the GitHub Discussions community. This is the fastest way to make the project usable for people in other languages and disciplines.

good-first-issue · docs

Now · Next · Later.

Honest about scope. Honest about what's funded. Honest about what's exploratory. No dates — direction.

// now

MVP-1 — Behavior pipeline

  • Apache 2.0 GitHub repo
  • Animal pack + site pack firmware on EmbedIQ
  • Python SDK and calibration pipeline
  • Anomaly-detection reference notebook
  • One named academic partner
  • Reference experiment + write-up
// next

MVP-2 — Chemical & physiological

  • VOC array (BME688 / SGP41)
  • PPG heart-rate / body-temp wearable
  • Optional EEG header (unpopulated)
  • Second reference application — conservation or veterinary
  • Community-contributed deployments
// later

V3+ — Generalize & scale

  • Multi-animal architecture (insect cohorts)
  • Hosted analytics tier (optional managed cloud)
  • Pre-trained model zoo
  • Cross-deployment dataset corpus

A research enabler, not a forecasting product.

This part is up-front, not buried at the bottom. The science around animal-based prediction is genuinely contested. We don't make claims the science doesn't support.

// what we are

AnimalSensePipe is the open infrastructure that lets researchers, conservationists, and curious builders study animal-augmented sensing rigorously — with shared hardware, shared firmware, shared data formats, and reproducible experiments.

We are not a disease-screening product. We are not promising disaster prediction. We are not competing with closed clinical players. We are not claiming animals predict earthquakes.

We claim only that the platform works — and that if the open questions in this field are ever going to be answered, the answers will come from shared, open, rigorous tooling. That's what this is.

Built by the team behind EmbedIQ and SensorPipe.

AnimalSensePipe is a Ritzy Lab open-source project. It is the sibling of SensorPipe — Ritzy Lab's industrial sensor-pipeline framework — and runs on EmbedIQ, Ritzy Lab's AI-first open-source firmware and gateway framework. Same architectural DNA, same Apache 2.0 license, same first-principles approach: open format, open hardware, observable by default.

Where SensorPipe demonstrates EmbedIQ's industrial relevance, AnimalSensePipe demonstrates its scientific and humanitarian breadth — a framework that scales from factories to wildlife. Both projects ship with reference hardware. Both are designed to be forked.

Animal welfare and ethical framing are central. Reference deployments are non-invasive (collar-worn, site-fixed) and follow established field-research protocols. The optional EEG header in v2 is a header, not a populated component — to leave room for future research while keeping the v0 kit clear of any invasive-BCI patent territory.

If this is the platform you've been waiting for — fork it.

The repo is public. The license is Apache 2.0. The first deployment costs about $120 in parts. If you want to talk before you build — about partnerships, deployments, or where this fits your research — write us.

View on GitHub [email protected]