Pavlov Edge is the sensing layer for the Smart Police Kennel. ESP32 nodes read temperature, air quality and motion off the floor, mesh them over ESP-NOW to a TLS bridge, and a Raspberry Pi running AWS Greengrass lands every reading in the cloud, at sub-second latency.
Every sensor reading travels the same route. Connectionless on the floor, mutually authenticated on the way out.
The signals that actually change a handler's decision, and the plumbing that keeps them flowing.
Temperature and humidity off the kennel floor every few seconds, with per-sensor health counters that survive reboots.
Regulator-grade ammonia (NH₃) and CO₂ readings. Ammonia is the controlled risk in enclosed housing; CO₂ is the ventilation proxy.
Dual PIR coverage flags activity in the run, decoupled from the cloud cadence so on-floor response stays instant.
An on-device camera pipeline runs object and pose detection in the browser. It counts visitors and watches the run, with no frame leaving the edge.
A graduated recovery watchdog catches silent "ghost online" failures. It bounces MQTT, then WiFi, then restarts, before a single reading is lost.
Mutual TLS to the cloud, X.509 certs in Secrets Manager behind a KMS key, and zero credentials in source. All infra is Python CDK.
Purpose-built nodes for the kennel: the environment, the feeding station, and the dog itself.
node-a
The environmental workhorse. It reads the kennel's air and movement several times a second, then meshes every sample to the bridge over ESP-NOW.
esp32-cam
A motion-triggered camera watching the run. It streams MJPEG over WiFi and serves JPEG snapshots on demand for the handler dashboard.
food-cart
A load-cell scale under the feeding station. It turns bowl weight into a welfare signal, tracking how much each dog actually eats and drinks.
collar · polar-h10
Sensing the dog itself. A XIAO-based collar pairs with a Polar H10 chest strap to stream heart rate, activity and temperature for the K9.
The five operators behind Pavlov Edge, from sensor floor to cloud.
Quiet when conditions are good. Loud, fast and verifiable the moment they aren't.