Garage state estimation in an uncertain world: sonar, ESP32, and HMMs

Sat, 21 Mar 2026 00:00:00 -0400

Introduction

I built a device I call Garage133 (see my earlier post ) to control our garage doors and monitor the state of the garage. For each side, a sonar sensor determines whether the garage door is open or closed, and whether a car is present. This allows our home automation system to alert us if a door is left open and provides the capability to close it remotely.

The image below shows the estimated garage state as rendered in our live Home Assistant¹ dashboard using the FloorPlan² card. The rendering includes door status, car presence,³ temperature, humidity, motion, charging, etc…

Home Assitant/Floorplan render of garage state

Garage133 basically works well. All we really need is an easy way to open and close the garage doors without overpaying for a closed, cloud-based system like MyQ⁴, and we have that. But the state detection logic proved surprisingly difficult to get right. This post explores why this is—the nature of sonar—and how we solved it using Hidden Markov Models.

Hidden Markov Model on Selective Appeal Blog

Garage state estimation in an uncertain world: sonar, ESP32, and HMMs

Introduction