Challenge
With EAN and Stanford, we studied HVAC control that reflects the comfort of individual occupants. Conventional HVAC operates from a temperature sensor mounted in the room, so it cannot account for how many people are actually present or how warm or cold each of them feels.
The constraint here was not the algorithm but the data. However capable the model, without data representing comfort at the level of an individual person there is nothing for it to act on. That data did not exist on site, and obtaining it first required a physical device that recognizes people in the space and measures the environment.
Approach
Alpaon handled the full path, from hardware to AI.
We designed and built a custom PCB that integrates several sensor modules, for temperature, humidity, and occupancy, onto a single board to capture the room environment. Rather than handling each sensor separately, we laid out the circuit so the required signals were captured on one board, obtaining the signals the model needed to specification.
We combined the environmental data with occupancy recognition from a vision AI to estimate each person's comfort, and fed that into the HVAC control input. Sensing, occupancy recognition, and HVAC control were arranged to operate as a single automated control loop.
Because one team developed both the hardware and the AI, we could tune sensing error and recognition accuracy from the circuit and the model at the same time, shortening the development cycle.
Outcome
The finished system controls heating and cooling autonomously based on occupants rather than the average of the room. The joint research was published at IEEE AIoT 2025, confirming the validity of the approach.
The project is a case of generating data that did not exist on site at the sensor level, and connecting it through to AI decisions and control within a single team.