From May 25th to May 27th, our team was an exhibitor at the Connected Conference in Paris. It was a great opportunity to meet many interesting people and present craft ai.
Each year, the connected conference builds a full-scale connected home within the tradeshow, filled with many connected things ready to play with. This year, craft ai was invited to take part in the connected home and orchestrate various connected devices. It was a perfect occasion for us to demonstrate our vision of the smart home, done right: an intelligent and personalized home that automatically adapts to the habits of its occupants. We showed how a smart home can be much more than individual connected devices, but rather a coherent experience for the user, orchestrated by AI.
craft ai is an AI platform developers can use to create services that adapt to each user. In the connected home, craft ai allows the home to learn the habits and preferences of its occupants via the data collected from the connected Things. The occupants don’t need to tell craft ai to learn their preferences because craft ai does it automatically, it learns for them!
craft ai learns behaviors by finding patterns depending on the daily usage of the occupants and the context: time of the day, the outside luminosity, temperature, etc. The models are continuously generated and used to automate the home.
In the case of this particular home, we interfaced with these devices:
- Philipps Hue lightbulbs,
- Somfy blind and window opener,
- The Keys door lock,
- Samsung TV,
- Nodon button that occupants toggle to tell they’re sleeping (in the absence of a sleep API).
Before the conference, we learned three users models, each of them with different habits and preferences: Gisele, Kevin and Robert. We used those models directly in the demonstration, but in a real life scenario the profiles would keep updating according to what users do in the home.
Through the use of Gablys, portable Bluetooth beacons, and Bluetooths hotspots strategically placed in the house, we could find the location of the inhabitants. With this final contextual information it becomes possible to predict the different behaviors from the pre-learned patterns by craft ai as the users move in their home.
In the morning, Kevin wakes up and goes to the kitchen to make breakfast. The bedroom blinds stay closed because Gisele is still asleep. However, Kevin does not want to hit something while he’s in the dark, so the light turns on dimmed.
When Kevin wakes up he likes to have a blue light in the bedroom. As craft ai has learned his current habits, the light is automatically adjusted without having to action anything.
Kevin moves to the living room and thanks to his Gablys beacon craft ai can locate him and open the living room blinds, knowing that Kevin likes to enjoy the morning sun.
Then, Gisele joins him and decides to watch her favorite morning TV show. So, she turns on the TV. As craft ai learned her habits, it sets a laid back atmosphere with a dimmed light colored in pink to wake up smoothly. The blind automatically goes down and the light turns to a pink color.
It’s finally time to go to work for Kevin and Gisele. When they leave their home, the blinds close completely and the light goes off. But on this day, they forgot to turn off the TV. Through learning habits, we are able to realize that this is abnormal behavior. craft ai understands this anomaly and sends a notification to the occupant. As part of the connected home of the show, we flashed the lights to signal an unusual thing, but this could be done by sending an SMS for instance.
To orchestrate and automate this smart home, we had to be very patient to make the devices work. Our system is based on context and habits learning of the occupants with their connected devices, so it was essential to have an indoor location system to know who is in which room and when. We had opted for a Bluetooth connection with Gablys beacons. Unfortunately, the Bluetooth signal wasn’t stable and regular enough in the noisy environment of the Connected Conference. So the demonstration didn’t always work properly and we had to keep trying. This was not perfect but it was the best option we found. We are still looking for a completely satisfying indoor positioning system..
This was the first time we showed how craft ai works in a real smart home instead of a virtual demo. Overall the response was very positive and people seemed to immediately get the demo. It allowed visitors to better understand what we do and especially to see physically what craft ai is capable of. With virtual demos, we usually have to give more explanations, especially about how this would translate to concrete use cases.
Thanks to ARUCO for giving us the possibility to take part in this project.