Thoughts and observations one year on.
The idea of making a simple internet enabled Granny Monitor for friends and family was sown when working with the charity Designability. I helped them build smart homes solutions for people suffering from dementia. These cost thousands of pounds. I wanted something that cost £100 quid and you could install yourself.
Once I discovered electricimp I knew I had the platform I’d been waiting for. I created a simple device consisting of a PIR, temperature and humidity sensor. Electricimp removes the whole burden of dealing with cloud connectivity, misbehaving WIFI routers and provides a rich event driven software platform & APIs.
The vital signs of daily life
I’ve long held the belief that placing a monitor in the kitchen was the best way to a get a sense that someone was looking after themselves. It’s the place where you make a cuppa first thing, cook a meal and make a hot drink before bed time.
Learning by building – using the data
Installing Mk1 monitor in my study we quickly started to collect data and the question soon arose about the best way to visualise all of this stuff. We started with a simple time series graph, which quickly showed we needed to distinguish between movement and activity.
Activity was simply a number of movements within a given period. This was to exclude false positives and establish that someone was around.
Talking this through with friend we both felt that some form of timeline was needed. The first cut was too noisy you couldn’t see the wood for the trees. So we created a filter that showed a single activity/movement for each hour. This has been refined over time the and current incarnation is built using JQuery Mobile.
Evidence of absence
We soon realised that showing no activity was equally as important, visually we show this by a brake in the timeline. I then installed monitor Mk2 in our kitchen and started to monitor the family.
Being parents meant we were out of the kitchen by 10pm and up again after 6am. So it was quite a surprise, to me, to find my mother in law was a night owl often active at 1am!
The other thing my partner and I’ve have recently noticed is that Grandma is being seen less in the afternoon. We create an alert if someone is up but not seen for 4hrs. It turns out she’s been having an afternoon nap more often.
Alerts and notifications
When electricimp offered their cloud Agents we had the platform to start creating a wider set of alerts and notifications.
At first we used the timeline to check on Grandma now we just rely on notifications. So much so they have become an integral part of our daily lives. My five year old says “Grandma’s up!” on hearing the special “ring tone” on her mum’s phone.
This has been positive for Grandma too. Recently she was called in succession by me and her daughter when we both got the “Not up!” alert. She said without prompting “I feel reassured that someone is looking out for me”. No worries from her about big brother a common concern among the elderly. There is sufficient evidence that fear of falling is a significant predictor of future falling, so we’re hoping that the monitor is helping in this regard.
We found the language you use in alerts is important, so we started subtlety changed things so we could detect a bit late from late when someone got up in the morning. We also had to allow for single and plurals “Dad’s up!” or “Mum & Dad are up!”
Temperature and humidity sensing
Temperature and humidity have a well documented and important role to play in well being. However one surprise was correlating humidity to activity. It’s shown us that you can imply that some form of cooking is going on. In bathroom it could show a bath or shower is running.
One thing I learnt quickly when first installing wireless sensor networks is there are three key indicators for reliability; energy level, signal strength and the time of the last transmission. The electricimp has exceptional WIFI performance better than any other device I have seen including commercial grade wireless building automation technology. Signal strength is a great commissioning tool as you can often get the device to work but you don’t know if your on the edge. We alert and display on all three key indicators of a device’s health.
Machine learning (ML)
Data science is out of my league but we have clear activity profiles. I’m sure you could predict or hint at when someone’s activity pastern might change. I need to find a friendly data scientist and get their interest!
That said I’m wondering if ML could deal with unexplained activity changes. Last summer Grandma started to get up early on certain days of the week. I asked her why? “ow I get up to get the milk in and go back to bed” she still has the milk delivered and didn’t want it going off.
Given that you know someone is up and about it would be a good time to call them a with a friendly reminder “remember mum, the health visitor is coming at 11 today”. Given we already use twilio this would be quite a simple addition and the phone isn’t something that the elderly are intimidated by.
I also use pushover for push notifications. Pushover allows for priority and alert acknowledgements which would allow us to escalate or create round robin alerts to ensure someone responds when a critical issue is raised. But I love the fact with pushover you can change the alert ring tone via the API the tug boat one still makes me jump on hearing it!
We’ve got a simple messaging function in the app that adds a comment to the timeline. It would be good to include SMS messages between carers in that too. Also at friends parent’s house we’re monitoring external access, so adding camera snap shots to the timeline would also make sense.
On the hardware side electricimp have made great strides in reducing imp power requirements. But I’m still short of my goal of one years of battery life. One way to squeeze a bit more life would be to batch event updates to the cloud if they were not critical by staying offline.
I’m also looking forward to the 3/4g version of the imp. We have some friends whose elderly parents live in two different locations so having one monitor that can take with them would be a great solution.
Monitor: electricimp & sensors