My first battery test was a triumph. Forty-eight minutes of navigation, phone unplugged with the screen off, and Google Maps got through the lot on about three seconds of CPU.
That would have been remarkable if anything had actually navigated. It was a Monday morning at the end of October, the phone was lying on the desk next to my Mac, and the app was still called trippered and still had an audience of one… me.
Why the battery mattered at all
I’ve already told the story of why I gave up on the app that drives my Tripper pod, and it was mostly the voice that did it. The battery was a close second. An hour’s ride would take more than a third of my Pixel 9’s charge, and on a day out that adds up fast. If my replacement couldn’t do a lot better, I’m not sure there was much point in it, even if I was the only one using it.
Back then the plan was to borrow Google Maps’ directions and pass each turn on to the pod, rather than build any navigation of my own, and there were two ways of getting at them. One was reading the turns off the notification Maps shows while it’s navigating. The other was a bit more exotic. Android Auto, the thing that puts your phone on a car’s dashboard, sends the car a tidy stream of structured turn data, things like the roundabout exit number and the distance to the turn in metres. And Open Headunit, an open-source app that plays the car’s part, has a Self-Mode in which the phone acts as its own head unit. So all that lovely data lands on the same phone, with no car in sight.
Self-Mode had already passed a pocket test, happily sending turns with the screen dark. So the next question was how much battery it would eat.
The fake GPS nodded off
You can’t ride a motorbike from a desk, so the rig was a little debug app feeding the phone a fake GPS fix every second, as if it were doing 15 mph along a six-kilometre test route. Maps followed it, Self-Mode passed the turns along, and I unplugged the phone, pressed the power button and left it to it. Then I went off and did other things for three quarters of an hour (I was moving my workspace onto a new dev server and reviewing a post for this blog at the time, which tells you roughly how much attention the phone was getting).
The triumphant three seconds came from my fake GPS falling asleep. The debug app wasn’t keeping the phone awake, so about a minute after I unplugged it the phone dozed off, the fixes dropped from sixty a minute to a handful, and Maps was left with no ride to follow. So it wasn’t being clever with the battery at all… it was having a nap along with everything else! On a real ride the GPS keeps the phone awake with every fix, so I suspect this was about as unlike a ride as a test can get.
The fix was a wake lock, which is how an Android app tells the phone “not yet, I’m busy” (search for Android’s partial wake lock if you want the details). The mock drive now holds one until it arrives.
Maps went to sleep with the lights
That fixed the rig. The rig’s operator took a couple more goes to get right. On the first I got distracted and forgot to unplug the phone at all, and on the second I misread which button I was meant to be pressing. Then, once it was finally going, I asked Claude whether it was normal that the “off” screen still showed the time, a couple of notification icons and the fingerprint sensor. It was. That’s the always-on display, and I’d just never had a reason to sit and stare at it before.
Twenty-one and a half minutes later I plugged it back in, unlocked it and was greeted by “Connection lost”, and it dawned on me I hadn’t heard a single direction for quite a while. The fake GPS had behaved perfectly this time, at fifty-eight or fifty-nine fixes a minute, and the phone had stayed awake throughout. But Maps had used under three seconds of CPU again, and spent nearly the whole run parked in the background. As soon as the lights went out, Self-Mode had stopped navigating.
Then the penny dropped, and it stung a bit more. All the screen-off tests where Self-Mode had kept going, the pocket test included, had been with the phone plugged into the Mac. On USB power it kept navigating, and on battery it didn’t. So my earlier good results had really been measuring my USB cable. And I’ve no idea what Self-Mode costs on battery, and perhaps I never will, since on battery it stopped before it could cost anything.
For comparison I ran plain Maps the same way, unplugged and screen off, and it kept navigating the whole time (all ten minutes of it), at roughly 5% of the battery an hour. Take that with a pinch of salt though, the phone only reports charge in five-milliamp-hour steps, so it could be out by fifteen percent either way.
Wide awake, and couldn’t count
Straight after that I swapped my daily phone for a spare Moto g04 I keep for development, so I could stop unplugging the phone I actually need. The Mac then refused to charge it from any of its ports, with any of the cables I tried. So it ran on wireless debugging instead, and kept locking its screen on me, and in the end it went on a wall charger, where wireless debugging kept turning itself off. I could almost hear it laughing.
By then Self-Mode was out anyway. The battery result was one more reason on a list that already included asking each rider to dig into Android Auto’s developer settings. I’d already started building navigation of my own that runs on the phone (how that went is a story for another post). The next evening the prototype had its first battery run on the Moto: Brecon to Llandovery at a simulated 30 mph, unplugged, screen off.
Four minutes in I was getting worried, and I typed to Claude “the phone remains unplugged, but I dont hear any voice for the directions… is it still running ok?” It was fine, three miles out of Brecon on the A40, and the last thing it had said was “Continue for 20 miles”, just after the Tarell roundabout. It had simply gone quiet for the rest of that twenty miles, as you’d hope it would. After two days of things falling asleep on me, the first thing that was wide awake and doing its job… I’d assumed had dozed off!
It kept going for all forty-one minutes, stayed on the route and arrived. How much battery did it use? No idea! The Moto’s charge counter believes a full battery is 667 mAh, a fraction of what’s really in there, so the battery stats recorded 0 mAh used, and the level sat stubbornly on 100% the whole way.
So where’s the number?
That’s three battery tests in two days, and each one measured something other than my app. The first measured my fake GPS, the second my USB cable, and the third ran on a phone that couldn’t count.
I did get a number in the end, later that same evening, by putting the prototype back on the Pixel 9, whose counter works. It came out a fair bit lighter on the battery than Maps. But the fake GPS was still doing the driving, so the phone’s real GPS receiver was probably switched off the whole time, and I’d imagine that’s one of the bigger drains on an actual ride. So it’s a desk number, and after those two days I’m a tad wary of any number when I haven’t checked the rig behind it.
I can tell you what I see on my own rides, though. With the old app, an hour on the bike would cost me more than a third of the battery. With Ridelark I can go out for a ride and still have something like 90% left when I get home. That’s one rider on one phone, an anecdote rather than a measurement, and it may not hold up on anybody else’s.
The closed beta is where that gets proven, or not, on other people’s phones on other people’s bikes, with a real GPS keeping them awake. And if one of the testers messages me twenty minutes up the A40 asking whether it’s still running… it probably is.





