Hello Reader,

Welcome back to The Build Log! We're almost ready to take the submarine on its mission under the sea, to explore a crashed WWII B-17 bomber. In this newsletter, we'll show you the tests we did before the mission, to make sure our underwater localization works well enough.

Quick recap - our localization system is based on an external compass in the AUV's antenna - that provides the current heading (along with sensors in the Pixhawk), and a DVL (at the bottom of the submarine), which is essentially a very accurate speed sensor. If we know the location where we submerged, we can integrate the heading and speed readings to estimate the current location.

Of course, the position estimation error drifts over time. We just need to make sure that it doesn't drift too much, so that we can actually record / take a picture of the airplane.

To test it properly, we used our usual trick - a surface mission on the surface, at sea, at the test site. That way, we can compare real-time GPS data, with the position estimated by dead-reckoning algorithm running on the submarine. We needed to constantly monitor the submarine on a stand-up paddle to not lose it, and make sure no boats crashed into it.

We call this mission, a calibration dog. It's because the shape kinda looked like a dog. It tests both the dead-reckoning-based localization system, and well as how well our navigation algorithm handles cornering. The first graph shows the "real" GPS position vs the DVL dead-reckoned position. The second graph shows the position error over time - higher value, means worse result. Lower value - better result.

As you can see, the results are worse than we expected based on previous tests (especially given the optimistic results we have shared in the last email) . The localization error grew to very high values (up to 37 meters!) only 7 minutes into the mission. Here is what we figured out:

1. The DVL was sometimes losing "bottom lock" some of the time - this is caused by bubbles generated by the front motors that go under the DVL;
2. The DVL was underreporting velocity - it was too low by roughly 5% - this caused the error to be large when far away from the starting point, and get lower again once it came back;
3. The GPS antenna sometimes went underwater - The blackouts in the GPS heading are caused by the submarine flying slightly below the surface than on it;
4. The compass heading was innacurate - probably the biggest factor. Sometimes it was off by a couple of degrees - by how much, depended on the direction the submarine was facing

In next week’s newsletter, we’ll show you how we fixed these issues and how the improved (!) results we achieved!

Cheers,

Piotrek, CPSdrone

PS: You can access all of the previous releases of the newsletter here.