“Usually people put the cheap and cheerful IMUs, which basically means that all of your sensors are lying to you. And the EKF's job is to sort out all the lies to create the truth, because every sensor has errors.”

…who's the head for state estimation, looks after our EKF architecture, which is state estimation. Working out where you are, which way up you are, and where you're going is really important to this type of vehicle. And so a huge amount of work goes into the state estimation. That's all Paul's area.

So state estimation is taking the input from all of these sensors and fusing them together to determine where in the world I am, which way up I am. Which my nose is facing and so on.

So you essentially run several instances of this algorithm or perhaps different algorithms to try and estimate the state of the vehicle. And there's some system of, is it based on voting or quorum or there's some other heuristics used to determine this sensor is telling us we're at this altitude…

Yeah, so the robustness is at several different levels. So the most common type of vehicle, I mean, some vehicles only have one IMU, but most commonly three IMUs. And by default, we arrange things so you run one EKF per, we call them lanes. There's basically one EKF per IMU. And then you have…

So to make that a little bit more concrete for the listeners, can you recall any examples where this EKF switching or the vehicle essentially flip-flopping between different opinions it has about where it is or which way it's facing? How does that manifest in flight and cause a real issue?