It's not really fair to call the Lighthouse system an "inside-out" tracking system. "Inside-out" generally refers to the tracking data collection and processing happening on data collected exclusively by the headset, without any specialized, external reference. That's not what Lighthouse does.
Lighthouse is nearly identical to outside-in camera tracking, with the one wrinkle being that the photons flow in the opposite direction. Instead of a fixed, designed constellation of point-like emitters on the headset, you have a constellation of point-like detectors. Instead of a stationary grid detector, you have a stationary grid emitter. But otherwise, the data is practically the same, the math is all the same, the calibrations are all the same, and the whole system doesn't work without those stationary, external reference points.
Similarly, ray tracing doesn't simulate photons leaving light sources, bouncing off surfaces, and arriving at a camera sensor. The simulation is of anti-photons, leaving the camera, bouncing off surfaces, and seeing what lights they hit. It's like conventional current actually being the opposite direction of electron flow. The systems can run forwards or backwards and get the same answer.
Actually inside-out tracking does a completely different thing. The acronym "SLAM" stands for "simultaneous locating and mapping". It's building up a coherent, consistent model of the world around it. It adapts to new surroundings.
Bump a Lighthouse emitter or CV1 camera out of position and everything stops working because the data no longer makes sense. Designing a Lighthouse headset or controller requires given the tracking code a 3D model of the position of all the detectors.
But move the furniture around in a room and SLAM catches up in a few seconds. SLAM also doesn't care about the shape of thing you're tracking. Hell, it really doesn't care all that much about the quality of the camera feed, other than being relatively high framerate and not very noisy.
It's simple enough to say that on the lighthouse system, the sensors are on the headset/controllers, and the beacons are cast from the lighthouse boxes.
On the oculus system the sensors were the external cameras, and the beacons were the LEDs on the devices.
My experience with both was that the oculus system did really well in a seated system but for room scale games the lighthouse system does better, especially when the controllers go behind you like in the valve archery game.
I haven't bought an oculus system since the DK2 so not sure how sophisticated it is now.
Windows MR (both VR headsets and the HoloLens), Magic Leap, Vive Focus, Pico Neo, and the upcoming Linx all use inside-out cameras, all with their own implementations. HTC Vive, Vive Pro, Vive Cosmos, Valve Index, Varjo XR-3 and VR-3, and PiMax headsets are the only ones using outside-in tracking anymore, and they're all using specifically Lighthouse.
First of all, you just don't really do that very often. People have rotator cuffs and elbows that make any action in those regions fairly uncomfortable.
Second, the all current VR systems primarily use inertial tracking. The visual tracking is only there to correct for drift out of the reference frame. Whether it's Lighthouse or Rift CV1 outside-in cameras or inside-out cameras on every other system, you can put your hands in the sensor blind spot for several seconds before it becomes a problem.
99% of the time, you're working with your hands in front of you. Lighthouse doesn't care about your hands in relation to your body. But it does care about your body in relation to the base stations. Lighthouse's blind spots are constantly changing over the course of your play session. Quest's are always in the same spot.
So many times I've found myself in a corner on the opposite axis of my base stations and my own body is blocking my controllers' view of the base stations. When that happens, you have to have enough awareness of what is going on to understand why your hands start slowly floating away while you are trying to work on something, having forgotten your orientation in the real world room. It's literally immersion breaking.
"Inside-out cameras can't track behind your head" is really not the problem that your random Valve fanboy on Reddit makes it out to be.
Controllers behind the head are tracked by some algorithm magic that fuses the last seen position by the cameras with accelerometer and gyro data for the blind spots. Seems to work like a charm. Probably not as good as full lighthouse system, but good enough.
Every extant tracking system uses IMUs as the primary tracking sytem. The Lighthouse base stations and the Oculus camera tracking are used to correct for sensor drift.
You need it to be this way, because the IMUs can run at fairly high frequencies (200 - 1000Hz), which is (in part) keeps latency low. The data paths and processing needed for the reference frame corrections are so complex that they can't be run anywhere near as fast. It's why the hand tracking on the Quest is so high-latency: there's no IMU on your hand.
And it's not "algorithm magic". It's mostly just Kalman Filtering.
Lighthouse is nearly identical to outside-in camera tracking, with the one wrinkle being that the photons flow in the opposite direction. Instead of a fixed, designed constellation of point-like emitters on the headset, you have a constellation of point-like detectors. Instead of a stationary grid detector, you have a stationary grid emitter. But otherwise, the data is practically the same, the math is all the same, the calibrations are all the same, and the whole system doesn't work without those stationary, external reference points.
Similarly, ray tracing doesn't simulate photons leaving light sources, bouncing off surfaces, and arriving at a camera sensor. The simulation is of anti-photons, leaving the camera, bouncing off surfaces, and seeing what lights they hit. It's like conventional current actually being the opposite direction of electron flow. The systems can run forwards or backwards and get the same answer.
Actually inside-out tracking does a completely different thing. The acronym "SLAM" stands for "simultaneous locating and mapping". It's building up a coherent, consistent model of the world around it. It adapts to new surroundings.
Bump a Lighthouse emitter or CV1 camera out of position and everything stops working because the data no longer makes sense. Designing a Lighthouse headset or controller requires given the tracking code a 3D model of the position of all the detectors.
But move the furniture around in a room and SLAM catches up in a few seconds. SLAM also doesn't care about the shape of thing you're tracking. Hell, it really doesn't care all that much about the quality of the camera feed, other than being relatively high framerate and not very noisy.