I do have experience in robotics. There exists robust solutions to depth estimation and position estimation that are deployed in the real world, in difficult applications where they have been generalized to work with any object you can draw a box around. There is a difference between a problem that is solved in theory and a problem that is solved enough so that you can buy polished products that implement a solution and work essentially without failure. I don't think you can compare optical phase detection in the context of position estimation and depth detection to clippy in the context of AI. Phase detection is quite literally a closed form optical solution to the problem of "how far is this object away from me" as long as the object is within a few thousand times the physical aperture of the lens. It's mature enough that you can use it to drive a motor in response to movements or "that bird", "that teapot", "the closest object in that clump of pixels", "the farthest objects in that clump of pixels", as well as calculate the velocity of the aforementioned object in three dimensions. For all intents and purposes, if you have a problem of the order "what is the distance of that object" as well as "how is that distance changing over time", then you can solve it, and indeed it has been solved to very high reliability, using phase detection. That's what it means for a problem to be solved.
In other words, I wouldn't call an implementation where you can click anywhere on an image and receive a distance, all of the time, with almost any lens imaginable in any environment where there are enough photons hitting the sensor "performed in a limited context".
The technology simply hasn't been used in mainstream robotics, mainly because it's patented and difficult to implement from scratch, but these are all implementation problems not fundamental problems.