I’m not sure that’s what you’re asking about, but it’s the first thing that came to mind :)
The research never felt like a chore, quite the opposite. Because I was writing down what I was learning, I felt I could fairly confidently speak at length on some aspects of housing design. As is often said: if you write down what you learn, you remember it better and it deepens your understanding.
Another assistant devised a clever solution to a problem in making practical measurements of shearing. The physicist wanted a way to fit nearly-spherical balls of clay into the middle of blocks of clay of a different color, so that he could pop in the sphere, shear the block, and then compare measurements of the resulting distortion with mathematical predictions. One problem was how to get mostly-spherical balls into a smoother, more-nearly-spherical shape, and the other assistant rigged up a clever device made out of some photographic development equipment that rolled a near-spherical ball of clay around in a bin for hours, resulting in a very-nearly-spherical ball.
More relevant to computing, about six or seven years ago I worked for about a year and a half on a subproject of Peter Neumann's Clean-Slate computing research project. The overall project was funded by DARPA, and pretty large. (I don't know its current status). Its mandate was to examine a group of proposed ways to address rampant security and reliability problems in computing.
I worked on maintaining and extending an implementation of a novel programming language intended to be the main system and application language of a new computing platform. I didn't design the language, or any substantial part of it; I just fixed bugs and proposed and added conveniences for programmers.
The language used both static and dynamic type disciplines, simultaneously, in order to address the areas that each discipline handled best.
The language, called "Breeze," was meant to represent some security and data-integrity invariants in the type system that are not normally handled that way. For example, Breeze made it a type error to transmit privileged data to a recipient or a process with the wrong privileges.
It was a terrifically interesting project. I enjoyed it a lot. I did notice at least one problem with approaching it as an academic research project. It sometimes seemed like the incentives to publish novel work made the researchers less curious than I expected about related work that had previously been done--I'm guessing because nobody wants to publish a paper that basically just says, "these guys tried this thing and it worked pretty well."
I can't be sure, but it seems like maybe this preference might have slowed practical progress in some areas. On the other hand, it could also be that the researchers had more substantial reasons for their lack of interest in some older work, and I just didn't know about them.