They want realistic randomness in the apartment layouts. This is a quick, effective way to get that. If they were honest with the hosts, it wouldn’t even be a bad idea.
You don't need that until the very end. They should be modeling many houses first, and they can get that by having employees measure their own house. They should also know something about the edge cases and have a lot of very unrealistic houses modeled.
Then they should have a lab with real furniture and movable walls so they can do controlled real world tests. Once the above tests are done you add confidence with random real world tests.
The types of problems seen here are things that your lab tests should fail and keep you out of real world tests. Particularly when the test subjects don't have some sort of test agreement
I assure you that if randomness is the order of the day, then involving handymen to arrange things is a sure-fire method of getting there.
The cheaper the handymen come, the better the randomness is.
One doesn't even have to tell them to be random; that part happens all on its own in the ways that real apartments ebb and flow.
And the handymen themselves can be rotated out every couple of weeks, as well. The cost of rotation is low. The handymen are plentiful, and they are happy to get hired for a day here or there for literal odd jobs.
(If anyone wants even more practical solutions for robot testing facilities that don't make headlines by pissing people off, please put your contact information in the space provided by pushing the "Reply" button. Thanks!)
Then they should have a lab with real furniture and movable walls so they can do controlled real world tests. Once the above tests are done you add confidence with random real world tests.
The types of problems seen here are things that your lab tests should fail and keep you out of real world tests. Particularly when the test subjects don't have some sort of test agreement