The were incompetent enough to go to real world testing when these issues would have been obvious from a basic model kitchen test. Obviously their bot is in the very early development stages where it can't do any of the basic things right, they're nowhere near the phase where they needed real world testing. You don't need a real house to tell that your bot keeps damaging furniture, floors, and other items. You iron that out in the lab, then go in a real setup.

And yet they weren't able to build a model house or even just some model rooms for a controlled environment and practice there full time first. They could have done round the clock testing, with full flexibility of the arrangement, no need to waste time moving hardware around and risk damage, no liability, and more. A fake house costs next to nothing. A (fake) model kitchen is cheaper than an Airbnb stay.

Have you seen how many public demos from manufacturers of advanced robots like Boston Dynamics are using "artificial" obstacles and layouts? It's obvious they did a lot of development in those conditions. You don't need someone's home to find out if your robot can grab a plate without destroying it, or climb a flight of stairs.