Just catching up to where I was 25 years ago. There are different kinds of AI. What I call the brute force method; what I call the free range method and what I call the deterministic method. The first tends to give mediocre results. The second is the dangerous one due to garbage in and garbage out. Especially if the exposure to garbage is intentional. The last one is the best for high level functional and military use. For every set of inputs, you know what the outcome will be. However it requires an expert in what you are trying to do and someone trained in the process.

A major restriction of the larger robots is the power source. To work on an assembly line or at McDonalds, they can be tethered. To work in a home, they can have a ready recharging station. Can this be solved? Sure, but I am not sure if I want to at this point.