Sending a robot into the household makes it much more difficult to predict which tasks will end up defined as "critical" by the people giving it orders. My first reaction to the bit about bringing beer was the same as yours, but it brings up some interesting issues.
Suppose the robot is cleaning a room when its human orders it to drop everything and go fetch some beer and food. The human is too busy watching the game, or playing bridge or whatever it is humans do, and doesn't know the robot was in the middle of cleaning the toilet. Will the robot realize it needs to sanitize its manipulators before fetching food, even if the human has placed all orders including beer into the "critical" category? This might be an easy decision to program into its code, but only if the designers have considered this possibility.
Another commenter here said the robot should let you stop taking your prescription meds if you'd rather spend the evening drinking. Your doctor might not want your robot to contribute to your unhealthy behavior. What should the robot obey: yours, or your doctor's orders? Or should it just obey you, but then quietly report your pill-skipping and beer-drinking to your insurance company?
In many other ways a household helper robot can get complicated to design, compared to designing an industrial robot for the factory floor. This may make it a good candidate for the "many eyes" model of open source design methods.