I also think this is a case of nothing more than operant conditioning.
I once saw a demonstration of how they trained a mouse to navigate an obstancle course (supposed to be a cross-section of inside a house wall with joists and so on) for some movie. When trained, the course took the mouse about 20 seconds to run and it looks like it must be one really smart mouse. But really, it learnt the route one step at a time over a period of days. I suspect a similar technique is used here.
I once saw a demonstration of how they trained a mouse to navigate an obstancle course (supposed to be a cross-section of inside a house wall with joists and so on) for some movie. When trained, the course took the mouse about 20 seconds to run and it looks like it must be one really smart mouse. But really, it learnt the route one step at a time over a period of days. I suspect a similar technique is used here.