Even if deep learning AI provides good results (~95%) when the robot is located and oriented as during the scan collect, a slight change in position or orientation may conduct to a wrong prediction.
Magnetic field inside a room is significatively variable depending on the environment. As we use tha compass to orientate the robot before doing scans, we know that scans will not be done in exactly the right orientation.
The robot moves are not 100% precise.
Combining AI deep learning and particles filter will face the above difficulties !