While the micromouse is exploring the maze, there are plenty of opportunities for it to make mistakes and end up out of position somehow. Equally, there are plenty of places where there are opportunities to correct any errors that have occured. The single cell cul-de-sac or dead-end is a perfect place for both.
There are several maze configuratons that leave the mouse builder hoping will not give trouble. One of these is a series of dead-ends in a comb. It is common to see mice have to turn into each of the one cell dead-ends in a long sequence. Steering and positional errors can accumulate quite quickly.
However, once in the dead-end, the micromouse has three walls in a known position and orientation and it can make use of that information to try and make sure that it exits the cell with a smaller error than it had on entry. Here, the forward sensors are particularly useful. They tell the mouse how far it is from a wall and the difference between them tells it something about the angular error that it might have.
As the mouse enters the cell, it can ensure that it stops the correct distance from the wall ahead. that is the forward error largely taken care of.
The angle indicated by the forward sensors at this point can be used to adjust the totl angle of the next turn to correct any remaining angular error.
To be on the safe side, you can do a quarter turn to face one of the side walls and again use the front sensors to adjust the distance. Remeasure the angle and use that to alter the next quarter turn and the mouse should be in the centre of the cell facing the exit.
As the mouse leaves the cell, there will be an edge detected by the side/diagonal sensors and another correction is possible for forward errors.
If you are clever, the difference in the detected position of those edges by each sensor can give you another lateral correction. That cannot be corrected just now but it can be held over for the next turn where it will become a forward error.
Clearly, the reliability of this will depend on the quality of data available from the sensors and that will depend on the brightness of the walls. Consequently, the mouse will not get a good correction if faced with an unusually light or dark wall. Still, any process that reduces error, when repeated, should give further improvements.
Remarkably, I have never put this on the mouse and so exploring has always been a nerve-wracking experience. Not that the rest of a run is any more relaxing.
Decimus2 now has a first draft of this and, although not perfect, it seems capable of doing a fairly good job of correcting some pretty gross errors. In this video, you can see that the exit position is quite consistent even if it is not perfectly centred.