For any robot that uses a downward facing camera to look for floor based reference marks, I’d expect that the same sort of techniques used by optical mice could provide a ‘ground reference’ for the motion of the robot.

In addition, I’d expect that any robot with compliant wheels could use this technique to judge load balance by determining the degree of ‘wander’ that is imparted by the difference in compression between the two wheels.

Optical mice manage to track the path of the sensor in great detail with a rather limited sensor, illuminator and low power computing resources. Using the sort of resources a wheeled robot has available, I’d expect that a downward facing camera could provide excellent path information on a real-time basis. This would either replace or supplement any positioning information provided by wheel rotation tracking. The camera information should be far more reliable as it is independent of any wheel slip or wheel compression due to loading.

Given differential side to side loading causing the radius of the drive wheels to vary as one side is compressed more than the other, I’d expect that over a bit of forward motion the system could judge weight imbalance. Examine the number of rotations of each wheel and the drift off to one side and from that information determine the effective degree of tilt. With known (or calibrated) material properties of the wheels you should be able to effectively judge the degree to which the loading of the robot is off center.