I asked this question on AI stack exchange but didn't get a response.
Basically I'm interested in how you deal with inherent error between your ideal discrete state and your actual state in control planning.
For example, suppose we have a map of an area which is split up into square blocks which represents the possible states of the robot. You then make a plan based upon its starting state - but the real state is somewhat offset from this ideal state.
How do you handle it? If you do a MPC and forward simulate the dynamics, error is going to accumulate - do you replan after every movement? How does it work?