Refining pose probabilities
We will generate new populations by resampling, choosing based on probability, and creating a particle filter, where our system is more likely to sample from poses that are more probable than others.
How do we make probabilities for poses? We incorporate observation models. An observation model compares predictions with an observation (e.g., a sensor) to see how likely the pose is. As the robot moves and our poses move, these weights will change.
This cycle of making many guesses, making updates, making observations, then making new guesses is the heart of Monte Carlo simulation.
In this section, we'll create and incorporate the boundary observation model, based on how close our robot is to the boundary using encoder motion only. Although being outside the barrier is a clear boundary, we'll later incorporate more sensor data that is fuzzier. This can then update poses' weights, so we can consider how likely they are, and resample a new generation...
