Integrating POMDP Models with PID Controllers with Application to Counting Doors on a Corridor

Abstract

We propose a hybrid planning approach integrating Partially Observable Markov Decision Processes (POMDPs) with Proportional-Integral-Derivative (PID) controllers for mobile robot navigation. The method leverages PID feedback components as observation variables in a POMDP model to improve decision-making under uncertainty. This work is tested on a Create3 robot tasked with identifying and counting doors in a corridor using only onboard sensors. Our approach improves sensor noise robustness, achieving a success rate of up to 85%, demonstrating the feasibility of integrating AI- based planning with low-level control.