Counting Doors by Integrating POMDP Models and PID Controllers

This paper presents an enhanced hybrid planning architecture that integrates Partially Observable Markov Decision Processes (POMDPs) with Proportional-Integral-Derivative (PID) control for indoor robot navigation under uncertainty. Our approach uses PID controller feedback—proportional, integral, and derivative signals—as observation variables in the POMDP model, exploiting well tested mechanisms of processing sensor data. A case study is conducted using an iCreate3 robot navigating a university corridor to deliver a load to a specified room by identifying door frames. We evaluate three progressively refined models with increasing sensor integration and probabilistic reasoning. Experimental results demonstrate that the multi-sensor fusion model achieves an 85% task success rate, outperforming simpler PID-only baselines. This framework provides a viable foundation for scalable, robust planning in lowcost robot platforms without advanced localization systems.