This paper investigates the problem of joint estimation of a pedestrian user path and the available WiFi access point locations. The observations are limited to unlabeled WiFi received signal strength (RSS) values. The problem is formed as a partially observable Markov decision process and RSS gradients are integrated to estimate and update the user locations along the path. The RSS data is modeled as a Gaussian process and gradient vectors are updated for each step based on the motion dynamics. Realistic assumptions and constraints are introduced to model the user's movement and reduce the computational complexity.