Discrete belief propagation network using population coding and factor graph for kinematic control of a mobile robot

Abstract

This paper presents a probabilistic graphical model in the form of a factor graph to perform hierarchical probabilistic inference by computing kinematics of an omnidirectional mobile robot. We propose applying population coding principles to encode messages transmitted within the factor graph to update the network's internal belief, as inspired by neuronal information processing. We examine two inference scenarios in this paper: first for single wheel motor control using real data from an omnidirectional mobile robot; and second for the robot's velocity and orientation in real-world coordinates using simulation data. The experimental results for the first scenario show that the factor graph can learn input-output relations almost perfectly and the simulation results for the second scenario demonstrate that the selected model in the factor graph is quite robust against disturbances due to noise during inference. The results of this study can be applied in more complex intelligence tasks, which build on top of this basic kinematics system.