Agent behavior monitoring using optimal action selection and twin gaussian processes
Luis Ávila, Ernesto Martínez · 2014
Abstract. The increasing trend towards delegating complex tasks to autono-mous artificial agents in safety-critical socio-technical systems makes agent be-havior monitoring of paramount importance. In this work, a probabilistic ap-proach for on-line monitoring using optimal action selection and twin Gaussian processes (TGP) is proposed. A Kullback-Leibler (KL) based metric is proposed to characterize the deviation of an agent behavior (modeled as a controlled sto-chastic process) to its specification. The optimal behavior specification is ob-tained using Linearly Solvable Markov Decision Processes (LSMDP) whereby the Bellman equation is made linear through an exponential transformation such that the optimal control policy is obtained in an explicit form.