Goal Recognition for Rational and Irrational Agents

Contemporary cost-based goal-recognition assumes rationality: that observed behaviour is more or less optimal. Probabilistic systems, however, generate probability distributions on the basis of suboptimality. We show that, when an observed agent is only slightly irrational (suboptimal), state-of-the-art systems produce counter-intuitive results. We present a definition of rationality appropriate to situations where the ground truth is unknown, define a rationality measure (RM) that quantifies an agent's expected degree of suboptimality, and present a novel self-modulating probability distribution formula for goal recognition. Our formula recognises suboptimality and adjusts its level of confidence accordingly, thereby handling irrationality---and rationality---in an intuitive, principled manner.