Towards Computational Modeling of Human Goal Recognition

Abstract

Recently, we are seeing the emergence of plan- and goal-recognition algorithms which are based on the principle of rationality. These avoid the use of a plan library that compactly encodes all possible observable plans, and instead generate plans dynamically to match the observations. However, recent experiments by Berkovitz (Berkovitz, The effect of spatial cognition and context on robot movement legibility in human-robot collaboration, 2018) show that in many cases, humans seem to have reached quick (correct) decisions when observing motions which were far from rational (optimal), while optimal motions were slower to be recognized. Intrigued by these findings, we experimented with a variety of rationality-based recognition algorithms on the same data. The results clearly show that none of the algorithms reported in the literature accounts for human subject decisions, even in this simple task. This is our first contribution. We hypothesize that humans utilize plan-recognition in service of goal recognition, i.e., match observations to known plans, and use the set of recognized plans to conclude as to the likely goals. To test this hypothesis, a second contribution in this paper is the introduction of a novel offline recognition algorithm. While preliminary, the algorithm accounts for the results reported by Berkovitz significantly better than the existing algorithms. Moreover, the proposed algorithm marries rationality-based and plan-library based methods seamlessly.

Language Model-Based Player Goal Recognition in Open World Digital Games

“…Plan-library based goal recognition encodes all possible observable plans, and a goal is inferred by searching a matched plan (Schmidt et al 1978;Kautz and Allen. 1986;Charniak and Goldman 1993;Mirsky et al 2016;Treger and Kaminka 2022). Recently, Rabkina et al (2022) proposed a cognitive model-based goal recognition framework using the analogy of similar cases retrieved from a plan-library and evaluated it using MineCraft and Monroe.…”

“…Planning-based approaches considers goal recognition as an inverse planning problem, while data-driven approaches use machine learning to recognize goals directly from observations. Planning-based goal recognition has generally been studied using theoretical problems and datasets (Schmidt et al 1978;Mirsky et al 2016;Treger and Kaminka 2022;Baker, Tenenbaum, and Saxe 2007;Ramırez and Geffner 2011;Pereira, Pereira, and Meneguzzi 2019), while data-driven goal recognition has primarily been used in simulation and game environments, such as an action-adventure game (Gold 2010), PDDLGym (Amado, Mirsky, and Meneguzzi 2022), and CRYSTAL IS-LAND (Mott, Lee, and Lester 2006;Ha et al 2011;Min et al 2014Min et al , 2016Goslen et al 2022a,b).…”

“…In planning-based goal recognition, a key assumption is optimality, which assumes that the actor is rational and tends to undertake an optimal plan to achieve a goal. However, humans are not always optimal (Treger and Kaminka 2022), and thus we should, in general, assume actions are suboptimal in environments involving human actors. Particularly, in open-world digital games, optimality is not always a valid premise, because players have incomplete knowledge of the gameworld and undertake inefficient paths toward accomplishing goals, while exploring and learning about the environment.…”