Stochastic Goal Recognition Design Problems with Suboptimal Agents

Abstract

Goal Recognition Design (GRD) problems identify the minimum number of environmental modifications aiming to force an interacting agent to reveal its goal as early as possible. Researchers proposed several extensions to the original model, some of them handling stochastic agent action outcomes. While this generalization is useful, it assumes optimal acting agents, which limits its applicability to more realistic scenarios. This paper presents the Suboptimal Stochastic GRD model, where we consider boundedly rational agents that, due to limited resources, might follow a suboptimal policy. Inspired by theories on human behavior asserting that humans are (close to) optimal when making perceptual decisions, we assume the chosen policy has at most m suboptimal actions. Our contribution includes (I) Extending the stochastic goal recognition design framework by supporting suboptimal agents in cases where an observer has either full or partial observability; (ii) Presenting methods to evaluate the ambiguity of the model under these assumptions; and (iii) Evaluating our approach on a range of benchmark applications.

Generalising Planning Environment Redesign

“…Most approaches to environment redesign assume the observer's objective is to modify the environment to facilitate recognizing goals and plans (Keren, Gal, and Karpas 2014;Son et al 2016;Mirsky et al 2019). Later works in GRD frame and solve this task under different observability settings Karpas 2015, 2016a,b), environment assumptions (Wayllace et al 2016(Wayllace et al , 2020Wayllace and Yeoh 2022), or observer's capabilities (Shvo and McIlraith 2020;Gall, Ruml, and Keren 2021). Unlike these works, we assume the interested party might want to modify the environment for tasks different than recognising goals and plans.…”

Can LLMs Fix Issues with Reasoning Models? Towards More Likely Models for AI Planning

“…One subclass of domain authoring problems is designhere, the task is not to author a new domain but to evolve an existing one to optimize certain criteria like making the task of recognizing the goals of agents in the environment easier (Keren, Gal, and Karpas 2014;Mirsky et al 2019;Wayllace et al 2016) or making the behavior of agents easier to interpret (Kulkarni et al 2019(Kulkarni et al , 2020. Here as well, search techniques reveal multiple possible design options that can be enforced on a domain to achieve the desired effect.…”

Block-Level Goal Recognition Design

“…Keren et al proposed using the WCD as a performance measure in GRD Karpas 2014, 2019). Subsequently, many works extended the WCDbased GRD model to deal with non-optimal agents (Keren, Gal, and Karpas 2015), non-observable actions (Keren, Gal, and Karpas 2016a), privacy preserving in GRD (Keren, Gal, and Karpas 2016b), stochastic domains (Wayllace et al 2016;Wayllace, Hou, and Yeoh 2017), game-theoretic GRD (Ang et al 2017), GRD for plan libraries (Mirsky et al 2019), partially-observable states (Wayllace et al 2020), incomplete information (Keren 2019), information shaping , stochastic domains with suboptimal agents (Wayllace and Yeoh 2022), interleaving between agents and observers (Gall, Ruml, and Keren 2021), and agents with multiple goals (Au 2022). (Keren, Gal, and Karpas 2020) is a survey of the works on GRD before 2020.…”