Cognitive Models of Decision Making Processes for Human-Robot Interaction

Abstract

A fundamental aspect of human-robot interaction is the ability to generate expectations for the decisions of one's teammate(s) in order to coordinate plans of actions. Cognitive models provide a promising approach by allowing both a robot to model a human teammate's decision process as well as by modeling the process through which a human develops expectations regarding its robot partner's actions. We describe a general cognitive model developed using the ACT-R cognitive architecture that can apply to any situation that could be formalized using decision trees expressed in the form of instructions for the model to execute. The model is composed of three general components: instructions on how to perform the task, situational knowledge, and past decision instances. The model is trained using decision instances from a human expert, and its performance is compared to that of the expert.

A Framework for Developing and Using Shared Mental Models in Human-Agent Teams

“…Others suggest developing (Neerincx, de Greef, Smets, & Sam, 2011; Ososky et al, 2012) or have developed simulated robotic SMMs (Kennedy & Trafton, 2007) but do not provide comprehensive frameworks for realizing robot SMMs. It is suggested that SMMs are necessary to make the robots true teammates (Ososky et al, 2012), and ACT-R (adaptive control of thought–rational) was used to demonstrate, in simulation, the use of mental models to improve performance (Libiere, Jentsch, & Ososky, 2013), although no comprehensive framework was provided for integrating SMMs into robots that can function in uncertain and dynamic environments.…”

An Investigation of Human Decision-Making in a Human—Robot Team Task

“…ACT-R is a theory of cognition that involves the integration of multiple information processing modules (e.g., declarative memory, perceptual-motor) to produce a holistic representation of human cognition (Anderson, Bothell, Byrne, Douglass, Lebiere, & Qin, 2004). Details of the computational decisionmaking model, and a preliminary report on model training using data from expert decision makers, can be found in Lebiere, Jentsch, and Ososky (2013). The current paper reports on an extension of the project reported in Lebiere et al (2013).…”

“…Details of the computational decisionmaking model, and a preliminary report on model training using data from expert decision makers, can be found in Lebiere, Jentsch, and Ososky (2013). The current paper reports on an extension of the project reported in Lebiere et al (2013). The goal of the extended research project was to validate the ACT-R model against the responses of novice human participants.…”

“…As shared mental models have been linked to successful performance in human-human teams (Mathieu, Heffner, Goodwin, Salas, & Cannon-Bowers, 2000), efforts to implement shared mental models in human-robot teams and to facilitate human-robot interaction are underway (Lebiere, Jentsch, & Ososky, 2013). Implementing shared mental models in robots involves the translation of non-computational knowledge held by human mental models to a computational representation, understandable by robots.…”

“…Computational cognitive models provide a promising approach to translating human mental models to robot mental models, and future SR team performance. A computational cognitive model's representation of human mental models can take many forms, including symbolic frameworks (e.g., decision trees), semantic networks, as well as statistical frameworks (e.g., Bayesian networks) and include representations of the cognitive processes that underpin human mental models (Lebiere, Jentsch, & Ososky, 2013). Because cognitive models represent the mechanisms that underlie mental models, they can be used to train agents to make decisions based on the actions and decision making of humans through a methodology called instance-based learning.…”