Formal Verification of Human-Robot Interaction in Healthcare Scenarios

Abstract

We present a model-driven approach for the creation of formally verified scenarios involving human-robot interaction in healthcare settings. The work offers an innovative take on the application of formal methods to human modeling, as it incorporates physiology-related aspects. The model, based on the formalism of Hybrid Automata, includes a stochastic component to capture the variability of human behavior, which makes it suitable for Statistical Model Checking. The toolchain is meant to be accessible to a wide range of professional figures. Therefore, we have laid out a user-friendly representation format for the scenario, from which the full formal model is automatically generated and verified through the Uppaal tool. The outcome is an estimation of the probability of success of the mission, based on which the user can refine the model if the result is not satisfactory.

A Deployment Framework for Formally Verified Human-Robot Interactions

“…As shown in the diagram, we have identified two macro phases to analyze interaction scenarios: the designtime phase and the run-time phase. The first one is briefly summarised in the following (interested readers can refer to [5] and [6] for a thorough presentation).…”

“…Within the HA network, the orchestrator shares its commands with the human through channels identically to what it does with the robot. The unpredictability of human behavior is embedded in the formal model in terms of stochastic features [5]. Specifically, once the command has been received, edges are also labelled by probability weights that determine whether the human will obey or disobey.…”

“…We have developed a model-driven framework to analyze human-robot interaction scenarios [5], [6]. The methodology covers scenarios that feature: a battery-powered mobile robot, one or multiple humans that need to interact with the robot, and a closed environment.…”

RoboMAX: Robotic Mission Adaptation eXemplars

“…These systems are deployed in uncontrolled real-world environments, where the robots are exposed to multiple sources of uncertainty that can only be handled through self-adaptation. In many applications, the robots are additionally required to perform complex missions, increasingly in collaboration with other robots [2,3] or humans [4,5]. As such, their control software needs to perform monitoring, analysis, planning and execution activities that are the hallmark of self-adaptive systems.…”

“…The development of the control software for robotic systems is a recognized challenge that demands new software engineering methods and techniques [6], not least to achieve self-adaptation. This demand has increased the software engineering community's interest in understanding the needs of the robotics domain [7], and in developing solutions that address these needs [8,4]. In this context, a major difficulty faced by researchers and practitioners has been the scarcity of model problems [9] that can be readily used to guide research into, and to support the development of, such solutions.…”