Proceedings of Australasian Conference on Robotics and Automation, 3-5 Dec 2012, Victoria University of Wellington, New Zealand. How People Naturally Describe Robot Behaviour J. P. Diprose & B. Plimmer B. A. MacDonald J. G. Hosking Department of Computer Science Department of Electrical & College of Engineering & University of Auckland Computer Engineering Computer Science New Zealand University of Auckland Australian National University [email protected] New Zealand Australia [email protected] [email protected] [email protected] Abstract behaviour based on an analysis of healthcare robots. Our end goal is to create an end user robot program- Existing novice robot programming systems are ming environment, initially for healthcare professionals complex, which ironically makes them unsuit- who do not have programming experience, and eventu- able for novices. We have analysed 19 reports ally for end users in other robot domains. However, a of robot projects to inform development of an prerequisite to this is to understand how people natu- ontology of critical concepts that end user robot rally describe robot behaviour. This involves collecting programming environments must include. This peoples descriptions of healthcare robot behaviour and is a first step to simpler end user robot pro- analysing them to uncover the key building blocks and gramming systems. relationships used in these descriptions. There are two main options for collecting data where 1 Introduction people describe healthcare robot behaviour: a case study As robots become more pervasive in human environ- based approach and a literature search. Given time and ments, there will be a plethora of new task demands resource considerations, a case study based approach on them. As a result it will be necessary for end users would only be able to gather data for a very small num- to be able to program new tasks in the field, even when ber of cases, meaning the results would only be applica- libraries and components are programmed in the factory. ble to the healthcare robots examined in the case study. However, there is no well established ontology of robot This is not ideal as our goal is to create an end user robot behaviour on which to base such end user programming programming system for the wider healthcare domain. environments. Instead, a literature search approach was chosen as it The most closely related works are ontologies designed can examine a larger variety of robots and the analysis to represent robotic systems [Schlenoff and Messina, of the data is likely to be more generalisable. The col- 2005; Ros et al., 2010; Paull et al., 2012; Balakirsky lection of sources describing healthcare robot behaviour et al., 2012; Varadarajan and Vincze, 2012; Schlenoff et is discussed in Section 2. al., 2012; Tenorth and Beetz, 2012], the most notable An approach based on grounded theory [Glaser and of which is a new IEEE working group formed to create Strauss, 1967] was chosen to analyse the qualitative de- the IEEE Standard Ontology for Robotics and Automa- scriptions of robot behaviour found by the literature tion (SORA) [Schlenoff et al., 2012]. SORA aims to search. Grounded theory is a method widely used in represent every part of a robotic system; from low level social science research to discover theory from system- devices, such as sensors and actuators; to the knowledge atically collected data [Glaser and Strauss, 1967]. The required for a robot to solve a problem, such as domain researcher builds up a theory by iteratively identifying specific objects, like a mine object for a mine hunting concepts and relationships that fit the underlying data. robot [Paull et al., 2012]. The problem with using these Eventually a conceptualisation of the data emerges. This ontologies as the sole representation for end user pro- allows us to create an ontology that is easy for end users gramming systems is that they are not being explicitly to understand, because we can use the methodology to designed for ease of understand by end users, which is explicitly generate our theory from descriptions of robot what we aim to address. behaviour that end users are expected to find compre- The objective of this study is to develop an ontology hensible. suitable for end users, that can be used to describe robot Grounded theory focusses on generation of theory, as Proceedings of Australasian Conference on Robotics and Automation, 3-5 Dec 2012, Victoria University of Wellington, New Zealand. opposed to verification of theory [Glaser and Strauss, Is physically embodied. 1967]; because of this, we have focussed on generating Makes conscious decisions (in a limited sense). our theory rather than verifying it at this stage. This doesn't mean that our theory is without evidence, in fact, Treats or cares for a person to benefit their health. Glaser and Strauss [1967] argue that theory based on Healthcare robots were excluded where: data is hard to totally disprove because it is thoroughly The \robot" is a virtual avatar. connected to the data. It is likely to last even though it will certainly be refactored in the future [Glaser and A human controls the higher level behaviour of the Strauss, 1967]. The analysis of the qualitative data and robot. the ontology it generated are discussed in Section 3. The robot treats a person using invasive surgery. Conclusions are presented in Section 4. Physical embodiment is a characteristic of robots that makes them unique compared with generic software sys- 2 Literature Search tems, hence the requirement that the \robot" must be a The design of a general robot programming system needs real physical entity (ruling out virtual avatars). to be informed by many examples of robot behaviour. The capabilities of existing robots may not meet Fortunately, many research groups have applied robots users expectations because technology has not advanced to the healthcare domain, so there is much literature enough yet. For example, a user might expect AIBO, a describing healthcare robots and their behaviour. This robot dog, to be able to sense touch all over its body. afforded us the opportunity to conduct a systematic lit- However, this technology hasnt been implemented in erature search for articles describing healthcare robot AIBO, let alone any robot dog. If we only included real behaviour. We expect this analysis to also reveal a sig- robots in our systematic search our ontology would dis- nificant amount of generic robot behaviour. count such natural expectations of users. To ensure that A systematic literature search is a formalised approach these natural expectations are represented in our ontol- to searching literature [Kitchenham, 2004]. It focuses on ogy, our definition of robot also allows fake robots to be a research question, inclusion and exclusion criteria to included. For instance, a robot could be a robot oper- construct a search query, a corpus of work to search, ated in a \wizard of oz" fashion (as is the case for most and a process for quality assessment and data selection. Autism robots) or even be a person acting like a robot for Subsection 2.1 describes the systematic literature search the purpose of discovering how people naturally interact methodology and subsection 2.2 presents the systematic with robots. literature search results. Search process 2.1 Methodology Several databases that are known to contain information The methodology for systematically searching literature related to healthcare robots were searched. These are: used in this study is based on guidelines provided by ACM Digital Library [ACM, 2012], IEEEXplore [IEEE, Wolfswinkel et al. [2011] and Kitchenham [2004]. The 2012], SpringerLink [Springer, 2012] and ScienceDirect goal of our systematic literature search is to obtain de- [Elsevier, 2012]. scriptions of healthcare robot behaviour. The steps in The search terms were based on the inclusion and ex- the systematic literature search are discussed below. clusion criteria defined earlier. The general structure of the query can be seen in Figure 1. The first part of Research questions the query contains several common synonyms for robot The research question addressed by the literature search (humanoid and android). The second part of the query is: contains several synonyms for healthcare (care, treat- 1. What healthcare robots have been created to date ment and words beginning with help). and what behaviour do they exhibit? The research question is concerned with finding as Figure 1: General structure of database search terms. many healthcare robots and sources describing their be- (robot OR humanoid OR android) AND (care haviour as possible. The goal of this question is to ob- OR treatment OR therapy OR healthcare OR tain a broad collection of material describing healthcare help*) robot behaviour that can be used in the analysis stage of the study (Section 3). Inclusion and exclusion criteria Quality assessment Healthcare robots were included if they discussed a robot Quality assessment was not important for this literature that: search. Even if a robotics project is poor quality, it does Proceedings of Australasian Conference on Robotics and Automation, 3-5 Dec 2012, Victoria University of Wellington, New Zealand. not mean it cannot contribute to our understanding of Table 1: Healthcare robots how people describe robot behaviour. On the contrary, it is desirable to be able to describe the behaviour of any Category Robot Reports Analysed robot, regardless of whether it resulted in a high quality Children AIBO [Robins et al., 2005b] scientific study or not. The only quality requirement was Kaspar [Amirabdollahian to exclude articles that were solely abstracts, because et al., 2011; Dautenhahn et they lack sufficient detail to be useful. al., 2009] Keepon [Kozima et al., 2007] Data selection Nao [Shamsuddin et al., 2012] During the data selection stage: the list of healthcare Zeno [Ranatunga et al., 2011] Robota [Robins et al., 2005a] robots was compiled and sources describing robot be- Elderly ASIMO [Honda, n.d.; Sakagami et haviour were selected.