Entertainment - Myth or Reality Alexander K. Seewald AustrianResearch Institute for Schottengasse3, A- 1010Wien, Austria [email protected],[email protected]

Abstract Actually, they resemble cats muchmore than dogs. Take This paper presents an overviewof the current field of the following scene: the struggles to walk towards entertainment based on experiencesas spectator the ball, eventually reaches it, lifts his paw,kicks the ball during the RoboCup1999 and building an experimental successfully into the goal - however, unfortunately it was based on the LEGOplatform with the wronggoal, even though the goal colors were blue vs. digital color cameraand various other sensors. RoboCatis yellow with an orange ball in front of a dark green a robot cat prototypethat showscat-like behaviourin the background, despite carefully controlled lighting real-life environmentof typical households.For behavioural modelling,the Hamsterdamarchitecture was chosen. While conditions and even black hoods for the humanplayers showing that Blumberg’s Hamsterdamoffers a new whohad to rearrange the robots about twice per minute programmingparadigm to design intelligent entertainment since they kept bumping into each other and into the robots, this paper also aims to decidewhether or not truly boundaries! It was still most impressing that somerobots intelligent entertainmentrobots are as of yet a myth. "died" in such a convincing, life-like way, showingerratic and spastic random movement before they froze and crashed completely. In virtual death they were most alive. Introduction Still, these robots offer somepromise: it is only a question First, let us consider the commercially active field of how far this hardware can go with the right software. entertainment robotics and some current developments in Unfortunately only and a few select scientific this field. Then we shall clarify the notion of an research groups are writing new programs. As of now, entertainment robot and introduce an architecture by there is no official programmingkit, just a kit to edit and (Blumberg 1997) to design entertainment robots in create new movementsequences. A short side-look at biologically plausible way, starting at an ethogram. eBaytells us that there are eight of the robots for sale on Nowadays strange things are afoot in the area of eBay, five of which were almost unused. The following entertainment robotics. We see a well-known Japanese quote is by an anonymousseller on eBay who clearly was firm sell as of now15,000 entertainment robots at a rather not satisfied with the return on investment of his AIBO. high price, whichnevertheless only reflects the integrated expensive hardware, with a very effective marketing policy Dadthought he had a great idea, but it didn’t work. creating a tenfold over-demand ~. We see upcoming Now, I’m selling my AIBObecause my kids want a approaches using wireless links to utilize the massive real dog... computingresources of today’s desktop computers to offer speech and face recognition. Wesee low-cost kits for robot While the idea of paying $2,500 for an electronic building which enable the technically inclined to create a product that doesn’t perform any useful function large set of toy robots for various purposes. Nonetheless, strikes us as strange now, the fact that the Aibo dog with all these achievements anyone shall be hard put to doesn’t do anything productive is central to the whole find a customer that can tell the difference betweena very idea of entertainment robots. Sony’s goal is not to expensive, state-of-the-art entertainment robot and a cheap create a robotic slave that will do your chores, but clone in terms of what it can and cannot do. So, from point rather an electronic companionthat will make you of view of the market, anybody can create an acceptable chuckle with an endearingturn of its head or a playful entertainment robot; it is the marketing that makes the pawswipe at a ball. difference. Whythen is this considered a scientific The previous quote by (Buskirk 1999) summarizesit quite question? Of course because we would like to have succintly: as of nowwe are unable to create a true robot something that customers CANdifferentiate from less able slave and this will be true for the near and medium-term products by its capabilities. This may or may not be future. In fact everyone would also be hard put to find a reasonableto expect in the near future - only time will tell. bipedal robot that can walk arbitrary stairs without Consider state-of-the-art robot technology the author knowingtheir height, length and position in advance. experienced as spectator at the RoboCup 1999 in So, while overselling the capabilities of entertainment Stockholm which featured a Sony legged-robot league. robots and massive marketing efforts distorts the view people have of entertainment robots, thus makingit harder to market an entertainment robot with realistically Copyright© 2001,American Association for Artificial Intelligence describedcapabilities, there still remainsthe question as to (www.aaai.org).All rightsreserved. how to design entertainment robots flexibly given the Atthe timeof writingSony has announced that it intendsto serveall technological restrictions - in order to once be able to ordersfrom now on, discontinuingthe so-calledAIBO-roulette.

5O4 FLAIRS-20Ol From: FLAIRS-01 Proceedings. Copyright © 2001, AAAI (www.aaai.org). All rights reserved. transcendend technological limitations. This paper will Background consider only intelligent entertainment robots, using the Sony’s AIBOis a four- with a head and following intelligence definition due to (Steels 1994). binocular vision and was programmedto chase and kick an Behaviour is intelligent if it maximisespreservation orange ball, to give paw, express joy, boredomand anger of the system in its environment. by responding to petting and hitting (1-bit feedback). Additionally they participated in the RoboCup1998 and For entertainment robots, the environment in question is continue to participate using various software usually a household. Intelligent entertainment robots are implementationsfrom three selected universities each year. thus those whoare able to keep their users interested for a Sony AIBOcombines a 180,000 pixel CCDcolor camera long period of time and are not thrown or stowed away, with a single point infrared distance sensor. The color never to be used again, like most other toys. camera is coupled with the CognaChromesystem orginally An intelligent entertainment robot by this paper’s from MITwhich thresholds the image to find blobs of definition has to be able to learn newthings or be updated color and returns the position of the biggest blobs. The so as to remain interesting. Anytoy with a fixed repertoire, AIBOchanges its personality only in discrete steps. however complex, is ultimately doomed. Twoapproaches triggered by an internal counter that is checked at boot to this can be seen on the market: robot kits (very time- time. It is not clear what the design paradigmof this robot intensive, but allow a wide variety of robots) and self- was, but it probably was Brooks’ subsumptionarchitecture learning robots such as the AIBO.The former is still hard (Brooks 1986). The robot can be taught by explicitly to build and program. Even a simple task can prove programming it new movements but can not learn by daunting to any robot, howeverwell designed, leaving the example. customerwith a strong feeling of failure. NEC’s R100 features voice keyword and face Self-learning is a promising approach but the maximum recognition, speech output and turns it head whenyou talk plateau that can be achieved this wayis rather low. Giving to it. It can control your household applicances such as a robot too muchfreedom to learn increases the probability 2, televisions and lights, notifies you of newemails and reads of radical screw-up which mayor maynot be considered them.~. It can also send video emails but no text based ones funny by the user. It may be noted that Sony’s AIBO The R100uses a wireless link to a desktop computer and seems to have twelve possible stages of development and uses offline computation. It is less of an entertaimnent only four distinctive adult stages - definitely not an unique robot than a personal, keyword-driven voice interface to personality for every one! your desktop computer. The idea of offloading Alternative methods not yet considered are computationwirelessly is a bit dated hut quite appropriate downloadablepersonalities and personality toolkits to set here. It is not clear what, if any, personality it has - by various parameters influencing the behaviour in continuous definition a slave should not have a personality at all. ways. For robots that are already wide-spread such as the AIBOit maybe a good idea to offer a programmingkit to design completely new personalities to volunteers and set RoboCat - Hardware up an exchange board on the internet. This way, many RoboCat is based on the LEGOplatform with EveBo! more personalities could be created and exchanged that controller board, digital color camermbumpers and bend would be possible if Sony kept the programminginterface sensors. RoboCatis intended to be a robot cat protol3~e under check. In any case there are already efforts under that shows cat-like behaviour in the real-life enviromneut way to create such a kit by volunteers. Another point is of a .typical household. There were two goals: to create an that a robot that is active for a longer time has morechance entertainment robot and to do so using a ethologically of doing something to surprise its owners. To simulate at plausible model, i.e: Hamsterdamwhich was previously least a 12h waking, 12h sleeping cycle or enabling the used~. for various virtual, computer-simulatedcreatures robot to reload himself, e.g, by exchanging an empty As personality metaphor for RoboCat the cat from accumulator pack with a full one instantly, would be a Whiskas, knownfrom various television advertisements, is rather simple wayto increase its ability to seemalive. taken to be a protot3qgically playful, very young and Summarizing, intelligent robots should have different inexperienced cat. This personality was taken as basis for robot personalities (e.g. by tweaking personality the implementation. During implementation and tweaking parameters or downloading new personalities), changing of personality parameters, various other personalities personalities over time (continuously, not in discrete appeared, e.g. a paranoically fearful cat that continues to steps!) and learning new behaviours or tricks. The moveback from an obstacle upon collision far longer than following will show that the Hamsterdamarchitecture reasonable - sometimes even until hitting the opposite (Blumberg1997) can accomodateall these points. wall.

3 This would need speech-independent flawless voice recognition of arbitrary phrases. i.e. of learning something unwanted,useless or just incomprehensible 4 most notably Silas T. Dogwhich could even be taught tricks similar to to humans;thus leaving the customer thinking "dumbrobot!". conditioning in real dogs, see (Blumberg1997).

NICHE AUTONOMOUSROBOTS 505 A design based on LEGOwas chosen because there were the author is mainly interested in human-robotobservation no other cheap robot hardwareplatforms yet available that and interaction. There are still some interesting offered sufficient flexibility for designinga cat-like robot. experiences from cat-robot confrontations although LEGOis cheap, flexible yet stable and allows fast complexsocial interaction does not emerge. rebuilding and prototyping. Sensors, actors and other Since RoboCathas no legs, only one posture element is devices can be attached by means of elastic and adhesive applicable. tape. It would also have been possible to use a Sony AIBO, ¯ STAND- Positioned with just four paws in contact but this was meantprimarily as low-cost approach. with the ground. Three kinds of TAIL MOVEMENTSwere implemented. SLAPwas not implementedsince the robot’s tail is unable to moveup or downso it cannot strike the ground. ¯ SWISH- A cat movesits whole tail rapidly from side to side ¯ TWITCH- A cat abruptly movespart of its tail from side to side or up and down ¯ QUIVER- A cat vibrates its tail while raising it vertically

FORWARD BACK Fig. 1: RoboCatprototype hardware F RoboCat plays with a hard blue ball that is found via L image recognition. In every image recognition cycle a 80x50 pixel frame is grabbed and then converted to A normalized RGB-scale. Then a box in color space is taken T to contain the pixels of the ball, the extent of which has E been empirically determined. All pixels of the image are then classified and from those considered to be part of the R ball the center of gravity is calculated, the result is the El presumedposition of the ball. CI

RoboCat- Software Fig.2: Cat faces with different ear shapes The behavioural structure and the motivational system was All four ear positions are implemented as a virtual created in four steps. simplified cat face shownon the integrated LCDscreen. 1. Specification ofbehaviour The ears can independently be shown BACKvs. 2. Reformulation in robot-centered terms (i.e. as seen FORWARDand FLATvs. ERECT,yielding four different from point of view of the robot) ear shapes, as shownin Fig. 2. Thevirtual cat face can also 3. Specification of needs 5 close its eyes andsniff its nose. 4. Design of the motivational system ¯ BACK- Ears are held at the rear of the head. First, the desired behaviours will be specified. These will ¯ FORWARD- Ears are held at the front of the head. then be reformulatedin teleological terms, i.e. specified in ¯ FLAT- A cat flattens its ears to its head such that a way the robot can understand. Afterwards a system of they tend to lie flush with the top of the head. needs will be designed such that the desired behaviour will ¯ ERECT- A cat points its ears upward. emerge. This design will lastly be mapped to the Four meaowsounds are implemented. Hamsterdam framework. ¯ MIOUW(MEOW) - A cat makes a distinct sound, usually whenit is trying to obtain something. Behaviour specification ¯ GROWL- A cat makes a low-pitched rumbling noise. Simulating cat-like behaviour is a challenging task. The ¯ YOWL- A cat makes a long drawn-out vocalization. paper (UK Cat Behaviour Working Group 1995) which ¯ PURR- A cat makes a low rhythmical tone from its contains an ethogramof the domestic cat is a good starting chest and throat, produced during both exhalation and point to select interesting behaviours. Let us restrict inhalation. behaviours to all solitary behaviour patterns and those social behaviour patterns concerning humanpartners since

5 Mysystem is based on the Hamsterdamframework from (Blumberg 1997).

506 FLAIRS-2001 avoiding collisions, whichmust be especially robust, since indicate that the robot had been pushed by an external the robots operate without supervision. It is notable that force), emergency-stopactivations, and the like. the collision avoidance code on these robots is by far the Like other aspects of the robot, and least changed over the course of their existence, vision evolved over the four robots. Chips used a specific confirming the functionality of the initial simple design. set of pink visual landmarks, one of which was three- The robots use ultrasonic range-finding sensors (sonars) dimensional, to provide corrections to simple encoder-only detect obstacles, and movearound them reactively, each methods for determining location. As we installed robots cycle choosing the appropriate motion vector to take based in more locations, we added new kinds of visual strictly on the most recently available sensor data, along landmarks, including sharp edges in intensity and with restrictions on howfar the robot is allowed to move rectangles of different color. Wealso added different out of its ideal (no-obstacle) trajectory. The code methods for using them, allowing multiple landmarks to be extremely simple, with no explicit mappingor modeling of tracked simultaneously (to deal with changing lighting the world or of the sensors themselves. It is also easy to conditions), and using landmarks to allow the robot to understand, and because of the lack of internal state, easy localize in more directions. Wealso used the same "try to debug (Nourbakhsh 2000). Because of the limited again" technique to make the landmark searching accuracy of sonar at close range, the robots will algorithm more robust. These changes are the subject of a occasionally becomestuck whenthey approach a wall too companionpaper being written concurrently. closely. Given the infrequency of this failure mode(less than twice per month), we feel the increased trust one can On Human-Robot Interaction have in the robot’s safety due to conservative motionto be Our second requirement was to deploy robots with worthwhile. compelling interactivity. As the science of Human-Robot There is a great deal more to autonomythan safety. A Interaction (HRI)is in its infancy, it is not surprising that robot must be able to interpret its own behavior, to the robot interaction componentwas entirely redesigned in determine whether or not it is functioning correctly. In each subsequent deployment. Even so, we have reached order for humansto be confident in its ability to run several qualitative conclusions, whichwe will discuss here. without supervision, a robot must be able to determine on An interview with the exhibits maintenancestaff of any its ownwhen a failure occurs. Early in the developmentof large museumwill drive home an important fact: people these platforms, we began using pagers, which the robot are basically destructive. Sometimesthis is purposeful can signal via electronic mail. The ability to recognize damage caused by malicious people. More frequently, failure and actively request help satisfies near-term curious individuals whoare trying to better understand the requirements for autonomy.Of course the ultimate goal is robot cause damage. For example, some will attempt to that the robot never needs to send for help at all, so self- push the robot off course to see if it will recover. Others repair becomesa second step to self-diagnosis. will push any large red emergencystop button to see what Initially, Chipssent for help quickly, giving up as soon happens. as a failure was detected. Soon we began adding Also, what attracts people varies greatly depending on diagnostic methods to reset subsystems that weren’t the context of a particular public space. Whenin an functioning correctly. This evolved into a general method "entertainment" space, such as a museum,people will be for autonomywithin our object-oriented architecture: every curious and attracted by new and unusual things. To that time a task is performedor an external piece of hardwareis end the physical appearance of the robot is very important. commanded,check the result for validity. If the result is But two other characteristics produce even better results: invalid, reset the device or situation and try again. motion and awareness. Whenthe robot is in motion, it Whendocking to recharge, for example, if the battery draws the greatest attention from nearby people. To voltage fails to rise whenthe robot believes it is plugged capitalize on this we madeAdam twitch and moveslightly in, the robot will reset the physical situation by backingout while delivering longer presentations. of the plug and into the hallway. Then, it will repeat the The most successful technique for attracting human docking attempt. This "try again" policy is effective in attention is for a robot to demonstrate an awareness of robotics because, although the code is deterministic, there human presence. Interactions between humans and is sufficient nondeterminismin the environment that the complex machines are typically initiated by humans. same code may have different outcomes. We have further Whena robot deliberately faces a person and says "Hello," refined this policy with the caveat that the failure modeof he or she is almost always both surprised and enthralled. an attempted task must be non-catastrophic for a retry to be In contrast to entertainment venues, more utilitarian possible. spaces such as shopping centers and office buildings elicit The robots have evolved to makeincreasing use of this far less pronouncedreactions. In these spaces people tend strategy, and nowdetect manyabnormal situations, many to have an agenda; they rush about and are less willing to of which are automatically corrected, including battery be side-tracked by a new and entertaining creature. Early overcharging and undercharging, frame grabber anomalies, indications show that some success is possible using a DVDplayer errors, bizarre encoder values (which would

516 FLAIRS-2001 Figure 2: Adam40-80

An Evolutionary Study The underlying goals of compelling interaction and Figure 1: Sweetlips maximal autonomy have remained constant throughout the The most recent Mobot robot, Adam 40-80, has creation of all four robots. However, each succeeding operated in a variety of venues, including the Republican robot has been the product of a complete re-design based National Convention, the Democratic National on lessons learned from the prior robots. Although some Convention, a shopping mall, the National Aviary and technical aspects have remained unchanged, such as the most recently the Pittsburgh International Airport (see Fig. programming language and robot chassis, virtually all else 2). Originally designed to promote Pittsburgh both has evolved in an effort to improve the autonomy and statewide and nationally, Adam’s charter is to engage interactivity of the robots. passers-by with both information and challenges such as a We are in the unique position of having an established trivia game. Instead of navigating a fixed tour route, Adam trajectory of real-world interactive social robots. Studying is also responsible for seeking out and approaching humans the evolution of this robot series promises to uncover in order to engage them most efficiently. Adam has valuable information for the young science of social operated in a total of 6 venues for approximately 21 days. robotics. In the following two sections we discuss the evolution of the autonomy and interactivity of the Mobot robots.

On Robot Autonomy The first requirement of a robot operating in a public space is safety, both for the general public and for the robot itself. At the heart of the matter is the robot’s method for

NICHE AUTONOMOUSROBOTS 515