Entertainment Robots - Myth or Reality Alexander K. Seewald AustrianResearchInstitute for Artificial Intelligence Schottengasse3, A- 1010Wien,Austria
[email protected],
[email protected] Abstract This paper presents an overviewof the current field of entertainmentrobotics based on experiencesas spectator during the RoboCup 1999 and building an experimental entertainment robot based on the LEGOplatform with digital color cameraand various other sensors. RoboCat is a robot cat prototypethat showscat-like behaviourin the real-life environment of typical households.Forbehavioural modelling,the Hamsterdam architecture was chosen. While showing that Blumberg’s Hamsterdamoffers a new programming paradigmto design intelligent entertainment robots, this paper also aimsto decidewhetheror not truly intelligent entertainment robotsare as of yet a myth. Introduction First, let us consider the commercially active field of entertainment robotics and some current developments in this field. Then we shall clarify the notion of an entertainment robot and introduce an architecture by (Blumberg 1997) to design entertainment robots in biologically plausible way, starting at an ethogram. Nowadays strange things are afoot in the area of entertainment robotics. Wesee a well-known Japanese firm sell as of now15,000 entertainment robots at a rather high price, whichnevertheless only reflects the integrated expensive hardware, with a very effective marketing policy creating a tenfold over-demand ~. We see upcoming approaches using wireless links to utilize the massive computingresources of today’s desktop computers to offer speech and face recognition. Wesee low-cost kits for robot building which enable the technically inclined to create a large set of toy robots for various purposes. Nonetheless, with all these achievements anyone shall be hard put to find a customer that can tell the difference betweena very expensive, state-of-the-art entertainment robot and a cheap clone in terms of what it can and cannot do. So, from point of view of the market, anybody can create an acceptable entertainment robot; it is the marketing that makes the difference. Whythen is this considered a scientific question? Of course because we would like to have something that customers CAN differentiate from less able products by its capabilities. This may or may not be reasonableto expect in the near future - only time will tell. Consider state-of-the-art robot technology the author experienced as spectator at the RoboCup 1999 in Stockholm which featured a Sony legged-robot league.
Copyright© 2001,American Associationfor Artificial Intelligence (www.aaai.org). Allrightsreserved. Atthe timeof writingSonyhasannounced that it intends to serveall ordersfrom nowon, discontinuing the so-calledAIBO-roulette. 5O4
FLAIRS-20Ol
Actually, they resemble cats muchmore than dogs. Take the following scene: the robot struggles to walk towards the ball, eventually reaches it, lifts his paw,kicks the ball successfully into the goal - however,unfortunately it was the wronggoal, even though the goal colors were blue vs. yellow with an orange ball in front of a dark green background, despite carefully controlled lighting conditions and even black hoods for the humanplayers whohad to rearrange the robots about twice per minute since they kept bumping into each other and into the boundaries! It was still most impressing that somerobots "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. Still, these robots offer somepromise:it is only a question how far this hardware can go with the right software. Unfortunately only Sony and a few select scientific research groups are writing new programs. As of now, there is no official programming kit, just a kit to edit and create new movementsequences. A short side-look at eBaytells us that there are eight of the robots for sale on eBay, five of which were almost unused. The following quote is by an anonymousseller on eBay who clearly was not satisfied with the return on investmentof his AIBO. Dadthought he had a great idea, but it didn’t work. Now, I’m selling myAIBObecause my kids want a real dog... While the idea of paying $2,500 for an electronic product that doesn’t perform any useful function strikes us as strange now, the fact that the Aibo dog doesn’t do anything productive is central to the whole idea of entertainment robots. Sony’s goal is not to create a robotic slave that will do your chores, but rather an electronic companionthat will make you chuckle with an endearingturn of its head or a playful pawswipeat a ball. The previous quote by (Buskirk 1999) summarizesit quite succintly: as of nowwe are unable to create a true robot slave and this will be true for the near and medium-term future. In fact everyone wouldalso be hard put to find a bipedal robot that can walk arbitrary stairs without knowingtheir height, length and position in advance. So, while overselling the capabilities of entertainment robots and massive marketing efforts distorts the view people have of entertainment robots, thus makingit harder to market an entertainment robot with realistically describedcapabilities, there still remainsthe question as to how to design entertainment robots flexibly given the technological restrictions - in order to once be able to
From: FLAIRS-01 Proceedings. Copyright © 2001, AAAI (www.aaai.org). All rights reserved.
transcendend technological limitations. This paper will consider only intelligent entertainment robots, using the following intelligence definition due to (Steels 1994). Behaviour is intelligent if it maximisespreservation of the systemin its environment. For entertainment robots, the environment in question is usually a household. Intelligent entertainment robots are thus those whoare able to keep their users interested for a long period of time and are not thrown or stowed away, never to be used again, like most other toys. An intelligent entertainment robot by this paper’s definition has to be able to learn newthings or be updated so as to remaininteresting. Anytoy with a fixed repertoire, however complex, is ultimately doomed. Twoapproaches to this can be seen on the market: robot kits (very timeintensive, but allow a wide variety of robots) and selflearning robots such as the AIBO.The former is still hard to build and program. Even a simple task can prove daunting to any robot, howeverwell designed, leaving the customerwith a strong feeling of failure. Self-learning is a promising approach but the maximum plateau that can be achieved this wayis rather low. Giving a robot too muchfreedomto learn increases the probability of radical screw-up2, which mayor maynot be considered funny by the user. It may be noted that Sony’s AIBO seems to have twelve possible stages of development and only four distinctive adult stages - definitely not an unique personality for every one! Alternative methods not yet considered are downloadablepersonalities and personality toolkits to set various parameters influencing the behaviour in continuous 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 up an exchange board on the internet. This way, many more personalities could be created and exchanged that would be possible if Sony kept the programminginterface under check. In any case there are already efforts under way to create such a kit by volunteers. Another point is that a robot that is active for a longer time has morechance of doing something to surprise its owners. To simulate at least a 12h waking, 12h sleeping cycle or enabling the robot to reload himself, e.g, by exchanging an empty accumulator pack with a full one instantly, would be a rather simple wayto increase its ability to seemalive. Summarizing, intelligent robots should have different robot personalities (e.g. by tweaking personality parameters or downloading new personalities), changing personalities over time (continuously, not in discrete steps!) and learning new behaviours or tricks. The following will show that the Hamsterdamarchitecture (Blumberg1997) can accomodateall these points.
i.e. of learning something unwanted,useless or just incomprehensible to humans;thus leaving the customer thinking "dumbrobot!".
Background Sony’s AIBOis a four-legged robot with a head and binocular vision and was programmedto chase and kick an orange ball, to give paw, express joy, boredomand anger by responding to petting and hitting (1-bit feedback). Additionally they participated in the RoboCup1998 and continue to participate using various software implementationsfrom three selected universities each year. Sony AIBOcombines a 180,000 pixel CCDcolor camera with a single point infrared distance sensor. The color camera is coupled with the CognaChrome system orginally from MITwhich thresholds the image to find blobs of color and returns the position of the biggest blobs. The AIBOchanges its personality only in discrete steps. triggered by an internal counter that is checked at boot time. It is not clear what the design paradigmof this robot was, but it probably was Brooks’ subsumptionarchitecture (Brooks 1986). The robot can be taught by explicitly programming it new movements but can not learn by example. NEC’s R100 features voice keyword and face recognition, speech output and turns it head whenyou talk to it. It can control your household applicances such as televisions and lights, notifies you of newemails and reads ~. them. It can also send video emails but no text based ones The R100uses a wireless link to a desktop computer and uses offline computation. It is less of an entertaimnent robot than a personal, keyword-drivenvoice interface to your desktop computer. The idea of offloading computationwirelessly is a bit dated hut quite appropriate here. It is not clear what, if any, personality it has - by definition a slave should not havea personality at all. RoboCat - Hardware RoboCat is based on the LEGOplatform with EveBo! controller board, digital color camermbumpers and bend sensors. RoboCatis intended to be a robot cat protol3~e that showscat-like behaviour in the real-life enviromneut of a .typical household. There were two goals: to create an entertainment robot and to do so using a ethologically plausible model, i.e: Hamsterdamwhich was previously ~. used for various virtual, computer-simulatedcreatures As personality metaphor for RoboCat the cat from Whiskas, knownfrom various television advertisements, is taken to be a protot3qgically playful, very young and inexperienced cat. This personality was taken as basis for the implementation. During implementation and tweaking of personality parameters, various other personalities appeared, e.g. a paranoically fearful cat that continues to moveback from an obstacle upon collision far longer than reasonable - sometimes even until hitting the opposite wall. 3 This would need speech-independent flawless voice recognition of arbitrary phrases. 4 most notably Silas T. Dogwhich could even be taught tricks similar to conditioning in real dogs, see (Blumberg1997).
NICHE AUTONOMOUS ROBOTS 505
A design based on LEGOwas chosen because there were no other cheap robot hardwareplatforms yet available that offered sufficient flexibility for designinga cat-like robot. LEGOis cheap, flexible yet stable and allows fast rebuilding and prototyping. Sensors, actors and other devices can be attached by means of elastic and adhesive tape. It wouldalso have been possible to use a Sony AIBO, but this was meantprimarily as low-cost approach.
the author is mainly interested in human-robotobservation and interaction. There are still some interesting experiences from cat-robot confrontations although complexsocial interaction does not emerge. Since RoboCathas no legs, only one posture element is applicable. ¯ STAND - Positioned with just four paws in contact with the ground. Three kinds of TAIL MOVEMENTS were 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 Fig. 1: RoboCatprototype hardware RoboCat plays with a hard blue ball that is found via image recognition. In every image recognition cycle a 80x50 pixel frame is grabbed and then converted to normalized RGB-scale. Then a box in color space is taken to contain the pixels of the ball, the extent of which has been empirically determined. All pixels of the image are then classified and from those considered to be part of the ball the center of gravity is calculated, the result is the presumedposition of the ball.
RoboCat- Software The behavioural structure and the motivational system was created in four steps. 1. Specification ofbehaviour 2. Reformulation in robot-centered terms (i.e. as seen from point of view of the robot) 3. Specification of needs 4. 5Design of the motivational system First, the desired behaviours will be specified. Thesewill then be reformulatedin teleological terms, i.e. specified in a way the robot can understand. Afterwards a system of needs will be designed such that the desired behaviour will emerge. This design will lastly be mapped to the Hamsterdam framework. Behaviour specification Simulating cat-like behaviour is a challenging task. The paper (UK Cat Behaviour Working Group 1995) which contains an ethogramof the domestic cat is a good starting point to select interesting behaviours. Let us restrict behaviours to all solitary behaviour patterns and those social behaviour patterns concerning humanpartners since 5 Mysystemis basedon the Hamsterdam framework from(Blumberg 1997). 506
FLAIRS-2001
BACK
F L A T E R El CI Fig.2: Cat faces with different ear shapes All four ear positions are implemented as a virtual simplified cat face shownon the integrated LCDscreen. The ears can independently be shown BACKvs. FORWARD and FLATvs. ERECT,yielding four different ear shapes, as shownin Fig. 2. Thevirtual cat face can also close its eyes andsniff its nose. ¯ BACK - Ears are held at the rear of the head. ¯ FORWARD - Ears are held at the front of the head. ¯ FLAT- A cat flattens its ears to its head such that they tend to lie flush with the top of the head. ¯ ERECT - A cat points its ears upward. Four meaowsounds are implemented. ¯ MIOUW (MEOW)- A cat makes a distinct sound, usually whenit is trying to obtain something. ¯ GROWL - A cat makes a low-pitched rumbling noise. ¯ YOWL - A cat makes a long drawn-out vocalization. ¯ PURR- A cat makes a low rhythmical tone from its chest and throat, produced during both exhalation and inhalation.
avoiding collisions, whichmust be especially robust, since the robots operate without supervision. It is notable that the collision avoidance code on these robots is by far the least changed over the course of their existence, confirming the functionality of the initial simple design. The robots use ultrasonic range-finding sensors (sonars) detect obstacles, and movearound them reactively, each cycle choosing the appropriate motion vector to take based strictly on the most recently available sensor data, along with restrictions on howfar the robot is allowed to move out of its ideal (no-obstacle) trajectory. The code extremely simple, with no explicit mappingor modelingof the world or of the sensors themselves. It is also easy to understand, and because of the lack of internal state, easy to debug (Nourbakhsh 2000). Because of the limited accuracy of sonar at close range, the robots will occasionally becomestuck whenthey approach a wall too closely. Given the infrequency of this failure mode(less than twice per month), we feel the increased trust one can have in the robot’s safety due to conservative motionto be worthwhile. There is a great deal more to autonomythan safety. A robot must be able to interpret its own behavior, to determine whether or not it is functioning correctly. In order for humansto be confident in its ability to run without supervision, a robot must be able to determine on its ownwhena failure occurs. Early in the developmentof these platforms, we began using pagers, which the robot can signal via electronic mail. The ability to recognize failure and actively request help satisfies near-term requirements for autonomy.Of course the ultimate goal is that the robot never needs to send for help at all, so selfrepair becomesa secondstep to self-diagnosis. Initially, Chipssent for help quickly, giving up as soon as a failure was detected. Soon we began adding diagnostic methods to reset subsystems that weren’t functioning correctly. This evolved into a general method for autonomywithin our object-oriented architecture: every time a task is performedor an external piece of hardwareis commanded, check the result for validity. If the result is invalid, reset the device or situation and try again. Whendocking to recharge, for example, if the battery voltage fails to rise whenthe robot believes it is plugged in, the robot will reset the physical situation by backingout of the plug and into the hallway. Then, it will repeat the docking attempt. This "try again" policy is effective in robotics because, although the code is deterministic, there is sufficient nondeterminismin the environment that the same code may have different outcomes. Wehave further refined this policy with the caveat that the failure modeof an attempted task must be non-catastrophic for a retry to be possible. The robots have evolved to makeincreasing use of this strategy, and nowdetect manyabnormal situations, many of which are automatically corrected, including battery overcharging and undercharging, frame grabber anomalies, DVDplayer errors, bizarre encoder values (which would
516
FLAIRS-2001
indicate that the robot had been pushed by an external force), emergency-stopactivations, and the like. Like other aspects of the robot, robot navigation and vision evolved over the four robots. Chips used a specific set of pink visual landmarks, one of which was threedimensional, to provide corrections to simple encoder-only methods for determining location. As we installed robots in more locations, we added new kinds of visual landmarks, including sharp edges in intensity and rectangles of different color. Wealso added different methodsfor using them, allowing multiple landmarks to be tracked simultaneously (to deal with changing lighting conditions), and using landmarks to allow the robot to localize in more directions. Wealso used the same "try again" technique to make the landmark searching algorithm more robust. These changes are the subject of a companionpaper being written concurrently. On Human-Robot Interaction Our second requirement was to deploy robots with compelling interactivity. As the science of Human-Robot Interaction (HRI)is in its infancy, it is not surprising that the robot interaction componentwas entirely redesigned in each subsequent deployment. Even so, we have reached several qualitative conclusions, whichwe will discuss here. An interview with the exhibits maintenancestaff of any large museumwill drive home an important fact: people are basically destructive. Sometimesthis is purposeful damage caused by malicious people. More frequently, curious individuals whoare trying to better understandthe robot cause damage. For example, some will attempt to push the robot off course to see if it will recover. Others will push any large red emergencystop button to see what happens. Also, what attracts people varies greatly depending on the context of a particular public space. Whenin an "entertainment" space, such as a museum,people will be curious and attracted by new and unusual things. To that end the physical appearanceof the robot is very important. But two other characteristics produce even better results: motion and awareness. Whenthe robot is in motion, it draws the greatest attention from nearby people. To capitalize on this we madeAdamtwitch and moveslightly while delivering longer presentations. The most successful technique for attracting human attention is for a robot to demonstrate an awareness of human presence. Interactions between humans and complex machines are typically initiated by humans. Whena robot deliberately faces a person and says "Hello," he or she is almost alwaysboth surprised and enthralled. In contrast to entertainment venues, more utilitarian spaces such as shopping centers and office buildings elicit far less pronouncedreactions. In these spaces people tend to have an agenda; they rush about and are less willing to be side-tracked by a new and entertaining creature. Early indications show that some success is possible using a
Figure 2: Adam40-80
An Evolutionary Figure 1: Sweetlips The most recent Mobot robot, Adam 40-80, has operated in a variety of venues, including the Republican National Convention, the Democratic National Convention, a shopping mall, the National Aviary and most recently the Pittsburgh International Airport (see Fig. 2). Originally designed to promote Pittsburgh both statewide and nationally, Adam’s charter is to engage passers-by with both information and challenges such as a trivia game. Instead of navigating a fixed tour route, Adam is also responsible for seeking out and approaching humans in order to engage them most efficiently. Adam has operated in a total of 6 venues for approximately 21 days.
Study
The underlying goals of compelling interaction and maximal autonomy have remained constant throughout the creation of all four robots. However, each succeeding robot has been the product of a complete re-design based on lessons learned from the prior robots. Although some technical aspects have remained unchanged, such as the programming language and robot chassis, virtually all else has evolved in an effort to improve the autonomy and interactivity of the robots. We are in the unique position of having an established trajectory of real-world interactive social robots. Studying the evolution of this robot series promises to uncover valuable information for the young science of social 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 AUTONOMOUS ROBOTS
515
