DEVELOPMENT OF A NURTURANCE EVOKING ROBOT

Transcription

1 10th International DAAAM Baltic Conference "INDUSTRIAL ENGINEERING" th May 2015, Tallinn, Estonia DEVELOPMENT OF A NURTURANCE EVOKING ROBOT Peltonen, O.; Orhanen, S.; Venäläinen, J.; Auvinen, M.; Sepponen, R.; Kiviluoma, P. & Kuosmanen, P. Abstract: A robotic companion can have the same health benefits as pets, including alleviating loneliness, lowering stress and elevating mood. The non-allergenic and controllable nature of robots makes them preferable over real animals. This paper describes a nurturance evoking social robot for eldercare, which is able to express its emotions and act like a real animal, is robust and fits in user s lap. None of the existing social robots combine all these criteria. The robot expresses emotions with actuated neck, ears, nose and tail, vibration and sound. Flexible mechanisms ensure safe movements and durability. Touch, motion, temperature and sound affect the robot s emotional state. Preliminary studies show good results in evoking nurturance. Key words: social robot, companion robot, eldercare, artificial intelligence. 1. INTRODUCTION In healthcare, a robotic companion can have the same health benefits as pets, including alleviating loneliness, lowering stress and mood elevation [ 1,2,3 ]. Although robotic and real pets have the same therapeutic effect, a robot is often preferred in healthcare, because it is not allergenic. Some patients might also be incapable of taking care of a real pet. Thus the need for social robots increases constantly, especially in the field of eldercare. A study shows that more than one third of elderly people suffer from loneliness [ 4 ], and the population in western countries is continuously ageing. It has been estimated that the proportion of people over 60 years old will be 37% of the whole population in Western Europe by 2050 [ 5 ]. There currently exists multiple commercial and non-commercial social robots with applications varying from eldercare to entertainment. These robots include Paro, Huggable and Probo. Paro is a commercially available artificial pet seal. Under the soft hygienic artificial fur, Paro features a range of sensors, which allow it to sense touch and warmth, recognise its name, greetings and praise. Paro has actuators to move its eyelids, head and flippers and a speaker to play authentic seal sounds. Paro s behaviour alters to reflect user s interaction preferences. [ 6 ] MIT s Media Lab is developing Huggable, a therapeutic robot companion with its whole body able to sense touch. To achieve touch sensing, Huggable utilises three sensor types all around the body: Quantum Tunnelling Composite (QTC) for force, thermistors for temperature and electric field sensing to determine for example if the object touching the robot is human. For motion, Huggable uses voice coil actuators, which provide smooth, soundless, lifelike motion. Huggable also utilises neural network to determine the nature of the touch. [ 1 ] Probo is a mammoth-like robot developed to be a friend for hospitalised children. It features a fully actuated head with a moving trunk and natural, motion tracking eyes [ 7 ]. To achieve safe movement, Probo utilises flexible joints and Bowden cables between the motors and the actuated joints. In addition to safety, the use of Bowden

2 cables allows the motors to be placed anywhere in the robot, thus keeping the head light-weighted. [ 8 ] In this paper, the development of a Social Panda robot (Fig. 1.) is presented. The goal of this research was to develop a robot that keeps company with elderly by evoking nurturance by the means of motion and voice. The robot had the following four requirements: it must be able to express its emotional state; it must contain an artificial intelligence (AI) to allow it to act like a lifelike animal; it must fit in the user s lap and be robust. All of these separate requirements have been met by the previously mentioned robots, but none of them combines all four criteria. This paper introduces a design of a social robot that combines all four aforementioned requirements. Fig. 1. Appearance and features of the panda robot. 2. METHODS 2.1 Overview The approach to make the robot fit one s lap, be expressive and to have lifelike artificial intelligence sets certain limitations and requirements for the design. For one, the hardware inside the robot has to be lightweight and small, selected actuators have to be able to accurately drive the used mechanisms to make the robot s expressions distinctive. Secondly the software design has to take human interpretations into account. To meet the set requirements, the robot is able to move its head, ears, nose and tail, emit sound and vibrate to mimic purr of cats. The feeling states of the robot will be expressed with combinations of these outputs, and slight randomness is added to the AI to achieve a lifelike animal behaviour. The robot was chosen to look like a panda for two reasons. Firstly, a study shows that it is psychologically beneficial to make the robot look like an animal that most people don t have experience on [ 6 ]. This choice of design is advantageous, because if users have no knowledge on how the robot should behave, they don t lose interest in it due to unsatisfied expectations. Secondly, pandas are commonly considered as sympathetic. A 19 cm x 24 cm x 30 cm WWF plush panda was selected as the exterior of the robot. 2.2 Mechanical Design The first prototype of the robot features 6 degrees of freedom (DOF) to obtain a lifelike and nurturance evoking appearance. The robot is able to move its ears (2 DOF), move its nose (1 DOF), tilt its head (2 DOF) and wiggle its tail (1 DOF). These movements are used to express the feeling state of the robot. The prototype is powered with an external power supply. To achieve safe movements and durability, flexible joints and mechanisms were used. The neck joint was manufactured from a coil spring and polycarbonate connecting plates. The joint is actuated by pulling monofil nylon fishing lines with RC servo motors (MG90S). Fishing lines operated with RC servo motors were also used to operate the mechanisms for the nose and ears. RC servo motors were chosen as main actuators for their affordable price, light weight, size and controllability. The skeleton of the robot consists of two modules, the head and the body. Main parts of the skeleton are shown in the Figure 2. The function of the skeleton is to house and protect the electronics and serve as a contact surface for the touch sensors. The skeleton was 3D-printed, which allows great freedom of design and produces parts quickly. The skeleton was designed as a

3 tight fit inside the plush panda so that it does not require any external attachment to the cloth exterior. Fig. 2. Illustration of the two-module skeleton. 2.3 Electronics and Sensors Arduino was chosen as the platform for the electronics, as it is well suited for prototyping: it features easy and fast coding, and the market is full of Arduino compatible modules. Arduino Nano V3 was chosen for its small size in order to fit in the robot. The robot s touch sensing was achieved by placing sensors made of Quantum Tunnelling Composite (QTC, Peratech) pills on the surface of the skeleton. QTC material changes its electrical resistance when pressure is applied, thus making it suitable for touch sensors. QTC sensors were chosen for their affordable price and small size. Thermistors were used to detect the warmth of the holding human, other possible temperature changes and to detect possible electrical failures. A passive infrared (PIR, HC-SR501) sensor module was used for sensing motion in front of the panda. Motion and orientation of the robot is detected by an IMU module (GY-85). Servo shield F08019 controls the 6 servos. The robot makes noises from a speaker, driven by a sound module (WTV020-SD) with audio stored on a SD-card. One eccentric rotating mass (ERM) motor (RF300) is used to create the purr vibration, purr effect is enhanced with corresponding sounds. In order for the robot to be useful in eldercare or as a scientific tool for studies, one has to be able to monitor the functioning and usage of the robot. To achieve this, the monitoring data is sent by a serial-interface WiFi-module (ESP8266) to a monitoring computer. 2.4 Software Architecture The architecture shown in Figure 3 has a hierarchical structure where high-level computation is executed on the top and, on the contrary, low-level computation on the bottom side. For example, software classes that represent different moods of the robot are on the top-side of the diagram, because they are more abstract concepts than, e.g., motor controller that lie on the bottomside. This kind of hierarchical software structure is common in robotics and software design in general, because it offers two main advantages. Firstly, the designer of the software has strong understanding of ownerships of instances. Secondly, it is modular, as modifications made to classes in one hierarchical level do not affect the whole software. This is because each level works as an interface to the next level, which effectively hides all the modifications from further levels. In fact, modifications to one level will only affect the levels next to it. The main loop that keeps the software running resides in the State machine level of the hierarchy. As the name of the level

4 implies, it also contains a state machine that forms the AI of the panda robot. The transition logic between states is implemented with hard-coded control structures. The mood-level includes the sequences of actions that create the characteristic behaviours of each mood. The mood-level uses the high-level hardware communication layer (HCL) that is an abstraction level of the hardware controllers. This high-level HCL provides simplified interface to the hardware. Finally, the low-level HCL communicates with the hardware, e.g., with motors and sensors via various communication protocols. The software is implemented with C++ programming language, because our hardware is Arduino-based and there exist plenty of ready-to-use libraries to control various devices in C User Survey The robot was tested with a user survey to investigate whether the robot fulfils the set goal, i.e., whether or not it can keep company with elderly. The participants of the survey used the robot for approximately 5 to 10 minutes each. Afterwards the participants took a questionnaire about the robot with 6 questions, which were answered on a three point scale: Disagree, No opinion and Agree. The questions asked in the user poll are described in Table 1. Number of Question Question I understood if the robot was happy or sad. The robot reacts to my care in a reasonable way. The robot wanted my attention. I enjoyed the company of the robot. The robot was easy to hold in lap. 6 The robot felt alive. Table 1. User survey questions. 3. RESULTS The user survey was conducted with elderly people in a retirement home called Kustaankartano comprehensive service centre situated in Helsinki, Finland. A total of 8 elderly people participated in the survey and answered to the questionnaire shown in Table 1. The results of the questionnaire are presented in Table 2. The age of the participants varied between 70 and 97, the mean being 85. Fig. 3. Hierarchical software architecture used in the panda robot. High-level computation is executed at the top-side and low-level computation on the bottom side.

5 Number of Question Agree [%] Disagree [%] No opinion [%] Number of participants 8 Table 2. User survey results. Question number is shown in the first column, see Table 1 for questions. In the second, third and fourth columns the distribution of users answers is shown. 4. DISCUSSION The following four requirements were set for the robot: it must be able to express its emotional state; it must contain an artificial intelligence to allow it to act like a real animal; it must fit in the user s lap and be robust. The results of the user survey indicate that most of the users were able to identify the emotional state (Question 1, Table 2) of the robot and the participants felt that the robot wanted their attention (Question 3, Table 2). The robot also fit in the lap of every user and was easy to hold (Question 5, Table 2), even though several of the survey participants had only one functional hand. The results also show that although most of the users felt that the robot was alive (Question 6, Table 2), only few of the participants felt that the robot reacted to nurturing properly (Question 2, Table 2). This suggests that the AI of the robot does not meet the set requirement. No specific experiment was conducted to determine the robustness of the robot. Nevertheless, tens of people have handled the robot using varying force without breaking it proving that the robot is robust. The robot fulfils most of the set requirements and the survey indicates that the robot evokes nurturance. However, it must be noted that the sample size for the user survey was small and for more reliable results new survey with more participants should be arranged. While developing the robot, fitting everything inside the robot was found challenging. The prototype was built mostly with separate easy to work sensor and drive modules. Building a custom PCB would save plenty of space. Fitting several motors and the speaker inside the head module increased the weight of the head, which forces the head to slouch. 5. FUTURE WORK The robot should be designed batterypowered as a cordless robot feels more like a real animal. Batteries can be fitted inside the legs of the panda. The robot would sound less robotic, if the noisy servo motors used in our prototype would be changed with silent actuators, e.g. voice coil actuators as used in Huggable. The robot would seem more alive if it followed interesting objects with its eyes and head using machine vision. The cameras could be also used for video monitoring the patients. Speech recognition could be beneficial as real pets react to their name and other commands. For active use in health care, the robot should have an antimicrobial fur. 6. ACKNOWLEDGEMENTS The authors would like to thank the personnel and the residents of Kustaankartano comprehensive service centre for participating in the user survey.

6 7. REFERENCES 1. Stiehl, W. D., Lieberman, J., Breazeal, C., Basel, L., Lalla, L., & Wolf, M. Design of a therapeutic robotic companion for relational, affective touch. In Robot and Human Interactive Communication, ROMAN IEEE International Workshop on. IEEE, 2005, Broekens, J., Heerink, M., & Rosendal, H. Assistive social robots in elderly care: a review. Gerontechnology, 2009, 8(2), Banks, M. R., Willoughby, L. M., & Banks, W. A. Animal-assisted therapy and loneliness in nursing homes: use of robotic versus living dogs. Journal of the American Medical Directors Association, 2008, 9(3), Savikko, N., Routasalo, P., Tilvis, R. S., Strandberg, T. E., & Pitkälä, K. H. Predictors and subjective causes of loneliness in an aged population. Archives of gerontology and geriatrics, 2005, 41(3), Lutz, W., Sanderson, W., & Scherbov, S. The coming acceleration of global population ageing. Nature, 2008, 451(7179), CORRESPONDING ADDRESS Panu Kiviluoma, D.Sc. (Tech), Post-doc researcher Aalto University School of Engineering Department of Engineering Design and Production P.O.Box 14100, Aalto, Finland Phone: , panu.kiviluoma@aalto.fi 9. ADDITIONAL DATA ABOUT AUTHORS Peltonen, Olavi, B.Sc (Tech) olavi.peltonen@gmail.com Orhanen, Samppa, B.Sc (Tech) smorhanen@gmail.com Venäläinen, Janne, B.Sc (Tech) janne.venalainen@outlook.com Auvinen, Matti, B.Sc (Tech) matti.a@outlook.com Sepponen, Raimo, D.Sc (Tech), Prof raimo.sepponen@aalto.fi Kuosmanen, Petri, D.Sc (Tech), Prof petri.kuosmanen@aalto.fi 6. Shibata, T. Seal-Type Therapeutic Robot Paro [WWW] ( ) 7. Saldien, J., Goris, K., Yilmazyildiz, S., Verhelst, W., & Lefeber, D. On the design of the huggable robot Probo. Journal of Physical Agents, 2008, 2(2), Goris, K., Saldien, J., Vanderborght, B., & Lefeber, D. How to achieve the huggable behavior of the social robot Probo? A reflection on the actuators. Mechatronics, 2011, 21(3),