ASME Early Career Technical Journal 2011 ASME Early Career Technical Conference, ASME ECTC November 4 – 5, Atlanta, Georgia USA

DEVELOPMENT OF A MODULAR COMPANION ROBOT FOR THE ELDERLY Jaime Mudrich, Andres Pacheco, Leonardo Ampie, Sabri Tosunoglu Florida International University Department of Mechanical and Materials Engineering Miami, Florida, U.S.A. ABSTRACT The goal of this project is to explore mechanisms that could be incorporated into the G.E.N.A.I. modular elderly assistance robot that would enable it to support its elderly consumer [1]. With today’s trend for advances in health care, people are living much longer than before. Increasing age is typically accompanied by increasing health care costs. To alleviate some of this stress, an autonomous elderly assistance robot has been prototyped. This work explores several mechanisms to enable the robot to shadow the consumer and concludes that triangulation via color-based cameras would be the optimal solution for the G.E.N.A.I. application. The cameras were not obtainable over the short scope of the project and so a triangulation system using ultrasound transceivers was employed instead. Testing is performed and while determination of distance to the target is accurate, the direction carries an error to the point of being unacceptable. The study reevaluates the method of calculation and switches from exact triangulation to a different system that closely approximates direction. Upon completion of testing, the “following” mechanism is incorporated into the G.E.N.A.I. robot with very successful results. The robot locates its target and maintains a desirable distance for effective service. Modularity is illustrated by integrating both a blood pressure diagnostic accessory and a medication organization accessory into the G.E.N.A.I. robot via universal “discs”. Upon realization that complete autonomy lacks essential flexibility, a voice interface is implemented. Finally, an ultrasound array proximity system is explored. Modularity is emphasized and continued “geriatric engineering” is promoted. INTRODUCTION As medicine rapidly improves, so does the life expectancy of our population. This means that we will have a larger and larger population of people who are steadily becoming less and less capable of caring for them. There are a multitude of options that address this problem. The dominant contemporary solutions are nursing homes for the elderly, and registered nurses that make house calls. The greatest complication with both of these solutions is the expense put forth to acquire these services.

ASME 2011 Early Career Technical Journal - Vol. 10

This challenging conflict is illustrated by the following statistical data: The 2010 MetLife Market Survey of Nursing Home, Assisted Living, Adult Day Services, and Home Care Costs reports an average annual cost for living in a nursing home to be approximately $40,000 per year [2]. The Employee Benefit Research Institute reports an average annual income of persons aged 65 and over to be $28,778 [3]. This means that on average, nursing homes are simply too expensive for the elderly or that an external financial contribution is required. The former case requires that assisted living be stricken as a solution to dependency in old age. The latter case imposes stresses on either the government or loved ones. This is undesirable. Another solution is hiring a registered nurse to make house visits. House visits include several types of services and therapies. Some examples of these therapies include occupational therapy, speech therapy, physical therapy, running errands, and social services. This approach eliminates the need for the elderly to trade their home in for an assisted living facility. It also reduces the amount of time required from loved ones for extra help. But at what expense? The National Clearinghouse for Long-Term Care Information reports that the average cost for a “home health aide” in the U.S., as of 2009, is $21 per hour [4]. Assuming one hour of assistance for five days every week over the course of one year, the elderly squanders approximately 19% of their annual income on health services. This is a better solution, financially, than an assisted living facility and it is less demanding on the care givers of the elderly. However, it is still very expensive and in many cases unsatisfactory or unavailable. This research’s proposed solution to the increasing health care need is automated assistance. This option is already in development for a hospital setting [5,6]. A follower, modular companion robot was designed and prototyped with the pupose of alleviating the medical service demand. G.E.N.A.I. is an acronym that stands for General Elderly Nursing and Assistant Instrument. Many of the regular tasks required by the elderly or disabled people can be performed completely by a mechanical device [7]. Housing these mechanical devices in an automated robot can eliminate the necessity for extra human assistance. Robots can be designed and programmed to perform an infinite amount of tasks. This versatility allows for the robot solution to

113

be applied to almost any unique disability-related challenge. For example, consider a person whose disability prevents them from climbing a set of stairs. This person can opt to “rent” a registered nurse for the steep rates mentioned previously, or instead a robot can aid in the task for no cost (after initial purchase) [8]. Robots are not cheap. Most elderly assistive technology comes with a respectable price tag. For example, consider power wheelchairs. The cost for these machines seldom dips below $1,000. However, the freedom gained by the investment more than justifies the expense. Furthermore, these mobility devices are quite often funded by the government as to limit the out of pocket contribution from the user. We believe that robotic assistance can be equally, if not more, liberating for the user than the power wheelchair. Costs will be comparable, if not less. And over time as the elderly population increases, cost of care will increase and perhaps the government will become interested in funding these sorts of devices as well. In the event that there is no government assistance, the cost of the robot over time is still less than a registered nurse or housing in an assisted living facility. MOBILITY Mobility is a critical component of the proposed G.E.N.A.I. elderly companion robot. To effectively assist the consumer, the robot must be able to efficiently navigate through different types of household configurations. Additionally, G.E.N.A.I. must be able to account for unpredictable obstructions, such as a person or a furniture rearrangement. Such constraints demand a versatile platform design. The platform must be fast enough to keep up with the consumer, taking into account the time required for readjustments. Agility is also critical to the platform, as the robot will be required to change direction frequently. Finally, the geometry of the platform must be considerate of snag prevention. To guarantee that G.E.N.A.I. will be able to stay near the consumer, the platform was designed with the capacity to move at a maximum speed of twice the average walking rate. Two common options for robotic propulsion are servomotors and direct current motors. Direct current motors are simple, powerful mechanisms, but require additional components for them to be useful to a microcontroller. Servomotors are more microcontroller friendly, but are significantly more expensive for the same power. After careful economic analysis, the direct current motors proved friendlier to the robot’s affordability. The selected motors were capable of traveling four feet in one second with no load. Such speed was well over the necessary load and so an engineering assumption was made that it would be able to keep up with the consumer, while carrying a significant payload. As an elderly companion, the G.E.N.A.I. robot is very likely to collect a great deal of responsibility and so must be reliable. There are several mechanical modes of failure that were considered in the design of the robotic platform. The first of these is the potential of the robot to be knocked on its side. Preventative measures were taken by carefully distributing the

ASME 2011 Early Career Technical Journal - Vol. 10

mass of the robot. By concentrating the mass of the robot close to the ground, it immediately becomes more resistant to knocks. To achieve this resistance, the heaviest component of the robot, the battery, was placed at the very bottom of the platform, only one inch from the bottom of its wheels. Additionally, a base was purchased that was much heavier than required. Figure 1 illustrates the base. The direct current motors had plenty of torque to spare and so some of it was traded for additional reliability. A second preventative measure was to limit the possible vertical distance at which a knock could occur. Tipping is a result of unbalanced moments. Moment has a direct relationship to the product of applied force and perpendicular distance from the point of consideration. While the force from a knock cannot be controlled, the maximum distance at which that force can be applied is controllable. The G.E.N.A.I. companion robot was designed to be as short as possible while still being functional. In the worst case, the force will be applied at the very top of the body. By reducing the robots height, the severity of the resulting moment is lessened and so tipping is further resisted. Constrained by functionality and reliability, the robot exhibits a height of approximately two feet.

Figure 1. Platform base A second mode of mechanical failure is becoming “snagged” by household obstructions and thus immobilized. If careful consideration isn’t taken, collision between the geometry of the robotic platform and geometries of other objects may result in the two locking themselves to one another. There are essentially two ways to prevent such interference. The first is to perform in-home consultations to minimize the potential of household configurations from resisting the functionality of the robot. This is costly and impractical for the desired “out-of-the-box” use of the G.E.N.A.I. robot. The second option is to attempt to prevent collision and to optimize the geometry of the body and base so that they are resistant to the various possible obstructions typically found in the household. This option is much more cost-efficient and in tune with the goals for the companion robot. Collision prevention is achieved via ultrasonic proximity sensors. This technology and its implementation will be expanded upon later. Several different geometries for the platform base were considered.

114

Amongst these were rectangular, triangular, and circular. The circular base was the best choice, owed to its absence of any edges. The body of the robot was designed to be cylindrical for the same reason. Without edges, the likelihood of G.E.N.A.I. to be caught on sharp corners is drastically reduced. Another preventative measure is wheel positioning. External wheels will easily catch on wall corners or table legs. To remedy this threat, the drive wheels were placed on the perimeter of the base, but inset via custom grooves in the base. Passive wheels were chosen that were short enough to fit underneath the platform and not be exposed to any stimuli outside of the platform perimeter. The completed base is depicted in Figure 1. OBJECT AVOIDANCE One of the greatest challenges pertaining to the G.E.N.A.I. companion robot is smooth passage across the household. The challenge is found in the numerous unknowns related to the configuration of an individual’s home. It is impossible to know the geometry of the consumer’s household. As such, the robotic platform must be equipped with enough sensors and logic to adapt to the more general cases. Typical collision avoidance systems are comprised of “time-of-flight” ultrasonic or electromagnetic proximity sensors, as well as a comparative logic that helps to paint for the robot a picture of its environment. This is also the case for the G.E.N.A.I. elderly companion robot. The best technology to use for household object detection turned out to be ultrasound. The most cost-efficient, commercially available choices were ultrasound or infrared. Infrared offers better accuracy than ultrasound, but has a very narrow beam. To effectively detect impending collisions, the wide ultrasound beam is preferable. Ultrasound easily acknowledges thin objects that infrared would miss, such as table legs. Typical ultrasound proximity sensors use a “time-of-flight” logic to figure the distance between the sensor and the next object in front of it. The sensor will first emit an ultrasonic pulse. The sensor then waits for the pulse to ricochet off of the next object in front of it and return. The sensor counts the time that passes between emission and reception. This time is halved (to account for the return trip) and then multiplied by the speed of sound at standard temperature and pressure to finally calculate the distance between the sensor and the next object in front of it. Due to the potential liabilities created by the services that the G.E.N.A.I. companion robot provides, it must be especially savvy in all of its functions. Efficient obstacle detection and avoidance is critical because of its contribution to the robot’s mobility. The ideal situation for object avoidance would be for the robot to store a map of the household in its memory. Such a map would prevent the robot from having to constantly take inventory of its surroundings and instead allow it to spend more time following the consumer. This situation would require costly memory expansion and the involvement of an experienced programmer. At this stage, such an option is only

ASME 2011 Early Career Technical Journal - Vol. 10

considered as a future expansion. On the contrary, the worst situation is the robot colliding with various objects and losing line-of-sight with the consumer. This situation obviously would not inspire the consumer with confidence, nor would it allow the platform to stay within the desired proximity of the consumer. The research conducted provided a solution between these two extremes. To accomplish efficient object avoidance behavior, an array of five ultrasound proximity sensors was employed. One of the sensors was placed on the front center of the robot, six inches above the platform base. The others were placed at the same height, on the same side, thirty degrees from one another, distributed symmetrically about the central sensor. Having a collection of sensors provided a better picture for the robot as to what its surroundings looked like. Planes were created via interpolation between data from adjacent sensors. These planes allowed the microcontroller to estimate where it would be able to pass the object. By executing this estimation from a short distance away, the platform was able to plan a more efficient trajectory to help keep up with the consumer. Figure 2, below, illustrates the configuration of the sensors as well as a basic expression of the data collection from a general object.

Figure 2. Object avoidance schematic Without an object avoidance system, G.E.N.A.I. would merely locate the consumer signal and chart a direct path to that signal. With the ultrasound array system, G.E.N.A.I. used the consumer signal coupled with the environmental data it gathered, to plot a more efficient path, keeping it a step ahead of its less capable predecessor. Again, the best solution would be a mapping system that would be less dependent on the robot’s sensory. Additionally, infrared packages exist in which a single sensor will sweep across one hundred eighty degrees, collecting high-resolution environmental data. This of course is far more expensive, but ought to be considered after the prototype stage. NAVIGATION SYSTEM Household navigation is the single most challenging aspect of the companion robot design. The difficulty lies in the many unknown variables pertaining to target recognition. A few examples of these variables include line-of-sight obstructions,

115

electromagnetic interference, and multiple targets. Numerous systems have been examined and compared. For the purpose of prototyping, the researchers opted to employ triangulation via ultrasonic transceivers. What follows is an explanation of the concept being used and its application. ULTRASOUND TRIANGULATION To localize the target, two ultrasound sensors were mounted on the front of the robot at a known distance from one another [9]. The consumer carries an ultrasound emitter on their person. When the signal strikes the robot, each sensor determines the angle at which the signal is coming from with respect to the frame connecting them. With these two angles, and the distance between the sensors known, the microcontroller can determine the coordinates of the signal relative to a frame fixed to the robot. Figure 3, below, illustrates the concept behind this technology.

Figure 3. Graphical representation of triangulation CONTROLLER SOFTWARE DEVELOPMENT The controller programming is what gives the robot life and allows it to make decisions. For the design project at hand, the study required a program that could effectively collect environmental data from the ultrasound sensors and process that data. Once the data was processed, the robot had awareness of its environment and the location of the person whom it was to follow. Finally, the microcontroller, with this information, was able to command the mobility components to shorten the gap between the robot and the target elderly consumer. The microcontroller first attempts to determine where the consumer is relative to the robotic platform. To accomplish this, the aforementioned ultrasound triangulation is employed. Each of two triangulation ultrasound transceivers takes a turn calculating the distance to the consumer. After this information is obtained, the following equations are used to determine the relative position of the elderly consumer with respect to the G.E.N.A.I. platform. Equation 1

Equation 2 where (xR, yR) is the relative location of the consumer in a coordinate system fixed to the G.E.N.A.I. platform. The left and

ASME 2011 Early Career Technical Journal - Vol. 10

right ultrasound data are denoted as “d1” and “d2”, respectively, and finally, “L”, represents the distance between the two triangulation ultrasound sensors. These values are passed to the mobility mechanism, finally allowing the robot to move towards the elderly target. The logic map, shown in Figure 4, is a symbolic representation of the microcontroller behavior.

Figure 4. Logic map for platform mobility MODULARITY The primary goal for the G.E.N.A.I. robot was to develop a robotic platform capable of satisfying a substantial variety of elderly needs. One possibility for accomplishing this was to stock the robot very densely with a large quantity of mechanisms directed towards the most general complications in the every day life of the elderly. Upon deliberation, it was concluded that such a method might lead to extra, unused components on the robot. Each individual has unique challenges and so the “general” design may not be sufficient. To the contrary, it may contain more features than required and so the design would not be cost-efficient for the consumer. For these reasons, the “general” design was exchanged for a customizable, modular design. A modular design for this application is desirable because of the great diversity in day-to-day complications for the elderly. For the consumer, a modular design means an initial purchase of a standard robotic platform. The platform contains the components that offer the robot the most potential for achieving a multitude of tasks. Examples of these components are the mechanisms that enable mobility, the object-avoidance components, and a navigational system. Mobility is required to remedy the decreasing mobility of the aged consumers. The companion robot can offer very little, statically. Object avoidance is required as a supplement to mobility. A household navigational system is critical and could also be said to

116

supplement the mobility of the robot. With mobility components, the robot may be able to move, but it will not be able to figure out where to go, e.g., towards the consumer, or a recharge station. These standard components give G.E.N.A.I. the capacity to perform a majority of the services provided by a home nurse. Purchasing the standard G.E.N.A.I. platform is like purchasing a drill. It affords you great torques and rotation speeds, but is essentially worth nothing without drill bits to transmit that power. The drill bits in this case are the accessories that will connect G.E.N.A.I. to the consumer through the unique services they provide. Some of the potential tasks considered that would motivate an accessory are medication control, alerts and reminders, at-home doctor appointments via video chat, diagnostics, and so on. To date, the actualized accessories are medication control, voice interface, and early stages of diagnostics. MEDICATION CONTROL One of the more feasible accessories to construct, and also one of the best to exemplify the potential of the G.E.N.A.I. robot, was medication control. It is very common that aged citizens will rely on a multitude of medication to combat deficiencies resulting from old age. Most of these medications are required to be taken on different days and different times. As such, medication scheduling can become rather complicated. A remedy to this complication can be found in commonplace organizers with pill compartments, ranging from morning to night and from Sunday to Saturday. Some are available with scheduled alarms to assist the consumer in remembering to take the appropriate medication. One drawback for these systems is limited range. The medication might be completely organized, and alarms set at the appropriate time, but an alarm may not be audible across the home (or even in the next room for that matter). In this very real circumstance, the G.E.N.A.I. companion robot provides a remedy. At its roots, G.E.N.A.I. is a household navigation follower robot. Ideally, it will always be in close proximity to the consumer where he or she will easily notice any kind of notification provided by the medication organizer. For prototyping, a simple, seven-compartment tray was fabricated and mounted on a one hundred eighty degree servomotor. The servo and tray were then mounted on a twelveinch disc that fit into the top of the robot body and rested on several supports installed around its inner diameter. (The twelve-inch disc represents the universal mounting system for all of the accessories.) The prototype medication organizer was also equipped with a buzzer and LED that were wired through the ceiling of the robot where they were easily heard and seen, respectively. At the appropriate time and date, the tray would turn to reveal the scheduled medication. The LED would flash and the buzzer would sound, reminding the consumer that it was time for their medication. After a time interval, the tray would conceal the previously exposed compartment and instead block

ASME 2011 Early Career Technical Journal - Vol. 10

the opening to the tray. Figure 5, below, depicts the conceptual model and prototyped medication control module.

Figure 5. Conceptual design for medication dispensing VOICE INTERFACE A completely autonomous design for the G.E.N.A.I. robot was desirable, but its consideration yielded too many complicated circumstances. It was quickly realized that effective autonomy would require employing a team of programmers. Instead, a human interface was introduced as another accessory for the G.E.N.A.I. platform. Such an accessory allowed more flexibility to the governing algorithms of the platform and so would be able to serve the consumer more effectively. Unlike the medication organizer, this accessory could not be fabricated so easily. The solution was provided in the form of the Parallax Say It module [10]. The Say It module identifies verbal commands and through programming can be related to the actions of the G.E.N.A.I. robot. The module comes with PC software that allows the programmer to create different words within the module memory for it to recognize. The user then calibrates that word to the spoken command. The code affiliated with the new commands are then generated and exported to the same format used by the governing microcontroller. One of the motivators for the voice interface was the situation where the user requires temporary privacy, such as in the restroom. Under the normal logic of the robot, it was supposed to continuously follow the consumer, unless the consumer was within a certain range of the robot. That included following right into the restroom as well. It would not be uncommon for the consumer to prefer that the robot wait outside during this time. To accommodate this preference, the voice interface was employed. A command called “wait” was created and linked to a subroutine. The subroutine waits ten seconds, allowing the consumer to shut the bathroom door, and then activates a motion detector that contains the doorway in its sensory cone. The robot was required to patiently wait where it was when the command was given until the motion detector sensor was fired, indicating the door opening once more. Another example of the usefulness of the voice interface was the situation where the robot blocks the consumer’s line of sight. This situation can easily occur as the consumer is walking past his or her television set to have a seat. If the path to the seat is between the seat and television, it is quite possible that the

117

robot will end up resting right in front of the television. To the robot’s understanding, it is doing exactly what it is supposed to do. To the user, this is clearly not what the robot should be doing. Fortunately, with the Say It module, the user can simply call out “Move!” and follow it with a direction to get the robot out of the way. Another case where the “Move!” command might prove useful is if the robot is blocking a tight passageway. Without the voice command module, the G.E.N.A.I. robot can still operate and make use of the other accessories that the consumer may choose to integrate. However, it is clear that having such control would be a desirable improvement over complete autonomy. Point being, that the modularity of the robot allows the consumer to purchase those accessories that fit their budget and needs, at their discretion.

considering that it is an invasive measurement, it was shied away from (for now) as a potential liability. In future work it will be desirable to include a data acquisition system that can collect the blood pressure measurements and then wirelessly communicate that data to the consumer’s physician. This accessory very clearly and powerfully illustrates the immense benefits of the G.E.N.A.I. robot. What should also be taken from this example is the fact that the diagnostic system is merely an accessory. It is not included in the standard, universal robotic platform. What this means is that if you happen to be a diabetic, then this accessory is perfect for you and you should purchase one. If you are not diabetic, then there is no reason to have this feature and so the consumer is spared the expense. Modularity is key to the success of such technology.

BLOOD PRESSURE DIAGNOSTICS The final accessory exhibited over the course of the research was a blood pressure diagnostic device. The device was store-bought and merely positioned within the robot. The controls for the device were mounted on the standard twelveinch disk and made accessible for the consumer. The cuff and tube used to collect the blood pressure was hidden within the body of the robot, accessible through a door in the back. Figure 6, below, illustrates the blood pressure module.

DISCUSSION The researchers believe that they have successfully prototyped a modular elderly assistance robot. G.E.N.A.I. is capable of following an ultrasonic signal through an environment laced with random obstructions, simulating a real household situation. The test environment was not nearly as complex as a consumer’s home could be, but still served as a successful proof of concept. The mechanism for following the consumer (ultrasound) is with many pitfalls, but again serves well as a proof of concept. The researchers are confident that additional temporal and financial investment will easily remedy such pitfalls. Figure 7 illustrates the completed prototype.

Figure 6. Blood pressure diagnostic module Very often people are required to perform their own diagnostics at home and then report that data to their doctors. For example, in the case of diabetes, sufferers of diabetes very frequently are required to measure their own glucose levels and blood pressure levels. Collecting these measurements requires a pair of machines, typically stored in one place. When time comes to take these measurements, the diabetic must stop what they’re doing, go to this location, take the measurements, and then return once again. To facilitate the meticulous measurements, the blood pressure accessory was integrated into the G.E.N.A.I. robot. Instead of having to return back to the location of machine storage, the consumer can simply turn to the G.E.N.A.I. robot that is standing right next to them. The back door is opened, the cuff is pulled out, and then the machine is turned on. It was very desirable to also implement a glucose-measuring device, but

ASME 2011 Early Career Technical Journal - Vol. 10

Figure 7. Robot body concept (left) and prototype (right) Modularity is illustrated via interchangeable accessories in the form of an automatic pill manager as well as a blood pressure diagnostic device. The intent of this research was to introduce a novel concept of modular elderly assistance and inspire future development. The researchers feel that they were successful in achieving this goal, in that the benefits of the modular design are obvious. Consumers are given the ability to customize the G.E.N.A.I. platform as required to maximize the economy of their investment. Future work will integrate the accessories into the body of the robot better than the initial prototype. Additional accessories

118

will also be designed and integrated to further illustrate the versatility and necessity of a modular companion robot. Furthermore, alternative localization systems should be investigated to replace the ultrasound system; environmental sensors with wireless communication may prove fruitful. A robust household layout mapping system would be beneficial to the mobility of the platform and should be invested in. ACKNOWLEDGMENTS The authors would like to cordially thank the Ronald E. McNair Post Baccalaureate Achievement Program for supporting the development of the G.E.N.A.I. modular companion robot for the elderly. REFERENCES [1] Ampie, L., Mudrich, J., Pacheco, A., & Tosunoglu, S. (2011) G.E.N.A.I. Companion Robot. Florida Conference on Recent Advances in Robotics. Gainesville, Florida [2] MetLife Mature Market Institute. (2010). Market Survey of Long-Term Care Costs. Connecticut. [3] Employee Benefit Research Institute “EBRI”. (2008). Income of the Elderly Population Age 65 and Over. Retrieved August 13, 2011, from www.ebri.org/pdf/notespdf/ EBRI_Notes_06-June10.Inc-Eld_COBRA.pdf [4] National Clearinghouse for Long-Term Care Information “NCLTCI”. (2009). Costs of Care. Retrieved July 15, 2011, from http://www.longtermcare.gov/LTC/Main_Site/Paying/ Costs/Index.aspx. [5] Pollack, M., Engberg, S., Thrun, S., et al. (2002). Pearl: A Mobile Robotic Assistant for the Elderly. AAAI Workshop on Automation as Eldercare. [6] Baltus, G., et al. (2000). Towards Personal Service Robots for the Elderly. Carnegie Mellon University. Retrieved September 12, 2011, from http://www.cs.cmu.edu/~nursebot [7] Guth, J. (2002). High Tech Medicine and Robotics. Retrieved September 13, 2011, from http://jhguth1942.tripod.com/scitechnews/id9.html [8] Chen, C., Liao, T., & Pham, H. (2010) On Climbing Winding Stairs for a Robotic Wheelchair. World Academy of Science, Engineering and Technology. 299-304 [9] Jimenez, A.R. and Seco, F. (2005) Ultrasonic Localization Methods for Accurate Positioning. Madrid: Consejo Superior de Ivestigaciones Cientificas. [10] Parallax “Say It” Module. Parallax 2011 catalog #30080

ASME 2011 Early Career Technical Journal - Vol. 10

119