Utilization of Ultrasound Sensors for Anti-Collision ... - IEEE Xplore

24

IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 13, NO. 1, MARCH 2005

Utilization of Ultrasound Sensors for Anti-Collision Systems of Powered Wheelchairs Tilak Dutta, Member, IEEE, and Geoff R. Fernie

Abstract—Anti-collision systems have been developed for use with powered wheelchairs in order to enable people with cognitive or physical impairments to safely operate a powered wheelchair. Anti-collision systems consist of sensors that have the ability to detect objects near the wheelchair and a computer that can stop the chair if a collision is determined to be likely. This investigation considered the suitability of using ultrasound sensors in such a system when encountering objects typically found within a home or a long-term care facility. An ultrasound sensor’s ability to detect an object was dependent on the object’s size, shape, specularity, reflectivity, and sound absorption characteristics. Ultrasound sensors, by themselves, were found to be unsuitable for anti-collision systems due to an inability to detect objects commonly encountered in the target environment (the home or long-term care facility) without increasing the complexity of the system to such a degree that it would be prohibitive to deploy this technology to the public. Index Terms—Anti-collision system, assistive technology, powered wheelchair, ultrasound sensor.

I. BACKGROUND

P

OWERED wheelchairs are commonly used by people who do not have sufficient strength to use self propelled wheelchairs. If a person with such a physical disability also has any significant cognitive, visual, or motor impairment, the use of a powered wheelchair is generally ruled out owing to the danger to the operator as well as to nearby bystanders. This is of particular importance in a long-term care facility where the population is highly susceptible to serious injury particularly if a collision causes a fall. A person with a cognitive, visual, or motor impairment may be able to operate a powered wheelchair if an anti-collision system, having the ability to sense objects (especially other people), and stop the wheelchair when necessary, were available. There have been a number of attempts to design autonomous/semi-autonomous wheelchairs or anti-collision systems for use with powered wheelchairs using ultrasound sensors as the primary method of obstacle detection [1]–[4]. However, there has been limited research into the suitability of ultrasound technology for anti-collision systems. Brown and Bradley determined that ultrasound sensors would only be appropriate if used in combination with a secondary sensing Manuscript received March 10, 2004; revised November 26, 2004; accepted November 30, 2004. This work was supported in part by the Canadian Institute of Health Research under Grant MOR-57696. T. Dutta is with the Centre for Studies in Aging, Sunnybrook and Women’s College, Health Sciences Centre, Toronto, ON M4N 3M5, Canada (e-mail: [email protected]). G. R. Fernie is with the Toronto Rehabilitation Institute, Toronto, ON M5G 2A2, Canada (e-mail: [email protected]). Digital Object Identifier 10.1109/TNSRE.2004.842366

system [5]. However, in [5], as in other investigations [1]–[4], there has been a lack of consideration given to the implications of deploying this type of technology into a typical seniors’ care environment. There are two aspects to this deployment that the authors feel are worthy of further discussion. First, most of the smart wheelchair systems being designed currently are too complex to become a commercial product within the near future. As Fioretti, Leo, and Longhi suggest, both the initial cost as well as the cost of maintenance of sophisticated sensors and computer systems required to integrate the various subsystems to the wheelchair should be considered [6]. Nisbet is of the opinion that the ultimate goal of research in intelligent wheelchairs should be the transfer of technology into commercial domains so that the public will have access to it. He goes so far as to say that the business of developing intelligent could be accused of being conducted for wheelchairs “ the benefit of engineers wishing to play with sophisticated technology, rather than people with disabilities [7].” Second, as is also suggested by Nisbet, some consideration needs to be given to the particular setting where the system is likely to be used [7]. Our target population is most likely to be found in their own residences or long-term care facilities. In long-term care facilities, elderly powered wheelchair users are operating in an environment where other vulnerable elderly residents are walking. These ambulatory residents typically use walking aids and are very unsteady on their feet. Even a minor collision would very probably cause them to fall. Studies have estimated 20%–30% of such falls will result in injury requiring medical attention [8]–[10], including 5%–10%, which will result in a fracture. Hip fracture, one of the most devastating consequences of a fall, occurs in about 1% falls in the community [11], [12] and this incidence may triple for residents of nursing homes [13]. It is known that almost none of these individuals who fracture a hip will regain their original mobility [14], [15] and up to 40% of them will die from complications within six months of sustaining hip fracture [13], [15], [16]. Clearly, grave consequences are likely in this at-risk population if the anti-collision system failed to detect their presence and, therefore, extra care needs to be taken to ensure that obstacle detection is consistently reliable. Providing increased freedom of movement for some residents at the expense of creating a potential of serious harm to other vulnerable residents would not be considered to be ethically acceptable. It may seem unreasonable to expect any anti-collision system to work perfectly, however, in this case, such an expectation may not be impractical because it is acceptable to limit the maximum speed of operation of the powered wheelchair to a very slow pace (0.3 m/s) and the area of operation can be necessarily limited

1534-4320/$20.00 © 2005 IEEE

DUTTA AND FERNIE: UTILIZATION OF ULTRASOUND SENSORS FOR ANTI-COLLISION SYSTEMS OF POWERED WHEELCHAIRS

25

Fig. 1. Cones of detection for objects with different visibilities. The objects are visible anywhere within the volume defined by the cone of detection. A large/hard object is detectable in a larger volume than a small/soft object, which can become invisible near the edges of the manufacturer’s specified cone of detection.

to the immediate environment of the nursing unit. Independent mobility, even with the limitation, is highly preferable to no independent movement. II. OBJECTIVE The objective of this investigation was to determine whether ultrasound sensors are suitable for an anti-collision system of a powered wheelchair when encountering objects typically found in the home or a long-term care facility. Although one specific brand of sensor was investigated, efforts were made to focus on fundamental characteristics inherent to the principle of ultrasound sensors and their most commonly used frequencies, (see Appendix for more details on ultrasound sensors as well as how the signals from the ultrasound sensor were interpreted) as well as to minimize the overall complexity of the resulting system to make real world implementation feasible. III. PRELIMINARY INVESTIGATION In order to determine whether ultrasound sensors are an appropriate technology for anti-collision systems, the characteristics of the Devantech SRF04 Ultrasonic Ranger (Norfolk, U.K.) were investigated. Although parameters for a cone of detection were given in the product literature, preliminary observations suggested that the shape of the cone of detection was a function of the object’s size, geometry, and sound absorbing qualities. The size of the cone of detection was used to quantify the detectability (or visibility) of an object to the sensor. It was expected that large, hard objects would be detected over a large angle whereas smaller or softer objects would be detected over a smaller angle as shown in Fig. 1. This prediction was tested using three spheres with diameters of 0.1, 0.2, and 0.3 m. A. Methodology Each sphere was suspended in front of the sensor in a plane one meter from the sensor. In order to map the transverse section of the cone of detection, the spheres were slowly ( 0.01 m/s) moved vertically in the plane to determine the point where the sensor no longer detected the sphere. This process was repeated for azimuthal angles of 0 –40 in increments of 5 (Fig. 2). The visibility of each sphere was measured first with the spheres tightly covered in a thin polyethylene sheet and then repeated while tightly covered with a thin layer of cloth. This procedure was repeated three times for each sphere.

Fig. 2. Three spheres (with diameters of 0.1, 0.2, and 0.3 m) were suspended, one at a time, in front of the ultrasound sensor in a plane 1 m away from the sensor. The sphere was slowly raised/lowered until the point where the sensor could no longer detect the sphere was reached. This process was repeated for azimuthal angles of 0 –40 in increments of 5 . The visibility of each sphere was first tested with the sphere covered in a thin plastic film and then while covered in cloth.

B. Results and Discussion In general, larger objects were detectable over a large area (i.e., they had a large cone of detection) whereas smaller objects were detectable over a much smaller area (i.e., they had a smaller cone of detection). As expected, the 0.3-m sphere was most visible and the 0.1-m sphere was least visible. Covering the object with cloth, a material, which absorbs sound well, made the object markedly less visible (Table I).

IV. EXPERIMENT The data from the preliminary investigation were used as a guide to help predict what types of objects the ultrasound sensor might not “see” well. In particular objects commonly found in the home or in long term care facilities were considered. The visibility of the following objects was tested: 1) simulated cane or table leg; 2) simulated human leg; 3) facing perpendicular to a wall; 4) wall at various angles; 5) table top.

26


TABLE I CROSS SECTIONS OF THE CONE OF DETECTION FOR THREE SHPERES

In addition, tests were done to determine 1) the effect of having two wheelchairs, each with its own ultrasound anti-collision system, in the same room and 2) the effects of different forms of environmental interference such as the noise from jingling keys or noises from compressed air lines. All tests were repeated three times. A powered wheelchair equipped with an anti-collision system to be operated by someone with a cognitive, visual, or motor impairment would be limited to moving at very slow speeds for the safety of both the operator and those nearby. Measurements done with the Nimble Rocket (Toronto, ON) powered wheelchair showed that an approximate stopping distance of 0.1 m was required to stop a wheelchair moving at 0.3 m/s (which was determined to be a safe speed for someone with a cognitive, visual or motor impairment). Note that for this stopping distance, we assumed objects in the surroundings were not moving. Thus, the system should be able to detect objects before the distance between the chair and the object is reduced to below 0.1 m. Allowing for people in the environment to be moving toward the chair would require the stopping distance to be reduced.

Fig. 3. Setup for testing of simulated table leg/cane involved the use of a 1.2-m steel tube with 0.025-m diameter to simulate a table leg or cane. The tube was systematically positioned at points on a polar grid to determine where the pole was visible to the sensor.

object was detected and, if so, what the error in the detected distance was. These measurements provided a transverse section of the cone of detection. B. Simulated Human Leg

A. Simulated Cane or Table Leg A 1.2-m long steel tube with 0.025-m diameter was used to simulate a cane or table leg. The sensor was mounted at a point 0.6 m above the floor and the tube was placed in front of the sensor in 5 increments along semi-circles located 0.5, 1.0, 1.5, 2.0, and 2.5 m from the sensor (Fig. 3) to determine whether the

A 1-m cylinder with 0.10-m diameter was loosely covered in cloth and was used to simulate a human leg. Pilot testing indicated that the model leg was an accurate representation for the purposes of determining the detectability of the leg by the ultrasound sensor. The procedure for testing of the simulated leg was identical to that of the simulated cane/table leg (Fig. 3).


27

compressed air line was discharged 1 m from the side of the ultrasound sensor. A set of seven keys was jingled 0.25 m from the side of the ultrasound sensor. V. RESULTS A. Simulated Cane or Table Leg The simulated cane/table leg was detectable in a cone that had an angle of approximately 27 at the cone’s vertex (Fig. 6). B. Simulated Human Leg The simulated human leg was detectable in a cone that had an angle of approximately 17 at the cone’s vertex (Fig. 7). Fig. 4. Setup for detecting a concrete wall at various angles. The sensor was kept 1.5 m from the wall and 0.6 m from the floor and was moved around in a semicircle to determine at which angles the wall was detectable.

C. Wall 1) Wall Surfaces: A number of commonly encountered wall surfaces were tested (with the sensor perpendicular to the wall at a distance of 1 m) for their ability to be detected by the ultrasound sensors. Painted dry-wall, concrete, wood as well as glass panes were tested. The conditions described in 2) and 3) were done with a concrete wall. 2) Moving Away From a Wall, Perpendicular : The detected distance to a wall was measured using the ultrasound sensor and compared with the actual distance for the following distances: 0.03, 0.05, 0.10, 0.15, 0.25, and then in 0.25-m increments to a maximum distance of 4 m. 3) Detection of a Wall at Varying Angle: The sensor was placed 1.5 m from the wall and 0.6 m from the floor and was moved around a semicircle (from 0 to 90 ) while recording the distances detected by the sensor at every 5 increment (Fig. 4). D. Table Top Two cases were investigated using a wooden table top (1.22 0.61 0.03 m) with flat edges to determine its detectability: 1) Case 1: Sensor 0.5 m away from the center of the table top [Fig. 5(a)]. 2) Case 2: Sensor 0.5 m away from a point 0.15 m from the edge of the table top [Fig. 5(b)]. Each case was tested with and without a table cloth starting with the sensor facing perpendicular to the table top then moving the sensor through 90 . The table cloth was placed on the table top such that there was 5 cm hanging over the side of the table. First, the sensor was at the same height as the table top, and was then raised above the table top.

C. Wall 1) Wall Surfaces: The ultrasound sensor was equally sensitive to all forms of wall surface tested (drywall, glass, wood, cement) as the detected distance was the same for all wall surfaces. 2) Moving Away From a Wall, Perpendicular: The sensor consistently measured a distance to the wall that slightly underestimated the actual distance (Fig. 8). 3) Approaching a Wall at Various Angles: The ultrasound sensor detected the wall over a wide range of angles. The wall was not detectable when the sensor was positioned within 7 1 from the plane of the wall (Fig. 9). D. Table Top The table top was very poorly detected. With the sensor 0.5 m away from the center of the table top, the bare wooden table top was visible up to 27 off the normal and the table top covered with a table cloth was visible up to only 9 off the normal. With the sensor 0.5 m away from a point 0.15 m from the edge of the table top, the bare wooden table top was visible up to 24 off the normal and the table top with a table cloth was detectable up to 5 off the normal (Table II). The results shown in Table II were for the setup with the sensor at the same height as the table top. As the sensor was raised above the table, the angle through which the table top could be detected was reduced dramatically to a point where it was invisible at 0.25 0.02 m above the table top. E. Multiple Sensors in the Same Room Interference, in the form of erratic distance measurements, was observed over an angle of approximately 10 from the normal in both directions. F. Environmental Interference

E. Multiple Sensors in the Same Room Two Devantech sensors were mounted 1 m apart, at the same height, facing each other. One sensor was held still, while the other was rotated about its center 90 in either direction.

Both the compressed air release and the jingling keys caused a significant amount of interference. VI. DISCUSSION

F. Environmental Interference

A. Basic Requirements

The effect of discharging a compressed air line as well as the effect of jingling keys near the sensor was investigated. A

The first factor to consider in deciding whether or not ultrasound sensors are suitable for use in an anti-collision system

28


Fig. 5. Two setups were used for testing the ultrasound sensor’s ability to detect a table top with dimensions 1.22 the center of the table top. (b). Case 2: Sensor 0.5 m away from a point 0.15 m from the edge of the table top.

2 0.61 m. (a). Case 1: Sensor 0.5 m away from

Fig. 6. Horizontal cross-section of the cone of detection for the simulated table leg/cane. In the range of interest, the cone of detection has an angle of approximately 27 .

is to determine whether such sensors can resolve distances accurately in the range of interest. The accuracy of the detected distance was good for our entire testing range (Fig. 8). B. Sensors on Front and Back of the Wheelchair Since the ultrasound sensor had the ability to provide accurate ranging information, the other characteristics for a simple anti-collision system were considered. Perhaps the simplest design for an ultrasound sensor anti-collision system is with a single sensor on the front of a wheelchair and a single sensor on the back. A setup such as this highlights one of the main negative characteristics of ultrasound sensors, which is the large variation in the size of the cone of detection based on the shape and texture of the object being detected (Figs. 6 and 7). The angle at the vertex of the cone of detection was 17 for the simulated leg (soft object) and 27 for the simulated cane/table leg (hard object) up to a distance of 1.5 m. Note that materials likely could be found that would cause an even wider variation in these angles based on specularity and reflectivity. To ensure that a soft object in the wheelchair’s path is detected, the smaller cone must span the entire area directly in front of the wheelchair. This sets the distance from the front of the wheelchair at which the sensor

Fig. 7. Horizontal cross section of the cone of detection for the simulated human leg. In the range of interest, objects outside an angle of approximately 17 were effectively invisible to the sensor.

must “look” for objects (using a 17 cone and a 0.61 m wide wheelchair, this gives a 1 m “sensing” distance). The problem that results with this setup is that the cone of detection for the hard objects (because it is much bigger) now “sees” objects that would not actually be in the wheelchair’s path. The implication of this is that no hard objects can be found in zones on either side of the wheelchair without causing “false stops,” as shown in Fig. 10. A 0.61-m (24 in)-wide motorized wheelchair equipped with an ultrasound sensor would require 0.20 m (8 in) on either side or a 1.01 m (40 in) wide path in order to avoid a false stop. Since standard doorways are 0.84–0.92 m (34–36 in) wide, this type of system would stop the wheelchair while approaching a viable doorway and would likely result in the user disabling the anti-collision system. To minimize the problem caused by the variation in the cone of detection, multiple sensors could be used. For example, if we consider a system with two ultrasound sensors on the front of the chair, (Fig. 11) we see that we only now require 0.10 m beside the chair in order not to experience a false stop (cf. our single


29

rapid ultrasonic firing (EERUF) [17], [18] but this is a significant increase in complexity of our overall system and results in longer processing times [4]. In fact, as Borenstein and Koren admit, the process of determining the timing pattern initially is a complex problem [17]. There are additional issues with the use of EERUF that will be discussed later. There are a number of other drawbacks to ultrasound sensors. There was a “blind spot” when the sensor was placed any closer than 7 to the wall due to specular reflection (Fig. 9). One implication of this result was a possibility of running into a wall while traveling down a long corridor if the chair veers slightly from the centre of the corridor. However, it is important to remember that guarding against collision with a wall is of secondary importance, and that our main consideration is the protection of other ambulatory patients and their walking aids. C. Sensors on the Side of the Wheelchair

Fig. 8. Actual distance versus detected distance while sensor was moved perpendicular to a wall. Resolution in measurements was 0.03 m.

6

Fig. 9. Results of approaching a wall at various angles. The ultrasound sensor 1 . was able to detect the wall for all angles greater than 7

6

sensor case where we needed 0.20 m). Increasing the number of sensors even higher will further reduce this “empty space” requirement beside the chair. The limit to how much this can be minimized is dependent on our minimum stopping distance of 0.1 m (as discussed earlier). Since objects must be detected at a minimum distance of 0.1 m (assuming computing time is instantaneous) there will be some finite amount of empty space needed next to the chair. The drawback to having multiple sensors on the same chair would be the increased complexity of the overall system, as well as a significant time delay. As Brown and Bradley write about a system with seven sensors, “[t]he principal disadvantage with this system is the time delay inherent in the signal traveling through the air. If the signal is sent in a number of directions, the time delay can build up to a second or more” [5]. Sensors must be fired sequentially to avoid crosstalk (one sensor receiving the echo of a neighbor). This problem can be addressed by applying error eliminating

The problem of running into a wall at a sharp angle may be solved by mounting additional sensors on the sides of the chair. However, this introduces some new challenges. One drawback of this is that it now becomes considerably harder to navigate through a doorway. If the sensors on the sides are set to stop the wheelchair if they come within 0.05 m of a wall, then we lose 0.1 m of the 0.3 m original clearance (0.9-m-wide doorway and 0.6-m–wide wheelchair). Again, note that this limitation of reduced clearance through a doorway is of secondary significance as collision with a door frame does not endanger anyone’s safety. There is also an issue with sensor minimum range capabilities. The Devantech sensor evaluated would work fine with this requirement since it can detect objects down to a minimum distance of 0.03 m; however, many popular ultrasound sensors have minimum detection range that would be prohibitive. For instance, as LoPresti, Simpson, Miller and Nourbakhsh found, the Polaroid 600 and Polaroid 900 have minimum detection distances of approximately 0.08 and 0.25 m, respectively [19]. Another point that this minimum range issue brings up is what would happen if clothing, or other material accidentally covered either the transmitter or the receiver. In both cases, the system would assume there are no obstacles present, which is of great concern. D. Other Obstacles Another poor characteristic of the ultrasound sensor was its marked inability to detect table edges, particularly for a table top with a table cloth (Table II). The covered table was only visible to the ultrasound sensor with the sensor perpendicular to the table, or a maximum of 5 off the normal. Note that this is the case with the sensor at the same height as the table top. If the sensor and table top do not line up, detection becomes even worse due to specular reflection. To minimize the effect of this limitation, a well designed anti-collision system utilizing ultrasound sensors would have to incorporate sensors at various heights. A possible solution to table-like obstacles, doorways as well as approaching a wall at a sharp angle is to create a modified environment with and tables designed to be visible to ultrasound

30


RESULTS

OF

TABLE II TABLE TOP DETECTION

sensors, wider doors and rough wall paper. This may be feasible as our target environment is the home or long-term care facility, which means that we can assume the user is limited to a given modified space. E. Interference

Fig. 10. To ensure that a soft object in the wheelchair’s path is detected, the smaller cone must span the entire area directly in front of the wheelchair; however this results in the cone of detection for hard objects “seeing” objects that would not actually be in the wheelchair’s path.

Fig. 11. Multiple sensor system. With such a system, 0.10 m is required on either side of the wheelchair to not experience a false stop.

Interference is another problem that ultrasound sensors are prone too. Interference can be caused both by other ultrasound sensors in the vicinity (known as crosstalk) as well as from other sources of high frequency noise in the environment. Such interference, if left unchecked would cause unreliable operation if two wheelchairs with such systems were ever brought in close proximity to each other. The outgoing pulse of sensor on one chair being interpreted as the returning echo of sensor on the other chair would likely cause both chairs to stop unpredictably. The other form of interference that would have a similar effect is environmental interference. In our testing both the noise from compressed air lines as well as jingling keys elicited erratic responses from the ultrasound sensor. These types of interference can be minimized by incorporating EERUF, which uses a comparison of consecutive readings to filter out noise from external sources. This system also incorporates the use of alternating delays between successive ultrasound bursts. This allows a faster firing of the ultrasound sensors, which solves the problem of the time delays mentioned earlier. However, even if EERUF is able to differentiate “good” and “bad” signals and allow for faster firing, it is likely that it would result in an overall sensing time that was on the same order as without EERUF. Without using EERUF, as Hwang, Chen, and Hong state, it would take 300–600 ms to fire all sensors once


while avoiding crosstalk [20]. It takes 100–200 ms for all sensors to fire once using EERUF with 12 sensors on board [21]. Let us consider a situation where the wheelchair turns 90 with and without EERUF. With EERUF, the first set of readings from the sensors would be taken to be external noise as a new set of objects are now in front of the chair and the system would wait for the next set of readings to come in before deciding that the new data is “good.” So we might require 200 ms to 400 ms to truly sense an object, which is on the same order as the sensing time without using EERUF. The other issue that comes up if there are two powered wheelchairs in the same vicinity each with its own anti-collision system, is that our assumed stopping distance will no longer be sufficient to allow two such wheelchairs driving toward each other to stop without colliding. VII. CONCLUSION Ultrasound sensors do have the ability to resolve distances in the range of interest. However, with single sensor detection, the variation in size of cone of detection for objects of varying characteristics coupled with an inability to detect walls at sharp angles and table edges make ultrasound a poor choice. With multiple sensors the problems of interference, significant time delays, and increased system complexity arise. EERUF can address the first two issues; however, its implementation significantly increases the complexity of our system. As well, if we consider overall “sensing time” using EERUF does not really solve our time delay problem. For these reasons, ultrasound sensors, by themselves, are not suitable for use with anti-collision systems of a powered wheelchair. Future work will involve evaluating the suitability of other sensor systems such as contact sensors, infrared sensors, active light systems, stereo vision systems, and others. The possibility of using ultrasound sensors in combination with other sensors will also be considered, along with the possibility of creating a modified environment in the home or long-term care facility. APPENDIX Ultrasound sensors have an emitter that sends out a burst of sound energy at a frequency of approximately 60 kHz, which is above the range of human hearing. The pulse travels away from the sensor at the speed of sound and may be reflected back by objects in front of the sensor in the form of an echo. By measuring the time it takes an echo to come back to the sensor, the distance to the object can be accurately calculated using the following: (1.1) where is the distance to the object, is the time it takes for the echo to come back, and is the speed of sound in meters per second in dry air given by (1.2) where is the temperature in Celsius. was measured to be 22 C in the testing environment, resulting in a speed of sound of 345 m/s.

31

The output from the Devantech sensor used in testing is a timing signal (square wave), which remained high for the time it takes for the burst of sound to travel out to the nearest object and come back to the sensor. This timing signal was viewed on an oscilloscope in order to find in (1.1). ACKNOWLEDGMENT The authors gratefully acknowledge the following people who have contributed to the investigation described: G. Griggs, P. Holliday, B. Row, and G. Shoham. REFERENCES [1] S. P. Levine, D. A. Bell, L. I. Jaros, R. C. Simpson, Y. Koren, and J. Borenstein, “The navchair assistive wheelchair navigation system,” IEEE Trans. Rehab. Eng., no. 4, pp. 443–451, Dec. 1996. [2] R. C. Simpson, D. Poirot, and F. Baxter, “The haphaestus smart wheelchair system,” IEEE Trans. Neural. Syst. Rehab. Eng., vol. 10, no. 2, pp. 118–122, June 2002. [3] G. Bourhis, O. Horn, O. Habert, and A. Pruski, “An autonomous vehicle for people with motor disabilities,” IEEE Robot. Automat. Mag, no. 1, pp. 20–28, Mar. 2001. [4] J. Urena, J. J. Garcia, E. Bueno, M. Mazo, A. Hernandez, J. C. Garcia, and V. Diaz, “New sonar configuration for a powered wheelchair,” in Proc. 7th IEEE Int. Conf. Emerging Technology and Factory Automatation, 1999, pp. 113–119. [5] M. Brown and D. Bradley, “Obstacle avoidance system for an electric wheelchair,” in IEE Colloquium Dig., vol. 55, Stevenage, U.K., 1995, pp. 4/1–4/4. [6] S. Fioretti, T. Leo, and S. Longhi, “A navigation system for increasing the autonomy and the security of powered wheelchairs,” IEEE Trans. Rehab. Eng., no. 4, pp. 490–498, Dec. 2000. [7] P. D. Nisbet, “Who’s intelligent? Wheelchair, driver, or both?,” in Proc. IEEE Int. Conf. Control Applications, Glasgow, Scotland, U.K., 2002, pp. 760–765. [8] C. I. Gryfe, A. Amies, and M. J. Ashley, “A longitudinal study of falls in an elderly population: I. Incidence and morbidity,” Age Aging, vol. 6, pp. 201–210, 1977. [9] M. E. Tinetti, M. Speechley, and S. F. Ginter, “Risk factors for falls among elderly persons living in the community,” New England J. Med., vol. 319, no. 26, pp. 1701–1707, 1988. [10] M. C. Nevitt, S. R. Cummings, S. Kidd, and D. Black, “Risk factors for recurrent nonsyncopal falls,” J. Amer. Med. Assoc., vol. 261, pp. 2663–2668, 1989. [11] M. B. King and M. E. Tinetti, “A multifactorial approach to reducing injurious falls,” Clin. Geriat. Med., vol. 12, no. 4, pp. 745–759, 1996. [12] S. R. Cummings, “Risk factors for fractures besides bone mass,” in Osteoporosis, S. E. Papapoulos, P. Lips, H. A. P. Pols, C. C. Johnston, and P. D. Delmas, Eds. Amsterdam, The Netherlands: Elsevier, 1996, pp. 137–146. [13] P. Kannus, J. Parkkari, H. Sievanen, A. Heinonen, I. Vuori, and M. Jarvinen, “Epidemiology of hip fractures,” Bone, vol. 18, no. 1, pp. 57S–63S, 1996. Supplement. [14] J. Magaziner, E. M. Simonsick, M. Kashner, J. R. Hebel, and J. E. Kenzora, “Predictors of functional recovery one year following hospital discharge for hip fracture: A prospective study,” J. Gerontol.: Med. Sci., vol. 45, no. 3, pp. M101–M107, 1990. [15] R. A. Marottoli, L. F. Berkman, and L. M. Cooney, “Decline in physical function following hip fracture,” J. Amer. Geriat. Soc., vol. 40, pp. 861–866, 1992. [16] S. Jaglal, P. G. Sherry, and J. Schatzker, “The impact and consequences of hip fracture in Ontario,” Can. J. Surg., vol. 39, no. 2, pp. 105–111. [17] J. Borenstein and Y. Koren, “Error eliminating rapid ultrasonic firing for mobile robot obstacle avoidance,” IEEE J. Robot. Automat., no. 1, pp. 132–138, Feb. 1995. [18] J. Borenstein and Y. Koren, “Obstacle avoidance with ultrasound sensors,” IEEE J. Robot. Automat., no. 2, pp. 213–218, Apr. 1988. [19] E. F. LoPresti, R. C. Simpson, D. Miller, and I. Nourbakhsh, “Evaluation of sensors for a smart wheelchair,” in Proc. RESNA Conf., pp. 166–168. [20] K. S. Hwang, Y. J. Chen, and H. C. Hong, “Autonomous exploring system based on ultrasonic sensory information,” J. Intel. Robot. Syst., pp. 207–331, Mar. 2004.

32


[21] User Guide for the Micro-Controller Interface Board (MCIB), J. Borenstein. (1999, Nov.). [Online]. Available: http://www.eecs.umich.edu/~johannb/MCIB_User_Guide.pdf [22] G. Lacey and S. MacNamara, “User involvement in the design and evaluation of a smart mobility aid,” Rehab. Res. Dev., pp. 709–723, Dec. 2000. [23] J. D. Yoder, E. T. Baumgartner, and S. B. Skaar, “Initial results in the development of a guidance system for a powered wheelchair,” IEEE Trans. Rehab. Eng., no. 3, pp. 143–151, Sep. 1996. [24] G. Pires, N. Honorio, C. Lopes, U. Nunes, and A. T. Almeida, “Autonomous wheelchair for disabled people,” Proc. IEEE Int. Symp. Industrial Electronics, pp. 797–801, 1997. [25] Y. Murakami, Y. Kuno, N. Shimada, and Y. Shirai, “Collision avoidance by observing pedestrians’ faces for intelligent wheelchairs,” in Proc. IEEE/RSJ Int. Conf. Intelligent Robotics Systems, Maui, HI, 2001, pp. 2018–2023. [26] Y. Kuno, T. Yoshimura, M. Mitani, and A. Nakamura, “Robotic wheelchair looking at all people with multiple sensors,” in Proc. IEEE Conf. Multisensor Fusion Integration for Intelligent Systems, 2003, pp. 341–346.

Tilak Dutta (M’04) was born in Sudbury, ON, Canada, on July 4, 1979. He received the B.A.Sc. degree in engineering science in 2003 from the University of Toronto, Toronto, ON, where is working toward the M.A.Sc. degree at the Mechanical and Industrial Engineering Department and the Institute of Biomaterials and Biomedical Engineering. He is currently a Research Assistant with the Centre for Studies in Aging, Sunnybrook and Women’s College Health Sciences Centre, Toronto. His research interests include rehabilitation engineering with a focus on the issues elderly people face.

Geoff R. Fernie received the Ph.D. degree in bioengineering from the University of Strathclyde, Strathclyde, Scotland, U.K., in 1973. He is Vice President, Research at Toronto Rehabilitation Institute and Full Professor in the Department of Surgery at the University of Toronto with cross-appointments that include the Institute of Biomaterials and Biomedical Engineering, the Graduate Department of Rehabilitation Science and the Departments of Mechanical and Industrial Engineering, Physical Therapy and Occupational Therapy. Dr. Fernie was the recipient of the 2002 Jonas Salk Award and the 2003 MEDEC Award in recognition of his contribution to the quality of life of people with disabilities through the development of innovative technologies.