SIRIUS: Improving the maneuverability of powered wheelchairs

Pmcsedngs ol the MO2 IEEE lntemabmal Canterem on Cmbd ApplicnDw September1820.2W2.G$sgow.~~d.UK SIRIUS: IMPROVING THE MANEUVERABILITY OF POWERED WHEELCHAIRS A. Civit-Balcells, F. Diaz del Rio, C. Jimenez, J.L. Sevillano, C. Amaya, S. Vicente. Escuela Supenor de lngemeria Infodtica Uruversidad de Sevllla. Avda Rem Mercedes dn. 41012 Sevilla Spain Fax +34-954552759. Email cint@alc uses ‘Abspact- The indoor maneuverability of powered wheelchairs may he diflicnlt or bothersome in several circumstances. In this paper, we describe an experimental powered wheelchair named SIRIUS, developed at the University of Seville, which introduces some simple but eKective navigation aids. Special emphasis is placed on the implementation of recorded trajectory playback and in the shared control modes, i.e. the chair’s guiding where both the user and the computer collaborate. Furthermore, SIRIUS is an open platform to essay another kinds of fnnetional or navigational aids, because its hardware architecture is based on a commercial PC. This would permit many devices that are frequently needed by the chair driver to be integrated smoothly into the chair controller. I. INTRODUCTION. Powered wheelchairs are one the most important elements in the field of rehabilitation. Indeed powered wheelchairs are not a luxury for most users but their only way to live independently. Much R&D work in the field of rehabilitation engineering is being focused on the improvement of power wheelchairs [l], [Z], [31, [4], [5]. In this context, automation, robotics and telematics should play a very important role in thc field of rehabilitation engineering. During the last decade advanced wheelchairs have experienced a spectacular improvement from the technical point of view, and many prototypes have been built to overcome the limitations of current wheelchairs, Some of them are: difficult adaptability, high cost, outdated designs, imprecise battery gauges, difficult handling and lack of portability [4]. Besides, a wide range of off-the-shelf electronic devices and sensors can be found today at very low cost. While, for example, most of these devices are being incorporated to automobiles as luxury components, they can play a crucial role in advanced wheelchair aids. Thus, we belief that researcbers should do a special effort to incorporate these devices to powered wheelchairs. An important effort is also required to bring these development to the market as, otherwise, they will not benefit the final users. During the last years in the Robotics and Computer for Rehabilitation Lab at the University of Seville we have developed a system named SIRIUS that embeds many * This wnlr was supported in part by the Spanish “Ministerio de Cimcia y Temologis” through project TlC-2MIO-0087-P443. mobile robotics features [6], [7]. It is based on a conventional wbeelchair but adding sensors and a controller; so it can be teleoperated or navigate in an autonomous way. Moreover SIRIUS can be used as a research platform to test new advanced features: GPS navigation systems, telecommunication devices, miiicial vision-based navigation, etc. In fact, we are working nowadays in a research project in order to incorporate telecommunication systems to standard wheelchairs, and the possibility of interacting with adaptive envkonments. In this paper, and using SIRIUS as a platform, we discuss bow the advanced navigation system (usually found in mobile robots) can help a wheelchair user and, also, how the user can aid an advanced navigation feature. In section I1 we briefly describe the main subsystems incorporated to SIRIUS for improving navigability. We discuss in section ID the navigation aids experimented until now and finally the conclusions are found in section N. n. THE Sms PROTOTYPE. SIRIUS is a wheelchair prototype that has been built based on a conventional wheelchair, and that remains open as a platform for testing different hnovations. The prototype is intended to be a low cost SOM MI domestic chair that should incorporate advanced navigation features. a mTIm -- W Fig. I: Subsystems of SIRIUS 0-7803-73S3/WS17.00 0 2002 IEEE 790 Besides we can not forget that a personal wheelchair can never be a fully autonomous system (like AGV, Automated Guided Vehicles), but a “semiautonomous-teleoperation” system. We mean that we cannot eliminate the user operation mainly because: a) complex navigation systems are prohibitive due to the intended low cost of the commercial wheelchairs, b) fully automatic systems usually are refused by users and c) In houses, there are small innicate spaces where the chair must navigate in. So the user will sometimes have to guide it (or show the trajectoly to the chair the first time), and we are developing a mixed or “shared control” system: control processing is shared between the user and the automatic sensor-based control system 111. The new hardware that the SIRIUS prototype includes has been incorporated into a conventional power wheelchair. The wheelchair has two independently rear driver wheels, which cause turning by their differential speeds, and two front castors or “free-wheels”. Below we describe the main components that have been added, and in Fig. 1 we show a schematic of SIRIUS. PC-bared orchifecfure. We have also reached the wheelchair prototype on the basis of a low cost high performance system. So we have “adapted” all subsystems to go around a common CPU board and bave simplified the electronic devices needed Our CPU altemative is a PC board based chair. If a x86 based main board is used, control circuitry can be incorporated as PCI or ISA based cards. This would also permit many devices that are kquently needed by the cbair driver to be integrated smoothly into the chair controller (for example, voice recognition systems, movement sensors, graphic user interfaces, etc.). Besides, as the computing requirements have increased, the main board has been easily updated with more powerful x86 processors. The only limitation is that low power versions should be used. On the other hand, PC architecture eases software development, and in fact, the PC solution for embedded systems is being more and ’ more common [8]. For PC adaptation we have used our previous experience in the conversion of robot control units, and have maintained SIRIUS as an open platform to improve multisensor coordination and navigation algorithms. In this sense the fmt purpose of SIRIUS was to determine what navigation features might incorporate a conventional wheelchair in order to facilitate the maneuvering in small areas. This will be discussed in next section. Power mofor drivers. The former chair had a pair of relays in order to change the direction of turn. This makes impossible to build a high performance control because the switching delays introduce non-linearity when crossing the zero. We incorporated a new power driver that ensures linearity in both directions and increases efficiency. The driver is based in an H-bridge of MOS transistors, with a pair of BJT transistors as the driver stage, that are commanded hy PWM waves. Optical encoders and counters. SIIUUS gets the position feedback through two optical encoders at each wheel, and saves this information in two 16-bits counters. That permits us a high precision determination of the chair position and an accurate recovering of trajectories in this research prototype (as we show in next section). Although these feedback sensors give high precision, we are considering a simpler feedback, that would avoid the encoders assembly in a prospective commercial chair. A possibility is detecting the intensity changes in the motors blushes, which precision will be accurate enough to allow the recovering of trajectories. User Inteflace. Designing a complete user interface needs a multi-disciplinary approach, inclndiig rehabilitation engineering, robotics, ergonomics, psychology, computer science, etc. For our prototype we did not consider all these aspects. We are using the original joystick and have included a 16 keys keyboard to select the different parameters and actions. We have added a LCD display to inform about the state of the wheelchair. Finally, a special button (the former horn) is used to play hack trajectories. Of come all the input information is digitized to be supervised by the CPU. Fig. 2: Possibilities for low-levelfeedback fiterno1 simple sensoring infomafion. Ultrasound calculating the world position in a changeable environment. transducers are the cheapest way of measuring distance and Although sonar information is not complete, their distance measures can be interestingly exploited as explained in the following section. was done by the user or by the CPU (see Fig. 2). When automatic feedback is not activated the only adjustable “parameter” is the power given to each wheel through the joystick or any other user interface (see Fig. 2, upper drawing). This makes the user the controller, continuously adjusting the joystick depending on the external conditions working with ’Ims protome and discussing them of inertia, floor slope, etc. For example, in the presence of with many and educators we have Of slopes the control system automatically changes the power the navigation aids to be included in order to improve the given to the wheels in order to maintain the desired speed, and to avoid the chair’s huns (more probable when rising a maneuverability of powered wheelchairs. We deliberately have avoided beacons, fixed paths or any slope). This “open-loop manual control” increases the other way of guiding the chair. Most of bibliography difficulty of handling the chair. In fact, with the usual open focuses on autonomous robots that follow fixed paths or are loop control the user “closes” the feedback path nI. NAVIGATION MODES AND EXPERIMENTAL RESULTS. guided by beacons or artificial vision. The last is nonviable in this moment because of its cost and computational requirements. On the other hand, we think that adjusting beacons or fixed paths in a domestic environment that may frequently change is not a good idea. This way of navigation only can fit in places where changes of the scene can be restricted and that have more open spaces like hospitals or industrial environments; this is not the case for a personal indoor wheelchair. The final prototype can he considered lie a teleoperated mobile robot with some amount of autonomy. The complexity of the interaction buman-computer in the navigation has implied the organization of the functionality in layers of control. We have followed the classical layers proposed in [9], but adding the user action as a new agent that can interfere with them. That is, from the most “intelligent” or highest to the lowest we tind Task planner. Obviously the user always must do this in wheelchairs. Trojectoiy planner. This can be done by the user (e.g. selecting a fued trajectory like a straight line). But the computer for example can do it if the user had selected a task lie “play back a trajectory”. In this case, this planner will retrieve the last memorized trajectory and it will send it to the lower level. Trajectoiy generator. It generates the next point in the trajectory sent by the immediate upper level. Although the CPU (lie in mobile robots) usually does this generation, if the user moves the wheelchair through the joystick, then hdshe is constmcting hidher own haiertnrv - 7 -----, . Controller. This level tries to converge real posture to the desired posture given by latest level. From the preceding. organization we can distinguish two kinds of feedback, which can be alternatively done by the user or the computer: 1) The low-level feedback, done at the controller level. 2) The high-level feedback, done at the trajectory planner level (in SIRIUS the task level feedback [cooper95]. Otherwise, when a computer is used to conml the wheelchair (see Fig. 2, lower drawing), the joystick command can be the linear and turning velocity (other possibilities include converting joystick commands into distance). This computer based feedback control makes the chair motion more intuitive and independent of external conditions. Here the speed of the drive wheels is sensed in order to obtain feedback information. Furthermore, the existence of an automatic low-level feedback can serve to tune system parameters to the user’s needs. This improves the adaptability of the chair making possible a “tunable control”. For future work even the input signals from the user interface could also be pre-conditioned, for example to neutralize intention tremor. The second group of essays was based in the need to relieve the user. We mean to say that in several situations, the difficult or bothersame guiding of the joystick (or any other interface) can be substituted for prefixed trajectories. We have concluded that three trajectories were especially interesting for the users: straight ahead or back, slight changes of path’s curvature, and 90 degrees turns. These trajectories are selected in SIRIUS through the keys 1 to 9 (‘2’ forward, ‘8’ backwards, ‘4’ 90 degrees left turn and ‘6’ 90 degrees right turn, as shown in Fig. 3). When moving in a corridor, the selection of a straight path can be very comfortable. In this case it is necessary to add some mechanism to neutralize the little orientation’s deviations of the wheelchair. We have found two simple mechanisms, which implantation will not increase the wheelchair’s cost: slight changes of path’s cuvature that can be chosen by the user (for example in keys 1,3,7 and 9 for SIRIUS), or continuous lateral distance measures with an ultrasonic sensor. Changes of lateral distance will mean changes of orientation. The last mechanism bas been fully tested in AGV’s, where the sonars are frequently the only sensor device to follow a wall [IO]. Sometimes they incorporate a full array of sonars around the robot (furthermore they are intended for other reasons like maps building). Although is always done by the user). In the first trials we show the difference when activating or not the controller feedback, i.e., when the low-level control some works like [ 111 have considered this pckbility, we think that this option is not obligatory in wheelchairs because of its complexity and due to the problems 792 associated with sonars multiple reflections and detection faults (e.g. if the object is as narrow as some table legs). Moreover building a map may not be a good idea in environments like some users’ homes, which may frequently change. (supposing the initial wheelchair orientation is facing the door under 30”, the maximum detection angle of sonars). In addition the user that guides the joystick has the impression that “someone” is helping himher. OOOOl Fig. 3: Correspondence behveen keys andprejtxed trajectories. The third group of tests was done with sensors for obtaining external feedback information. Due to the high precision of odometric internal sensors (optical encoders at each wheel), we do not consider complicated sensoring to get complete information, but the simplest ones, which are sufficient to help the internal sensors’ guiding. We finally decided to incorporate four ultrasonic sensors: two frontal (at left and right sides) with high reach distance and two lateral. The lateral ones were intended for guiding SIRIUS through a corridor as explained before. The hntal ones can be inactive or prog”ed in two main ways: Obstacle detection and progressive speed reduction used to approach an object facing it. We mix two different effects: velocity reduction and tum. The speed is reduced as we approach the obstacle, and each wheel’s speed reduction is proportional to the inverse of the distance to the object detected by each sonar. The different speed reductions translate in small huns that faces the chair to the object. Obstacle avoiding without changing the main wheelchair orientation (very useful to cross a door) saving the user from bothersome guiding. Here we also mix the velocity reduction and the huning, but in a different way. The speed is reduced as we approach the obstacle, and at the same time when one of the sonar detects a wall but the other does nof a small tum is performed to avoid the door walls (for example, the automatic right turn in Fig. 4). The small tums are recorded; when the obstacle are lost the old orientation is resumed (the iinal left tum in Fig. 4). This is done without user intervention. However if the user is guiding the joystick, the automatic correction of orientation helps the user to cross the door. This is simply done by adding the joystick command to the automatic command (with a scale factor). The result is that is almost impossible to collide with the door walls Left sonar I detection Fig. 4: Obstacle avoiding helps to cross a door. However ultrasonic sensors suffer from several drawback they do not work properly in very noisy environments, they have a limited sensing range with significant minium distance, etc. For these difficulties it is easy desirable that the sonars information will be combined with other sensors. The last group of navigation aids is the most automated playing back recorded trajectories. The main intention of this is to avoid the user the difficult maneuvering of reverse driving, and may be very useful in small areas like bathrooms. Therefore the CPU is always memorizing the trajectory done by the chair (no matter which agent has done it). For example, in Fig. 5 the CPU will have memorized the path from point 1 (initial position) to 2 (final position in the bathroom). When the user wants to play back the trajectory (e.g. when he wants to go out the bathroom), he/she pushes a button and the wheelchair began to automatically move in the reverse direction and recovering the last movements. For example, in Fig. 5 SIRIUS recovers the trajectory from point 2 to 3 (position after first recovery). When the user decides that the path recovery was enough (e.g. if the chair is out of the bathroom at point 3), he/she can press one of the stop buttons, and the chair leaves this automatic mode and stops. From here the CPU continues again to record the trajectory and the recovered path was lost (fragment 2 to 3 in Fig. 5). Therefore if in some meters the trajectory recover button was pressed again, the recovery will be done forgetting the previously recovered path. Then if the button were pressed in point 4 of Fig. 5, the recovery will pass through points 4, 3 and 1, forgetting the previously 793 recovered trajectory (inside the bathroom). This permits to “nest” the repetition of trajectories. 4. Final Position 3. Position afler I. Initial Position first recovery f 2. Final position in the bathmom f 1 Fig. 5: Playing backpaths in SIRIUS: thepossibilify of nested recoveries In order to play again a previously recorded trajectory we must choose a control law for the two driven wheels, considering SIRIUS as a mobile robot. This is a classical problem in non-holonomic systems lie mobile robots [12] and several solutions have been proposed. We have preferred to use the “path following” approach which ignores the time to follow the trajectory [12]. In our system this time is of no concern, and this way we avoid the dradacks of considering desired time evolution (“trajectory tracking” approach) such as discontinuities in transient responses. The whole procedure for selecting a good control law for unicycle robots (like SWS) is explained in [ 131 and it is far from the scope of this paper. The experimental results of paths recoveries are good enough when these paths does not have many rotations. But we have observed that the orientation errors due to slippage have to be considered. Although the control law ensures that robot follow the intemal memorized path with little errors, slippage errors could accumulate and the deviation from the original path may be 4 or 5cm for each 1Om of path. Note that the controller tracks the path using intemal odometric feedback, i.e. it can not estimate slippage errors. Therefore we must introduce some mechanism of extemal feedback to correct this problem. In this field there is lots of research in fully autonomous AGV’s, but all of them implies a complicated sensoring information and processing, as one can imagine. Although there have some trials of this kind in advanced wheelchairs (ulfrasonic array in [Ill, stereovision in [3], etc.) we have incorporated a very simple solution: the user himherself can introduce little corrections to the automatic recovery. That is, if the user sees that the chair is deviated some centimeters for example to the right, he/she can push the joystick to the left to inform the computer of the deviation. The more helshe pushes the joystick, bigger will be the correction introduced in the trajectory. It is a fact that the user usually overestimates the deviation and then in a few meters hdshe acts again (in the opposite direction) until the chair recuperates approximately the original path. This humancomputer cooperation can be considered as another kind of shared control [I]. Original path ----_ Path recovering Right deviation: the user pushes the joystick to the left Fig. 6: Shared control in a path recovery. N. CONCLUSIONS. the maneuverability of powered wheelchairs. All these features have been included without eliminating the user 794 navigation systems, b) the user’s rejection to fully autonomous systems, and c) the difficulty small spaces where the chair must navigate. The main conclusions are centered in a) The advantages of closed loop control (via internal adometric feedback) that reduces bothersome guiding in many cases; b) The benefits obtained when mixing computer and human control. Here we present two cases: computer correction of the user guiding via simple sonar information (very efficient to cross doors) and user correction of automatic path recovery (correction of slippage deviations, very difficult to detect via external feedback). REFERENCES 1. Cooper, R A. Intelligent Control of Power Wheelchairs. In: IEEE Eng. in Medicine and Biol. July, 1995. 2. Bourhis, G., Pino, P. Mobile Robotics and Mobility Assistance for People with Motor Impairments: Rational Justification for the VAHM Project. In: IEEE Trans. on Rehabil. Engin. Vol. 4, No. 1, March 1996. 3. Yoder, J.D., Baumgartner, E.T., Sk, S.B. Initial Results in the Development of a Guidance System for a Powered Wheelchair. IEEE Transactions on Rehabilitation Engineering., Vol. 4, No.3 September 1996. 4. Civit Balcells A., Diu del Rio F., Sevillano J. L., Jimknez G. SIRIUS: A Low Cost High Performance Computerized Wheelchair. Proc. of the Int. Workshop on Medical Robots, pp. 23-30. Vienna. October 1996. 5. ANSI/RESNA. Subcommittee on Wheelchairs and Transportation (SOWHAT) Meeting Minutes. Mesa, AZ, 1995. 6. Civit-Balcells A., Diaz del Rio F., Sevillano J. L., Jiminez G. SIRIUS: A Low Cost High Performance Computerized Wheelchair. Proc. of the Int. Workshop on Medical Robots, pp. 23-30. Vienna. October 1996. 7. Civit-Balcells, A., Diaz del Rio, F., Jmhez, G., Sevillano, J.L. A Proposal For A Low Cost Advanced Wheelchair Architecture. The 4th European Conference for the Advanc. Techn. AAATE Conference 1997. ISBN 4274901831C3047 (Ohmsha). ISBN 9051993617 (10s Press). Oct 1997. Thessaloniki. Greece. 8. Hennesy, J. The Future of Systems Research. IEEE Computer. Val 32. August 1999. 9. Cox, I. J. Blanche - An Experiment in Guidance and Navigation of an Autonomous Robot Vehicle. IEEE Trans. on Robotics and Automation, Vol. 7, N0.2, April 1991. 10. Lee, D. The Map-Building and Exploration Strategies of a Simple Sonar-Equipped Mobile Robot. Cambridge University Press. 1996. 11. Fioretti, S., Leo, T., Longhi, S. A Navigation System for Increasing the Autonomy and the Security of Powered Wheelchairs. IEEE Trans. Rehabilitation Eng., Vol. 8, No. 4, Dec. 2000. 12. Canndas de Wit, C., Khennouf, H., Samson, C., Sordalen, O.J. Nonlinear Contml design for Mobile Robots. In Recen!s Trenh in Mobile Robofs. Ed : Y.F. Zheng. World Scientific 1993. 13. Dim del Rio, F., Jimenez, G., Sevillano, J. L., Vicente, S., Civit Balcells, A. A Path Following Contml for Unicycle Robots. Journal of Robotic Systems 18 7, 325-342.2001 795