Low cost ultrasonic
There is now considerable interest in the use of ultrasonic technology as a means for spatial sensing as an alternative to optical sensing (Kay, 1999a).
The University of Portsmouth has been investigating new control methods for tele-operated vehicles. A powered wheelchair has been used as an example of a special case of a tele-operated vehicle where the operator moved with the vehicle (Stott, 1997; Stott and Sanders, 2000). This work was not aimed at skilled users but at potential users lacking in spatial awareness or the cognitive skills required to drive a powered wheelchair. This work could also be applied to situations where an able-bodied user of a tele-operated system had impaired perception.
Automated guided vehicles (AGVs) have been used to transport goods and raw materials within industrial areas (Goodwin, 1992). Advanced AGV systems have been able to navigate around temporary
obstructions or take different routes to avoid other AGVs. Similar technology has been applied to advanced wheelchair systems (Stott, 1998). To achieve a local avoidance manoeuvre AGVs have been capable of detecting the environment. Sonar sensors using the time of flight principle to measure range have been widely used on AGVs (Hinkelet al.,1988). Some vision-based systems have used a stereoscopic view to detect obstacles but these systems can require time-consuming data processing. As an alternative a laser range finder could provide accurate and fast range information
(Gutmann and Schlegel, 1996).
An automated vehicle can be given a task to complete; such as ``go to room A''. An intelligent wheelchair, when driven by an operator may find the task loosely defined and the task may change during execution. An intelligent wheelchair must make assumptions and re-assess the desired trajectory in real-time but the system must be affordable. It is possible to create a powered wheelchair that is capable of intelligent assistance for a user using computers (Beattie, 1995), or a human operator via a telemetry link (Gundersenet al.,1996). These systems are not in general use, probably due to their high cost and The authors
Ian Stott, David SandersandGiles Tewkesburyare all based in the Department of Mechanical and
Manufacturing Engineering, University of Portsmouth, Portsmouth, UK
Keywords
Teleoperation, Ultrasonic, Navigation, Sensors
Abstract
Describes a new reliable low-cost ultrasonic ranging system to assist in steering a powered wheelchair. Detection algorithms have been created and implemented on a micro controller based stand-alone system suitable for a tele-operated vehicle. The detection uses the gradient of the echo envelope and is resistant to noise and inconsistencies in the detection circuitry. The sensor array was considered as separate sensors, working independently so the system could quickly gather separate sets of range information. These sets were overlaid on to a 2D grid array. The new system is cheaper and simpler than available systems for powered wheelchairs.
Electronic access
The research register for this journal is available at
http://www.mcbup.com/research_registers/aa.asp
The current issue and full text archive of this journal is available at
http://www.emerald-library.com
Received: 8 March 2000 Accepted: 12 April 2000
Sensor Review
Volume 20 . Number 3 . 2000 . pp. 227±234
specialist nature. A track guidance system was implemented at The Chailey Heritage School (Langner, 1996) and has assisted some wheelchair users to progress from dependence on the track for guidance to fully independent movement using an ultrasonic system. The work described in this paper intends to bring high technology assistance closer for many powered wheelchair users or even those who would not normally be able to use a powered wheelchair safely without assistance.
A new sensor system for a powered wheelchair needed to be accurate, rugged, reliable, suitable for attaching to a wheelchair, and most importantly cheap. A human operator was usually the most accurate source of data but the accuracy of a human could be impaired by disability. In these cases a new complementary sensor system was required to assist the operator.
Ultrasonic sensing
An ultrasonic sensor system was selected to provide information to a wheelchair
controller. Ultrasonic range-finding and object-detection systems have used several principles, including: modulated waves (Teshigawara et al.,1989), frequency sweeps (Lindstedt, 1996) or echolocation (Stott and Sanders, 1996; Stottet al.,1997). Kay used a novel sensor array and constant transmission frequency modulation (CTFM) sweeps to assist blind people in detecting their
environment (Kay, 1999b). A CTFM system examines the response of the environment to a frequency swept sine wave. The shape of the envelope returned to the receiver is
determined by the properties of the
environment. It is possible to recognise objects by the frequency spectrum of the received signal. McKerrow (McKerrow and Harper, 1999) successfully differentiated between types of leafy plants with their system.
Single frequency pulse-echo systems can encode information on the pulses in order to reduce interference between different systems. Kleeman (1999) used a known separation between two pulses to identify different systems. The receiver signal was rapidly sampled and the separation read as each pulse train was received. Pulses from other sensors were rejected. Accurate angular resolution was also possible by using two receivers mounted
close to the transmitter and measuring slight time delays between pulses.
Advances in ultrasonic systems have improved reliability and accuracy. This work aimed to use a cheap sensor system that could be afforded by disabled users and to allow for the system shortcomings with in-built intelligence. Although many techniques existed, the pulse-echo or time of flight technique was the simplest method and was used in this work.
Time of flight sensors can generally detect an object but the sensors tend to have limited directional capabilities due to divergence of the transmitted beam (Sanders et al.,1998). The location of the object may be estimated to be lying on a locus described by an arc with radius equal to range, the sides of the arc defined by beam width. Accurate beam widths for transmitters used in this work were obtained experimentally inside an anechoic chamber and are shown in Figures 1 and 2. Symmetry is assumed about the transmitter axis. The data was gathered by measuring an isometric power line around an operating transmitter. Figure 3 shows a test under way in the anechoic chamber.
Figures 1 and 2 show a directional radiation pattern with the transmitter pointing along the y-axis. Half of the beam pattern is shown. Figure 1 shows the beam pattern obtained when the transmitter was mounted in a hard
mounting. Figure 2 shows the transmitter when tested under the same conditions but this time the mounting was faced with 5mm of polyurethane foam shown in Figure 4. The pattern on Figure 1 shows a side lobe starting at 500mm from the sensor face. The foam facing on the mounting reduced the lobe but made the beam wider. The wider beam is consistent with predictions using Huygens principle where each point along an
advancing wave front may be considered as a new source of sound (Frederick, 1965). As the wave front exits through the foam layer then the wave fronts from these sources combine and form a new wave front which continues in the same direction as the original, except at the edges. Here the sound
forms a wave front which originated from the corner and this wave front radiates in all directions. This explains why sound is diffracted around solid edges and narrow beams are difficult to create.
Side lobes could effect performance. Side lobes are the effect produced by constructive interference patterns and appeared as fingers of higher intensity sound radiating from the transmitter. Side lobes caused objects to be detected to the side of a sensor. This caused the system to report a detection in front of the sensor when the object was to one side. Although the beam pattern was wider in the case of Figure 2, the reduction of the lobing effect may defer the disadvantage of the wider beam width experienced with the foam modified beam pattern.
The hardware
The hardware was designed to be simple and low cost. Simple transmitter and receiver circuits were employed.
An experimental test rig was created to test the systems. It consisted of an input/output interface connected to a desktop personal computer. Ultrasonic transmitter and receiver boards were constructed and connected to the interface. A block diagram of the test rig is shown in Figure 5.
Figure 4A transducer mounting with a foam baffle fitted to the transmitter side
Figure 2Directional radiation pattern for a sensor mounting faced with 5mm of foam
A detection was associated with a sudden rise in the voltage measured at the receiver. The receiver voltage returned to a reference level after the pulse reflected from the object had passed. The pulse could be recognised by several characteristics. These included:
. amplitude of the receiver signal exceeding a pre-set value;
. shape of the pulse;
. gradient of the leading edge of the pulse exceeding a pre-set threshold.
A typical receiver voltage signal is shown in Figure 6.
The pulse of returned energy is prominent although noise can be seen along the
reference level. A detection method that used a pre-set threshold voltage was considered along with a method for detecting the shape of the pulse. By inspection of the pulse envelope, it was seen that the pulse had a steep rise (attack) and a gentle decay. The
noise however tended to have a more gentle attack and decay gradient. The most obvious attribute of the pulse was the attack gradient. Figure 7 shows a plot of the gradient of the trace shown in Figure 6. An advantage was that the reference voltage of the detector was not critical. Using this method, detector boards could be swapped without having to recalibrate a reference voltage.
Methods based on gradient were reliable and convenient and were selected as the detection strategy for this work.
A CMOS oscillator circuit tuned to 40KHz. excited the transmitter transducer and its duration was controlled by a micro-controller. The duration of the pulse could be set by external commands or calculated dynamically by the micro-controller in response to changing system demands.
The receiver transducer was connected to a rectifying amplifier and peak detector. The peak detector modified the amplified receiver signal to make it suitable for sampling. The smoothed and amplified signal was converted to a sampled digital signal by an analogue to digital converter (ADC). The
micro-controller demanded ADC samples at appropriate times, analysing the data between samples.
The transducers were cheap and readily available but were tuned to a single frequency and emitted a wide beam. Transmitters and receivers were mounted in pairs. Two mountings were tested, a solid wood fibre based mounting, and a similar mounting
Figure 6Typical receiver voltage plot
faced with a 5mm layer of polyurethane foam (Figure 4).
A block diagram of the ultrasonic system is shown in Figure 8.
The software
The hierarchical structure of the code is shown in Figure 9. This is similar to the hierarchical structure described by (Sanders, 1993). The supervisory level controlled the frequency and order of execution for the lower levels and could interface to other devices to allow integration of the system with other controllers and systems.
Ultrasonic sensors tend to be noisy and return misreads. Methods for filtering out misreads can be used to improve the reliability of the system. A method based on
histogramic in-motion mapping (Borenstien and Koren, 1991) was used in this work to improve sensor reliability.
The area in front of each sensor was divided into a simple grid of three areas as shown in Figure 10. The areas were denoted as near, middle and far and were a simple
representation of the ultrasonic beam and stored as an array in the micro-controller
Figure 7A gradient plot of a receiver voltage
Figure 9Hierarchical structure of the code
Figure 10A simple grid with a beam pattern overlay
memory. When a range was returned from the system, it was classified as being in the near, middle or far grid element. The array element representing the area in which the object was detected was incremented by three. The other array elements were decremented by one. The arrays had a maximum value of 15 and a minimum value of 0. An object occupying a grid element would cause that element to quickly ramp in value to the maximum. Random misreads in the other elements incremented that element temporarily, but the value of the false reads were decrimented each time the system updated. If the object moved to a different element, the new element quickly ramped up to its maximum value and the old element ramped down to the noise level. A reliable range could be acquired within half a second.
A single array
Reliable measurement of range was possible from a minimum of 75mm from the front of the transducers to a maximum of 2m. Longer ranges were sometimes possible with a large target with a good reflecting surface.
Figure 10 shows the simple three-element grid that was used as an example and shows a typical beam pattern for the system.
A three-element grid was sufficient for demonstrating a histogramic method for improving the reliability of the system. The number of cells selected can be dependent on the application. It was simple to modify the software to vary grid parameters and size.
A characteristic of ultrasonic transmitters is the divergence of the beam and their tendency to exhibit side lobes. It was impossible for a single sensor to provide angular information and the location of an object was often uncertain. Figure 11 shows a good reflector located within a powerful side lobe of the transmitted beam. The dashed line shows an arc of the possible location of the object. An object with poor reflection characteristics would not necessarily be detected if it were in a similar location to the good reflector.
Multi-sensor arrays
A sensor array acting in isolation had limited usefulness. More information could be gathered about the environment if more than
one set of sensors were used. Three sensor pairs were mounted as part of an array which created a nine-element grid in front of the sensors. Figure 12 shows the three-sensor array. It can be seen that grids and the beam pattern overlap. However it could not be assumed that an object detected in one grid square would be detected by the corresponding transducer in an overlapping grid.
Detections were reinforced where detection in one grid was matched by a corresponding detection in an adjacent grid. Conversely where an object appeared in one grid but not the adjacent one less confidence existed in the accuracy of the detection.
Conflicts between sensors were useful to gain more information about the environment so that intelligent assumptions could be made.
A representation of the environment in front of the sensor array was then available for analysis. Confidence existed in the accuracy of the grid as the histogrammic mapping reduced the effects of noise on the array. Path planning algorithms could then be used to form a strategy and suggest safe routes (Sanders, 1998).
Figure 11Unreliable detections due to side lobes and the characteristics of the target
Using the information from the sensors
In the case of a tele-operated vehicle, the vehicle received commands from an operator. If the operator could not be relied on for accurate control of the vehicle, the vehicle could help by interpreting the wishes of the operator and responding accordingly. The simple nine-element grid was effective in suggesting ``no-go'' areas for the vehicle. However, there are occasions when a tele-operated vehicle will be required to touch or even crash into an object so the vehicle system must not assume complete control in these situations but allow the operator to over-ride the guidance system.
There were four inputs to the new system; three sensor inputs and one operator input. The sensor system could provide
environmental information. Figure 13 shows an environment where the far-centre and the middle-right grids are ``no-go'' to the
wheelchair. If the vehicle was not moving then the situation could be considered as stable and no action was required. Considering three variations of operator inputs; forward-left, forward and forward-right, Figure 14 represents a strategy employed in this situation. The algorithms could be expanded to include additional variables such as vehicle speed or degree of turn demanded by the operation. The positions of the objects were dynamic with respect to the vehicle and as the vehicle turned, the vehicle trajectory was adjusted.
The prototype system has been used in a noisy environment and although the number of misreads increased, the histogrammic mapping filtered out the miss-reads and a reliable range was acquired. An early
prototype was filmed in operation and a video
is available from the authors. The early prototype used simple avoidance algorithms and the work is now concentrating on advancement of the control algorithms to include higher levels of intelligence within the control loop.
Discussion
A low-cost ultrasonic range finding system has been created. The new system is
significantly cheaper than other systems being used with powered wheelchairs. A simple gradient detection algorithm was used successfully without the need for complex filtering or adjustable gain amplifiers. A micro-controller was used to set timings and analyse the returning waveforms.
The recent increase in the availability of small, cheap and powerful computers has opened up new possibilities for low cost sensor systems. The inadequacies of the sensor system can be offset by the intelligence that can be built into the controllers.
Other advanced wheelchair projects such as the NavChair and SENARIO have used sensor systems and computers to enhance
Figure 13No-go areas on the grid
their usefulness although they are not in general use. A reason for the lack of transfer from industry to rehabilitation technology is the extra cost this technology transfer adds to the price of a powered wheelchair. This work has aimed to produce solutions to
navigational problems at a cost affordable by rehabilitation technology users. If suitable equipment is available at an affordable cost, more potential wheelchair users may be given the chance to drive powered wheelchairs.
Intelligent algorithms can be used for controlling the sensors and to fuse information from several sensor reads. Histogramic detection algorithms can be an effective way to reduce the effect of noise on ultrasonic data. More algorithms could be used to search for objects closer or further away, the controller could automatically adjust the pulse length or output power as the environment changed.
Sensors can be grouped together and data fused together by a micro-controller before passing the new data to a vehicle controller.
Small systems of this type could be suitable for many applications where a reliable, low-cost solution is required.
Work is now investigating the information that can be obtained as the blocked cells shown in Figure 13 change over time.
References.
Beattie, P. (1995), ``The SENARIO (sensor aided intelligent wheelchair navigation) project (tide project 1047)'', digest paper no 1995/055,IEE Colloquium On New Developments In Electric Vehicles For Disabled Persons, Professional Group P6 IEE, London, 17 March 1995, pp. 2/1-2/4.
Borenstien, J. and Koren, Y. (1991), ``Histogramic in motion mapping for mobile robot obstacle avoidance'',IEEE Tran on Robotics and Automation, Vol. 7 No. 4. pp. 535-9.
Frederick, J.R. (1965),Ultrasonic Engineering, J. Wiley & Sons Inc., New York, NY, Cat card No. 65-14257, USA.
Goodwin, M.J. (1992), ``A flexible manufacturing system for education'', MPhil thesis, University of Portsmouth.
Gundersen, R.W., Smith, S.J. and Abott, B.A. (1996), ``Applications of virtual reality to wheelchair remote steering systems'',Proc of the 1st European Conf. On Disability, Virtual Reality and Associated Technologies, Sharkey, P.M (Ed.), Maidenhead, pp. 47-56.
Gutmann, J.S. and Schlegel, C. (1996), ``AMOS: comparison of scan matching approaches for
self-localization in indoor environments'',1st Euromicro Workshop on Advanced Mobile Robots (EUROBOT '96), IEEE Comput. Soc. Press, Los Alamitos, CA, pp. 61-7.
Hinkel, R., Knierieman, T. and Puttkamer, V. (1988), ``MOBOT II ± an autonomous mobile robot for indoor applications'',19th Int. Symp. and Exhibition on Robots, Sydney, Australia, pp. 491-504. Kay, L. (1999a), ``Sonar-now a viable alternative to
vision'',Sensor Review, Vol. 19 No. 3, pp. 178-9. Kay, L. (1999b), ``ACTFM acoustic spatial sensing
technology: its use by blind persons and robots'', Sensor Review, Vol. 19 No. 3, pp. 195-201. Kleeman, L. (1999), ``Real time mobile robot sonar with
interference rejection'',Sensor Review, Vol. 19 No. 3, pp. 214-21.
Langner, M.C. (1996), ``The development of special mobility systems at Chailey Heritage'',Proc. Of The 1st Symposium on Powered Vehicles for Disabled Persons, PCR, University of Portsmouth, pp. 12-17. Lindstedt, G. (1996), ``Recognition using frequency sonar sweep'',Proc. Eurobot 96, IEEE Computer Society, pp. 46-53.
McKerrow, P.J. and Harper, N.L. (1999), ``Recognising leafy plants with in-air sonar'',Sensor Review, Vol. 19 No. 3, pp. 202-06.
Sanders, D.A. (1993),Making Complex Machinery Move: Automatic Programming and Motion Planning, a post graduate text published by Research Studies Press and by J. Wiley & Sons Inc.
Sanders, D.A. (1998), ``Route planning'',Intelligent Mobility, Vol. 8, PCR, pp. 44-9.
Sanders, D.A., Stott, I.J. and Tewkesbury, G.E. (1998), ``Strategies to assist a person in navigating a powered vehicle through a doorway'',Proceedings of Advanced Robotics ± Beyond 2000, Chapter 11 ``Mobile robots I'', NEC Birmingham, pp. 19-21. Stott, I.J. (1997), ``Computer assistance for users of
powered wheelchairs'', MPhil thesis, University of Portsmouth, UK.
Stott, I.J. (1998), ``A review of advanced and intelligent wheelchair research'',Intelligent Mobility, Vol. 2, PCR, pp. 4-12.
Stott, I.J. and Sanders, D.A. (1996), ``Modification of a wheelchair trajectory by the Integration of ultrasonic sensor data'',Proc. of 1st Symposium on Powered Wheelchairs for Disabled Persons, University of Portsmouth, PCR, pp. 4-7. Stott, I.J. and Sanders, D.A. (2000), ``New powered
wheelchair system for the rehabilitation of some severely disabled users'',The International Journal Of Rehabilitation Research, accepted for publication in Vol. 23 No. 3.
Stott, I.J., Sanders, D.A. and Goodwin, M.J. (1997), ``A software algorithm for the intelligent mixing of inputs to a tele-operated vehicle'',Journal of Systems Architecture, Vol. 43 No. 1, pp. 67-72. Teshigawara, M., Shibata, F. and Teramoto, H. (1989),
