By Jesse Jackson and Jerry Burton
[Editor: This is a reprint of two articles from the March and April 1993 Encoder.]
A primary problem in mobile robot design is endowing robots with enough environmental awareness to make intelligent movement possible (like running slaloms or mazes). One step toward solving this problem is to acquire information about ranges and bearings to nearby objects and to interpret that data.
Though there are many approaches, they can be grouped into two categories. The first are passive devices, such as stereoscopic vision and swept-focus ranging systems. The second are active devices, such as Laser, microwave and ultrasonic rangefinding systems.
Our club robot, RSSCy, has been a work horse for all kinds of experiments. In this article we will cover what we've discovered from experiments with RSSCy and other RSSC members' robots.
Since a 20° F temperature differential contributes only 7.8 inch error in 35 ft we will ignore temperature and altitude effects.
We used ranging modules made by Texas Instruments for use with the Polaroid elecuostatic transducer because of their low cost, high reliability, and ease of interface. Polaroid offers both the transducer and ranging module circuit board for only $40 (1993 pricing) a set when purchased in quantities of ten. Texas Instruments developed an improved version of the circuit board which greatly reduced the parts count and power consumption, and simplified computer interface requirements. Texas Instruments no longer produces the chips but they can be obtained directly from Polaroid.
The Polaroid ranging module is an active time-of-night device developed for their cameras to allow automatic camera focusing. It determines the range to a target by measuring elapsed time between the transmission of a "chirp" of pulses and the detected Echo. The one millisecond chirp consists of four discrete frequencies composed of 8 cycles at 60 KHz, 8 cycles at 56 KHz, 16 cycles at 52.5 KHz, and 24 cycles at 49.41 KHz. This pulse train is designed to increase the probability of signal reflection from a wide range of targets. Certain conditions and surface characteristics of the target can in fact cancel a return of a single-frequency waveform.
This one millisecond length of the chirp can also be a source of potential error. Sound travels roughly 1100 feet per second at sea level or about 13 inches per millisecond. The timing of the transmit pulse to the echo always begins at the start of the chirp. The uncertainty and thus the potential error arises from the fact that it is not known which individual pulse of the four frequencies making up the chirp is the first to return to trigger the receiver.
The second important characteristic of the Polaroid system is both the gain and the Q of the amplifier increase as a function of time following chirp transmission. This ensures a high signal-to-noise ratio by increasing receiver gain levels to compensate for inverse fourth power fading of echo returns.
The ultimate goal is to obtain a range map of objects surrounding the robot. To do this you must either sweep the ultrasonic beam over the coverage area, or you must have multiple transducers. There are advantages and disadvantages of both approaches. The power and the real time constraints seem to make a mechanically positioned sensor the less desirable of the two concepts, because of the slew time delay and energy used while the sensor is being repositioned to provide range in a new direction. On the other hand, using multiple transducers entails overhead in terms of physical space, transducer power consumption, interface circuitry, and cost. Our club robot originally had one transducer on the head (355 degree rotation) with three other fixed position transducers, one downward-looking at the fore to detect dropoffs and two side looking 45" collision detectors. Our next effort was a hybrid six transducer array. Two transducers mounted on the robot's rotating head platform can be panned ±170° from centerline and tilted ±75° from horizontal. The two transducers beams overlap, allowing increased resolution by "beamsplitting." Two similiar dual transducer arrays cover the sides of RSSCy. The scanning array complements the fixed arrays ensuring full coverage.
Jerry Burton's new robot K9 uses eight transducers which can be sampled under software control. It is also a hybrid with two transducers on moveable "ears," plus one forward-looking transducer, one rearward-looking, and four side-looking (two on each side). To reduce K9's cost, Burt uses a multiplexing scheme; only one transceiver board is required for eight transducers. The Transmit/Receive signal of the transceiver board is multiplexed through eight relays. We chose TTL compatible reed relays because the pulse voltage to the transducer is 300 volts or more. The controller, a 68HC11 drives a 3 to 8 mux to select the appropriate relay and transducer. We think that a PIC could handle the specialized job of selection, timing, and I/O to the Robot Main Processor, further reducing costs.
With all these transducers, we've got the problem licked, right? Not hardly.
|Rarely does range measured correspond to the distance along beam centerline, as shown in Figure 1. Ultrasonic beams aren't like pencil thin laser beams. At a distance of 15 ft. from a flat target, with an angle of incidence of 70 degrees the theoretical error could be as much as 10 inches, since the actual line of measurement intersects the target surface at point B as opposed to point A. The problem is further complicated for surfaces of irregular shape.||
|The width of the beam causes not only uncertainty in the measured object range, but also uncertainty in the angular resolution of the object's position. A very narrow vertical target such as a long wooden dowel would appear to tbe sensor to be a much larger obstruction. Using a 1-in. diameter vertical dowel as a target, the effective beam width of the Polaroid system was found to be 36 in. at a range of only 6 ft. Worse yet, an opening such as a 36" doorway isn't discernible at all to the robot from 6 ft, away. At that distance the beam is wider than the door opening (figures 2a and 2b). The width of the beam also limits the ability of the acoustic sensor to look deep into a corner. The return from the side Iobes are reflected first thereby reducing the range reading. Corners become "rounded."|
Some experimenters have reduced beam width by focusing the acoustical transmission using various combinations of reflectors, tubes and horns. The reflectors do work but they can be ten times the diameter of the transducer, too unwieldy for a small robot.
Tubes and horns also work and they can be more compact. We've tried long horns, short horns, funnels of many sizes, shapes, and aspect ratios. Our most important discovery of this exercise was that the hero or tube must be lined with a soft felt material to atteuuate the side Iobes of the main beam. By these techniques, the beam width can be reduced to about 10 degrees or less.
The beam width can also be reduced or angular accuracy increased by reducing the sensitivity of the receiver circuitry. The side lobes of the transmitted wave (see fig 3) are about 30 dB down in power from the main lobe. Whith receiver gain reduced the angular resolution can be increased by a factor of two or more.
We have been toying with the idea of generating a variable gain receiver, driven from the processor. The program would allow full gain until receiving a response, then processor would reduce gain to improve on the angular accuracy of the response. The processor should be able to determine if the return was from the main beam or from a side lobe.
|Another significant error is specuIar reflection, which occurs when the angle of incidence of the beam falls below a certain critical angle. Below the critical angle the reflected energy does not return to strike the transducer (Figure 4). This occurs because most targets are mirror smooth at the 0.25 inch wavelength of ultrasonic energy. In specular reflection, the angle of reflection equals the angle of incidence; in diffuse reflection, energy is scattered in various directions by surface irregularities. The critical angle is thus a function of the operating frequency chosen and the smoothness target. For the sensors used on RSSCy this angle is approximately 65 degrees for a flat target surface of unfinished plywood such as would be encountered for maze walls or for dividers in the wall following contests. In Figure 4a the ranging system would not see the target and would indicate instead maximum range, whereas in Figure 4b the range reported would reflect the total round trip through points A,B, and C as opposed to just A and B.|
The relatively long-range capability (approximately 35 ft.) of the Polaroid system makes it well-suited to gathering range data for both navigational planning and collision avoidance. Navigational planning involves determining the actual location of the robot and subsequently calculating the appropriate commands to move it to a new location and orientation.
The simplest case reduces the problem to two dimensions with a priori knowledge of the surroundings in the form of a memory map, or world model. The task becomes one of trying to correlate a realworld, sensor-generated image to the model and extracting position and orientation accordingly. Several factors complicate the problem.
For one, the real environment is three dimensional, and although the model represents each object as its projection on the X-Y plane, the sensor may see things differently, complicating the task of correlation. Second, large computational resources are required and the process is time consuming, requiring the robot to stop and think occasionally. Also, acquiring the data can take several seconds using ultrasonic ranging techniques, due to the relatively low velocity of sound waves in air. More important for the purposes of this discussion, however, are the effects of the various error sources previously described, which can act collectively to impede a solution.
Figure 2a depicts the results of 30 range values taken by a single sensor mounted on an azimuth table, with the sensor approximately 5ft, from the wall.
The exceptional quality of the plot is due primarily to the nature of the walls having a pebble grain finish that provided excellent beam return properties. The proper identification of the open doorway and the excellent correlation with the actual map would provide the robot with a highly accurate "fix". It should be noted that the room was fairly uncluttered, which is not always the case presented to our robots. In Figure 2b, the sensor was repositioned 7ft. from the wall and was unable to detect the opening.
For such situations, the robot needs help from other types of sensors of some type of narrowing or focussing of the beam.
Collision avoidance is a little easier to address, in that angular accuracies are less important and the computational overhead nowhere near so great. The intent is simply to be aware of obstructions in time to alter course. For this application the sequential array can outperform a single sensor, in that the array permits range measurements to be made in many different directions very quickly and with minimal-power consumption.
|Sequential arrays can use beamsplitting to improve the angular resolution, already shown to be some what poor for a single transducer. Beam splitting involves the use of two or more range finders with partially overlapping beam patterns. Figure 5 shows the simplest case of two transducers, twice the angular resolution can by obtained, along with a 50 percent increase in coverage area. The technique is simple. As the target is detected by both sensors A and B, then it (of at least a portion of it) must lie in the region of overlap shown by the shaded area. If detected by A but not B, the target lies in the region at the top of the figure, and so on. Increasing the number of sensors with overlapping beam patterns decreases the size of the respective regions and thus increases the angular resolution. The sensor pattern used on RSSCy allowed for an angular resolution of less than 20 degrees when locating a i-in, vertical dowel 9 ft. from the robot, a signifrcant improvement over the 30 degree resolution of a single transducer. It should be noted, however, that this increase in resolution is limited to the case of a single target in relatively uncluttered surroundings, such as a box in the middle of the floor. No improvement is seen for the case of an opening smaller than an individual beam width, such as the doorway illustrated in Figure 2b. The entire beam from at least one sensor must pass through the opening without striking either door post in order for the opening to be detected, and the only way to improve resolution for this case is to decrease the individual beam widths by changing transducers or through acoustical focusing, which sometimes is impractical.|
|Timing Operations To simplify the circuitry, all timing and time-to-distance conversions are done in software. Three control lines interface the Polaroid ultrasonic circuit board to a microprocessor. The first of these, referred to as VSW (Figure 6), initiates operation when brought high to +V. A second line, XLOG, signals the start of pulse transmission, while MFLOG line indicates detection of the first echo. The controlling microprocessor must therefore send VSW high, monitor the state of XLOG, commence timing when transmission begins (approximately 5 ms later), and then poll MFLOG until an echo is detected or time out when sufficient time elapses to indicate the absence of an echo. The eight ultrasonic ranging units are interfaced to the microprocessor through a 3-circuit 8-channel multiplexer using single pole relays operating in the digital mode, as shown in Figure 7. This way the microprocessor "sees" only one ranging unit at a time through the multiplexer. Three I/O lines from the 68HC11 handle this enabling function. The binary number placed on these I/O lines by the microprocessor determines which channel is selected. Three other I/O lines carry the logic inputs to the microprocessor for VSW, XLOG, and MFLOP. The implementation of the sequential ranging array using a small dedicated microprocessor offers several advantages to the mobile robot design,|
Using RSSCy electronics as the base for the paralleling experiment, two transducers were connected in parallel to the drive electronics. No degradation of the operation was observed. This technique can be used to increase the detection coverage tie. wider angular range) beyond that of a single transducer. Each transducer looks to the drive electronics an a capacitor. We would think that even more additional units could be driven simultaneously as long as the drive capability of the small transformer is capable of applying the 350 volt pulse to all the units.
We suggest obtaining one of the Polaroid evaluation kits for your club if members are interested in getting into ultrasonic acoustical ranging. The kit that has been available for years is a self contained kit with the timing board, readout board ultrasonic transducer and battery pack. The read out board provides distance infermation on a small 3 digit LED unit. The information books included with the kit detail the timing circuits and the readout board capability. It also gives detailed directions on modifications for changes in range and readout. Polaroid has just announced a second developers kit that contains the timing board, microprocessor board, software, ultrasonic transducer and battery pack. In place of the read out board it includes a microprocessor board with all the support software to write programs that allow control or modification of all the parameters needed in ultrasonic acoustical ranging systems.
Several prepositioned sensors seem to be superior in performance to mechanically positioned single sensor systems, in that they allow data to be taken at a faster rate, with less power consumption, and with fewer errors associated with actual sensor position. Improvements in angular resolution can be gained through the use of beamsplitting techniques, and temperature/altitude correction can be employed to increase range accuracy. Properly designed horns or tubes can be used to reduce the beam width of the transmission thereby increasing the angular accuracy of the return. Gain compensation (reduction) can also help increase the angular accuracy by eliminating responses to side lobe energy. The array concept exploits the properties of ultrasonic ranging for collision avoidance or object tracking, where absolute accuracy is less important than relative information. For other applications in which precision is an important factor, such as navigation and map correlation in cluttered environments, complementary sensor sets with appropriate characteristics probably should be added. The relatively long wavelength, poor angular resolution, temperature dependence, and slow speed of sound in air can be significant drawbacks in the field of precision navigation or mapping. Infrared and laser based rangefinders could be considered as alternative or supplementary approaches. Ultrasonic sensors when carefully applied, still remain the lowest cost and simplest means of obtaining both collision detection and longer range navigation information for our robots.
Lists Panasonic Ultrasonic Microphone three types separate transmitter and receiver
Masse Products Corp of Hingham, MA lists narrow-beam/no side lobe ultrasonic transducers in the 50, 100, 150, and 215 IrHz transducer frequencies.
|604142||Transducer lnstrument grade|
|615077||6500 Ranging Board|
|614095||Coil for Ranging Board|
|614096||Transformer for Ranging Board|
|614904||Digital Chip TL851 (6" to 35')|
|614905||Polaroid Digital Chip (2'to 24')|
|614906||Polaroid Analog Chip|
Jesse Jackson is a Registered Professional Engineer and is the Editor of the Robot Builder newsletter. Jerry Burton is the president of the Robot Society of Southern California (RSSC) and owner of Processing Innovations, a company manufacturing smart modems.