ORACLE - Our First RobotORACLE - Our First RobotConstruction and Testing
Neither Roy Leslie nor I knew much about robotics or hydraulics when we started so we designed ORACLE from basic principles of mechanical and control engineering. Fortunately the Department of Mechanical Engineering was well equipped for robotics; a workshop with highly skilled technicians and superbly maintained precise machine tools. Through on-going research in the dynamics and control of very long ore carrying trains, the department was also well equipped with minicomputers; both Roy and I had accumulated an understanding of operating systems programming, computer interfacing and data communication techniques.
We decided to tackle the design in stages, gaining experience along the way. Roy started with the follower actuator (picture above), and I tackled the basic software for following an unknown surface and following previously recorded motion paths. Both of these were ready to demonstrate to David Henshaw by the time of his visit in December 1977.
With new confidence, Roy started to wrestle with the compounding design complexities of the virtual centre wrist mechanism (photo above). I use the term 'wrestle with' because the process of selecting and designing a special mechanism like this involves endless iterations of guess work, drawing, cardboard models, analysis, re-drawing and calculation, only to find some detail which has to be changed. It may be a retaining screw which would knock against a bearing at one end of the movement, or a part which would snag wool or a bearing too small to stand the loads.
Computer and software
We chose a Hewlett Packard 21MX-E minicomputer with 64k bytes of memory (below), similar to others already in the department. With a peak throughput of just under a million instructions per second, and about 50,000 arithmetic operations per second, it was slower and had less memory than the original Apple Macintosh home computer. At $25,000 it was expensive but we expected that computers would be much cheaper by the time our research was finished.
I decided to structure the software as multiple independent real time tasks, each operating at a different frequency. The highest frequency level would read sensor signals and adjust the follower actuator set point at an update rate of about 250 Hz. The next level would calculate set points for the other actuators at 125 Hz. A slower level still would calculate the inverse Jacobian matrix, solve the arm geometry, and calculate motion control in cartesian space, to follow the recorded path. This principal real time level would run at about 14 Hz. Finally, motion planning programs and a program to interpret operator commands would run in background. The choice of update rates was an educated guess. But like many other aspects of the first software, they remained almost unchanged through the life of the project.
These update rates were too fast for a commercial real-time operating system, though the RTE real time multi-tasking system supplied by Hewlett Packard was efficiently written. A special purpose operating system was needed. Fortunately, I had already developed a suitable operating system for another project; my system simulated the RTE environment with minimal efficiently written code for just a few tasks (Trevelyan, 1981).
The structure which emerged in the first two weeks has served us well. It was changed in 1988 to implement recursive motion control - see chapter 11.
Arm mechanism
With four joints in the wrist and follower actuator, we needed at least two more joints - six being the theoretical minimum. But the theory ignores the effects of mechanical restrictions on joint movements. Our best attempt at the virtual pivot mechanism provided ±45° of rotation to compensate for slope changes. Even if we had found a way to extend this, we would have found that our fast response actuator would have become much less useful at greater angles. Our four-joint 'surface follower' mechanism still needed another five degrees of freedom for ideal performance, a total of nine joints in all. The only compensation was that the motions needed to provide the extra freedom could have a lower bandwidth - 3 to 6 Hz would be enough.
With more than six joints, the arm mechanism would be kinematically redundant. Part of the redundancy arises from the need for fast dynamic response - the bandwidth problem described above. The follower can be treated as a specialized shearing end-effector for a five-joint robot, and in this way, the apparent redundancy does not pose additional control problems. Unfortunately the weight of the structure, controls and services needed for nine actuators would be a problem, so I looked for ways of using less.
The follower assembly was longer than I expected; the distance from the cutter to the mounting point would be about 500 mm. The workspace for the arm carrying the follower assembly had to be large enough to include the envelope of the follower as it moved around the sheep.
Even a large sheep without its wool looks quite small. With twelve months wool, up to 130 mm long, a small sheep looks big. Now, to imagine the space needed for moving the follower around the sheep, think of a sheep with 650 mm of wool. It would be huge. That is the working space of the arm which we needed to carry the follower actuator.
One theoretically attractive solution was a vast horseshoe shaped arch over the sheep. This arrangement is used by medical CAT scanners. Only two movements would be needed - one around the arch and the other to move the arch along the sheep. While this would give access all over the upper surface of the sheep it presented overwhelming mechanical difficulties. Apart from the sheer size and weight, the sheep centre would always have to be close to the arch centreline - and sheep simply aren't cylindrical. Further, a large structure would be needed around the sheep - it would be difficult to gain access for loading and removing the sheep from its cradle.
(Paul Marsh and Roy Leslie review ORACLE arm under construction)
In the end I chose a mechanism which would allow us to shear the back and one side of the sheep. Responding to the politically imposed focus on the back and side of the sheep, and also to increasing urgency, I chose a design which would be simple, cheap, and quick to assemble. After all, this was to be a test rig - not at all a prototype shearer. We had always assumed that the next machine might look quite different. I built a model using my children's Fischer Technik parts. Only four extra actuators were needed making eight in all. It was ungainly, but my linkage used components familiar to farmers - welded steel and bearing blocks for the pivots. I naively assumed that farmers would relate to it better. It would be a stark contrast to Roy's precision machined surface follower mechanism. Only later did we learn how important appearance can be.
It was March 1978 and Jim Blair raised a concern that was to echo through our monthly project meetings. He proposed building a simpler test rig so we could have some experimental results before the Corporation reviewed the project at the end of the year. He was also worried about spending $200,000 before a single piece of wool was shorn. I argued strongly to continue with the ORACLE design; I thought that we might otherwise become trapped with a second class machine, like CSIRO. I cannot remember Roy Leslie's view, but he probably supported me. Jim and Prof A-W deferred to us as our salaries depended on the review so we had the most vested interest. We pressed on, but the apprehensions remained.
Electronics and wiring
There was a secondary benefit flowing from our choice of hydraulic actuation; the servo valves worked with simple amplifiers which even mechanical engineers could design. My incidental experience with computer interfacing left only the sensors as an area where electronics expertise was needed. Duncan Steven, a lecturer in the Department of Electrical Engineering, provided the help needed and invented a robust, compact and sensitive capacitance sensor with a range of about 15 mm.
Our grant allowed us to take on an electronics technician, Rob Greenhalgh. Rob took on the construction of the computer interfaces, sensors, and servo control circuits. Although the electronic circuits on ORACLE were quite simple, the wiring was not. Eight hydraulic actuators needed three wires for each of two tracks on the potentiometers, three wires for each of two pressure transducers, and two wires for the servo valve. Then there were the sensor circuits too, power supply wires, monitoring points, the remote control button box - the list seemed endless. All the wires came together in a junction box at the end of the trolley Rob had built to carry the computer and electronics circuits. It was made in a hurry, as it ought not to have been, and was too small. But we only discovered that when two thirds of the wires were in place, and undoing them all seemed worse than pressing on.
When we decommissioned ORACLE years later, we found that the trolley was too wide to pass back into Rob's workshop without first being dismantled. When we asked Rob how he moved it out, he simply smiled.
Assembly and Test
Two of the mechanical technicians in our departmental workshop were assigned to Roy to make the mechanical parts and assemble them (the large components were subcontracted to a local workshop for welding and machining). They were soon busy covering my 'agricultural' linkage with hydraulic hoses and tubing. As the weeks went by, my neat linkage disappeared under a growing mass of valves, valve blocks and tubing. The plumbing for a single actuator looked simple enough, but with eight actuators to be joined to their respective valves, check valves, pressure relief valves, filters and so on, pipes were running everywhere.
(Rob Greenhalgh and Roy Leslie with ORACLE - now festooned with pipes and wires)
Most of the 'robot' had been assembled by September 1978. Just as the tension around the imminent withdrawal of CSIRO from the project mounted, so did our own problems. First among these was a series of protracted delays in the delivery of key components. Perth is at the end of a long supply line from the 'technology' centres of the USA and Europe. Parts which would be available from a local warehouse in 'Silicon Valley' could (and sometimes did) take months to reach us through a series of agents and 'middle men'.
By the beginning of October, enough components had arrived to begin the first tests with hydraulic power. The first pressure test on a new machine is always a moment filled with apprehension. Countless pipes have been bent and fitted to manifolds and hose fittings - the hoses have all been carefully cut to length and had their special fittings attached. Stacks of valves, relief valves, check valves and adaptor plates have been bolted together, with all the holes and oil passageways correctly aligned. Then every part has been completely stripped and scrubbed clean, dunked in an ultrasonic cleaning bath and rinsed again to remove microscopic metal filings which would clog and scratch the delicate servo valves. Then oil has been flushed through the pipes for hours to clear out the last of the dirt and contamination. Finally, Roy carefully removed the blanking plates, bolted the delicate servo valves firmly in place, and turned on the main pump.
As Roy gently increased the oil pressure, we anxiously looked for the faintest signs of a leak - an oil smear growing on the side of a valve, a drip appearing at a connection.
At 25% of full pressure, we stopped and checked for leaks. Roy tried the manual control levers on the two main arm actuators, but the pressure was still too low to move the heavy linkage. Still no leaks.
Roy closed the pump bypass valve, and the pressure climbed. Full pressure, and still no leaks. The two main actuators were still fully retracted. Cautiously, he nudged one of the manual control levers - no response. A little further, and then a touch more brought a squeal of protest from the relief valves as the main arm mechanism leapt up. Roy, startled as we all were, let go the valve handles, and the arm silently sank back to its rest position. The manual control valves were just a little touchy, but that could easily be fixed.
Roy gingerly adjusted each of the servo valves. Each valve had an adjusting screw which could be used to slightly open the valve one way or the other. Roy gently turned each of the adjusting screws; each actuator in turn moved gently but firmly to the end of its stroke, raising the arm up under its own power for the first time.
The noise of the pump was something we would all have to get used to. Driven by a 5 kilowatt electric motor, the air was filled with a beating whining hum. Talking was an effort, and as the pump reached the maximum pressure, there was a sudden loud clunk and change in the beat as the automatic bypass valve opened. When the pressure in the high pressure accumulator fell, the bypass valve closed with another clunk and the hum strengthened as the pressure rose again. The clunk-clunk every few seconds irritated all of us, and probably ORACLE too, so Roy soon installed a more expensive proportional control valve which continuously adjusted the pump for the right flow.
With the pressure testing over, peace returned to the laboratory for the long job of connecting the computer and the electronics. ORACLE used three heavy cables with 40 wires each. Ordinary computer cables or telephone cables contain many more wires, and are far lighter and more compact. But a robot moves, so the cable has to be designed for repeated bending, rolling and twisting, being trodden on, and carried alongside oil hoses which crack like a whip when pressure waves radiate away from the control valves. Once again, our supply problems were critical. The 'ideal' cables were available, but on a six to nine month delivery from the USA and a minimum length of 1000 metres. We needed just 30 metres. So we had to make do with an oversize industrial cable designed for mining equipment. It was robust but we had to bend it more than it was designed for, and that meant that the wires inside would break, sooner or later.
Once again, moments of apprehension for all of us. The first test under computer control had to be taken slowly, the servo valve cables being plugged in one by one. In spite of all precautions, there was every chance that a valve connection had been accidentally reversed. If this happens, the valve passes oil to the wrong end of the cylinder and the actuator moves faster and faster away from its intended position until it thuds against the end stop! Sure enough, two connections had to be changed over.
Finally, with all the valves connected and checked, the first moves could be programmed. At long last, after months of preparations, the arm lifted smoothly up, across, down and back again. Days of tuning followed to readjust the actuators, correct minor software errors and to check the arm for correct positioning. The first of many parties of visitors arrived to take a look. One of the first groups included the vVice-Chancellor of the university, Bob Street, a professor of physics and a keen supporter of the project. ORACLE seemed to sense his affinity, and at the start of the demonstration sequence, leapt up towards a very startled Vice-Chancellor, stopping centimetres away before gracefully completing the programmed sequence. We were just as surprised, but ORACLE obstinately refused to repeat its sudden leap when we later tried to find out why it had put on such a performance.
It was March 1979 before the last of long awaited hydraulics connectors arrived. When Roy announced that he was ready to power the surface follower mechanism for the first time, we all gathered around the robot to watch. After so many months of waiting, the robot was finally complete, and we all expected to be shearing our first sheep within a few days. But we were to be disappointed.
Following our success with the surface follower actuator, Roy had designed special miniature rotary actuators for the wrist joints. Although they had been completed months before, we needed special miniature hydraulic hose fittings which had taken so long to arrive before they could be tested. One actuator steadfastly refused to move, no matter what pressure was used and the other was very 'sticky'. Roy soon realized that we had a major problem - the friction inside the actuators was mainly due to the pressure acting sideways on the shafts, and the force on the vane would always be less than the hydraulic pressure.
Roy discussed the problem with our technicians who took the parts away and cut minute oil passageways around the curved edges, from one side to the other. Two vanes could be used, and the oil pressures on each side could be balanced. The yaw actuator needed to turn more than 180° so an extra vane could not be used. But Roy was able to smooth the parts enough to remove most of the sticking. Three weeks later, we tried again, but soon realized that the oil passageways were so narrow that the bandwidth would be much less than we had hoped for.
Meanwhile, another cloud had appeared on the horizon. Reports from CSIRO had raised doubts over the cutter mechanism on which we had all pinned our hopes. Shearers had found it very hard to push through the wool, even at low speeds, and it would not work at all well at the speeds we wanted the robot to run at. Yet the surface follower mechanism had been entirely designed around the cutter, with its small motor. To fit a conventional cutter to the robot, we would need to completely remake the surface follower mechanism, a minimum of six months work. But then we also had misgivings about our own rotary actuators so Roy decided to start work on a new design.
March passed into April, and we desperately needed results for the first Australian Wool Harvesting Conference in May. I fitted a dummy sheep in front of the robot: a steel drum covered in foam rubber, brass mesh and conductive plastic sheeting to simulate sheep skin. My first attempts to shear the drum were far from encouraging. The approach was smooth enough, but as soon as the cutter touched the plastic, it jumped up and down, bucking wildly, and before long I had gaping tears in my simulated sheep, a broken comb, and injured pride. Even the steel drum had been punctured! Repairs took two weeks, and by then I had made several changes to the software. Days before the conference, I was able to make a videotape showing ORACLE shearing its dummy sheep.
In all of this we were joined by Stewart Key, a recent UWA graduate, whom we had persuaded to leave the corporate ladder of the Ford Motor Company and return to WA. We also took on our own mechanical technician, Ian Hamilton. Ian had learned his trade as an aircraft toolmaker in Britain and we soon appreciated our good fortune to have such a high degree of technical skill in the team.
We mounted our first sheep in a rush, and made scant preparations for what was to have been trial number 1. I followed my routine procedure for the dummy sheep tests - all we wanted was one blow. But it was too rushed. I mistyped the command to start shearing and ORACLE lunged forwards towards the sheep. Roy lunged almost as quickly for the emergency control levers, but not quite quickly enough. The cutter pressed heavily into the side of the sheep before Stewart rescued the unfortunate animal. Happily, the sheep made a quick recovery in Jim Blair's back yard, but we went to the conference without shearing results.
We put on brave faces, and Prof A-W skilfully parried the disparaging comments from the conference floor by reciting the following poem ('Grooks' by Piet Hein, Hodder Paperbacks, 1969).
Put up in a place
where it's easy to see
the cryptic admonishment
T.T.T.
When you feel how depressingly
slowly you climb
it's well to remember that
Things Take Time.
We need not have been so worried - CSIRO had withdrawn and the Corporation had placed their hopes for robot shearing with us. The conference turned out to be a great encouragement, particularly as it came just as our difficulties seemed to be overwhelming (Trevelyan and Leslie 1979; Leslie and Trevelyan 1979).
Yet worse was to come.
I needed more time to rework the software to avoid another frightening plunge towards the sheep. I re-arranged the way robot movements were programmed so that there was no possibility of mistyping a command in the tense atmosphere of a trial (Trevelyan, 1981a).
Meanwhile, Roy took the cutter mechanism off the wrist and tested it on a sheep to make sure it would work. He soon discovered that it would not work at all! Stewart and Roy soon realized that the sand and grit in the sheep's wool was to blame. The cutter had been tested in Geelong on nice clean white sheep, washed by frequent showers of rain. Our sheep had come from semi-desert pastoral country and the wool was full of dust and sand.
Roy finally made the cutter work by bending the pivot, forcing the comb
and cutter together more tightly. It seemed an unthinkable mutilation of
such a precisely and carefully made machine, yet it was effective. For
the first few months of shearing trials at least, we would have to make
do. At least the experience confirmed our earlier decision to re-design
the surface follower and fit a more conventional cutter as soon as possible.
|Tour Page|James Trevelyan's
Page| UWA Home page|
Web site design: James Trevelyan jamest@mech.uwa.edu.au
September 1997