| POWER TRANSMISSION AND MOTION CONTROL |
This section was edited by Associate Editor Alan S. Brown.
WHAT ROBOTS LEARN
As the tasks that we ask of robots—especially service robots designed to help humans—grow more complicated, teaching robots how to perform those tasks has grown ever more complex. That’s why robot developers may want to look at the work of a group of Stanford University researchers building autonomous helicopters. They have taught their robotic helicopters to fly the same way humans learn—by copying someone who knows how to do it.
These robots don’t just fly. They perform a whole range of aerobatic stunts, such as traveling flips, rolls, loops with pirouettes, stall-turns with pirouettes, inverted tail slides, and maneuvers with such exotic names as the knife-edge, slapper, hurricane, and the tic-toc, which involves swinging from side to side. Not only do the robotic helicopters perform these tricks, but they do them better than their human pilots.
Just staying aloft is impressive. Unlike airplanes that glide on wings, helicopters are inherently unstable systems. All they really want to do is fall out of the sky. “If you don’t provide feedback, it will crash,” said Pieter Abbeel. He and fellow graduate students Adam Coates, Timothy Hunter, and Morgan Quigley developed the helicopter under the direction of Andrew Ng, an assistant professor of computer science.
At first, Abbeel and Coates tried to teach helicopters aerobatics by writing computer code for each specific maneuver. This proved workable (but not elegant) for novice-level flips and rolls. It failed entirely for the tic-toc and other complex maneuvers. Nor could the robot simply copy the moves of the team’s expert radio control pilot, Garrett Oku. Performance changed with wind speed, sudden gusts, temperature, and humidity.
Finally, the programmers developed artificial intelligence algorithms to analyze Oku’s routines. Even though Oku’s piloting varied with each flight, the AI system was able to abstract the ideal trajectory Oku sought. Eventually, the autonomous helicopter learned to fly the routine better and more consistently than Oku himself.
The helicopter carries accelerometers, gyroscopes, and magnetometers that measure its position, direction, orientation, velocity, acceleration, and spin 20 times per second. The AI system analyzes the data in order to adjust the vehicle’s flight path.
According to Ng, the goal is to create autonomous helicopters that can search for land mines in battle zones or map the spread of wildfires in real time. “In order for us to trust helicopters in these sort of mission-critical applications, it’s important that we have very robust, very reliable helicopter controllers that can fly maybe as well as the best human pilots in the world can,” Ng said.
Stanford’s autonomous helicopters are a step in that direction. They also showcase the type of algorithms that may simplify teaching robots to provide complex services to their human masters.
Editor’s Note: This month’s cover article, “Moving on Their Own,” examines technical issues being researched to make possible future generations of autonomous robotic devices, capable of taking on increasingly complex tasks.
TRACKING IN TWO DIMENSIONS
Optical encoders are devices that sit on moving parts and continuously monitor their position by reading marks on a ruler-like incremental track. The information goes to a controller, which uses it to adjust the part’s speed and location.
But suppose you want to control a gantry or stage in two dimensions? The usual solution is to mount two encoders on the stage. Now, encoder specialist Dr. Johannes Heidenhain GmbH of Traunreu, Germany, offers a second option. It has developed a new encoder, the 1Dplus, which simultaneously tracks a part’s position along a grid that defines a part’s x and y axes.
 A new optical encoder tracks a part’s position in two dimensions simultaneously, and can detect linear guiding errors and thermal drift.
The 1Dplus does this by using two or three optical scanning units. This may sound like an obvious solution, but the problem in the past has always been accuracy. The 1Dplus is accurate to +_- 1 micrometer.
“The sensor is the easy part,” said Kevin Kaufenberg, a product manager at Heidenhain’s U.S. subsidiary. “The difficult part is creating long, accurate y-axis marking lines that run parallel to one another and perpendicular to the incremental track.” The lines themselves are only about 200 nanometers thick, spaced about 8 micrometers from one another, and run the entire length of the track.
It took nearly two years to figure out how to etch such fine lines onto a scale. While most of the technology is proprietary, Kaufenberg said that the process begins with a precision-flat glass substrate. The company uses a system similar to those used to define the circuitry of semiconductor chips to etch a grid of lines into the glass.
An encoder with two scanning units can read the grid and simultaneously measure both the x and y positions. Adding a third sensor enables the controller to calculate the angle of rotation of the bracket that houses the encoder and use the information to compensate for linear guiding errors and thermal drift.
“Let’s say you have an H-stage gantry,” Kaufenberg said. “With three sensors, you could measure any deflections in the middle bar of the H. If you’re using air bearings, you can make sure the center bar is not skewed. If you’re doing a repetitive precision task all day and the gantry starts to heat up, you can also measure and correct for heat-induced deflections.”
Kaufenberg sees potential applications in everything from wafer inspection and wire bonders to medical testing and the production and testing of large flat panel displays. While the 1Dplus costs more than a conventional one-dimensional encoder, it costs significantly less than the two encoders typically used for the same job.
POSITIONER LIGHTENS UP
Baldor Electric Co. of Fort Smith, Ark., has developed a new x-y positioning base that integrates two linear stepper motors into a single plane in a lightweight honeycombed platen, or base. Ordinary positioning bases use two separate sets of motors and tracks, one on top of the other. This makes them bulky and heavy. The new Honeycomb series is thinner, lighter, and flatter, and you can hang it upside down without any additional bracing.
According to product manager John Mazurkiewicz, Baldor’s previous line of dual-axis linear motors was up to 10 inches thick and weighed as much as 140 pounds. Five of the seven Honeycomb platens are only 1.1 inches thick, and the other two are 1.8 inches thick. They also weigh 70 to 80 percent less than similarly sized x-y positioners.
Like other linear motors, the Honeycomb motors have superb accuracy. According to Mazurkiewicz, they move at speeds up to 1.5 meters (5 feet) per second. The positioner has a resolution of 2.5 micrometers and repeatability of better than 2 micrometers (unidirectional) and 5 micrometers (bidirectional) in a four-phase power configuration.
The platen consists of an aluminum honeycomb core sandwiched between two sheets of aluminum. This is the same type of design used to stiffen aircraft wings and tails. Compared with platens made from steel or cast iron, it is much stiffer and flatter. It is also much less massive, so it is easier to mount in tight spots.
The design is straightforward. The moving stage, or forcer, integrates two linear stepper coils mounted at right angles and floats on an air bearing. It acts as the motor’s rotor. The platen contains permanent magnets and acts as the stator. The air bearing supports quite a load, as much as 400 pounds if it is mounted upside down.
According to Mazurkiewicz, Honeycomb positioners cost 50 to 60 percent less than conventional two-axis/two-motor systems. Equally important, they have the speed and accuracy needed for 3-D prototyping and rapid manufacturing, test and inspection measurement, and pick-and-place applications.
ENCODERS FOR DRAG RACING
Magnetic rotary encoders made by Great Britain's Renishaw plc are used as position sensors in a wide range of applications, from Formula One race cars to oil fields. Among the more colorful uses has been in the clutch housing of one of the world’s fastest motorcycles.
 Jaska Salakari of the Salakazi Racing team in Finland competes in the Super Twin Top Fuel class of motorcycle drag racing on a bike made by KTM Power Sports AG in Mattighofen, Germany. Leading vehicles in this class finish a quarter-mile in under 7 seconds.
This takes reaction times measured in milliseconds, and a driver moving that fast over such a short distance doesn’t have time to react and adjust the clutch during a race. Salakazi Racing has equipped its dragster with an automatic Prowork three-disc, four-stage clutch fitted with a Prowork digital controller. When the rider snaps open the throttle, the controller engages the clutch according to how it has been preprogrammed.
The program has to be right because every revolution that a wheel spins without traction is time lost on the track. The team has placed a Renishaw magnetic rotary encoder to monitor the clutch so that after each race the team can evaluate its settings and fine-tune them, if necessary.
 A small encoder enables a motorcycle racing team to analyze clutch shifts during races that last only 7 seconds.
The team uses a compact RM22 system designed and manufactured by Renishaw’s Slovenia-based partner company, RLS d.o.o. The solid-state encoder has a small actuator magnet that fits to the end of a shaft and a Hall sensor chip embedded in epoxy inside a metal housing/mount. The encapsulated design provides resistance to shock, vibration, acceleration, and deceleration.
Heat is also an issue. The RM22 has a maximum operating temperature of only 125°C, and the engine gets much hotter than that. This is where the device’s small size comes into play. According to the team’s technical chief, Petri Mäkinen, the 22-millimeter-diameter unit is small enough to fit inside the machined aluminum clutch housing. The housing is thick enough to provide some thermal protection, and the team removes the encoder quickly after each race to download its information.
MOTOR UNTRACKS MAGNETS
A new linear motor turns the usual recipe for such devices on its head. Linear motors are essentially rotary motors with their stators rolled flat. The shuttle, or primary section, contains an electromagnet. It rides along the secondary section, or stator, containing permanent magnets, like a train riding on a track. The attraction and repulsion created as the shuttle’s electromagnet switches polarity propels the primary section forward.
The new motor, made by Siemens Energy & Automation Inc., achieves motion in a very different way. The new 1FN6 synchronous linear motor series packs both permanent magnets and electromagnets into the primary section. The secondary track is simple steel and houses no magnets of any kind.
“We found out that the position of the permanent magnets did not matter as long as they were within the flux loop,” said Jeff Gerlach, a Siemens consulting business developer. So if the permanent magnet is not on the track, what makes the shuttle move? “The teeth along the secondary track,” Gerlach said. “The electromagnets in the primary induce a magnetic field in the teeth of the secondary, and the shuttle is attracted toward that. There are no magnets in the track, no copper wire, no magic—just magnetic steel.”
This seems like a complicated way of doing things, so why go through all the bother? It comes down to economics, said Siemens’ applications engineering manager, Stephen Czajkowski.
 A new-style linear motor houses all its magnets in its shuttle, simplifying installation of its plain steel secondary section.
Eliminating magnets from the secondary track not only eliminates expensive permanent magnets from long secondary sections, but also simplifies installation. “Workers have to take off their watches, glasses, and anything else that might get magnetized during installation,” he said. Nor do the new tracks attract ferrous chips and debris thrown off by machining. The new technology also removes the need for water lines, tanks, and pumps to cool magnets along the length of the track.
“The system is really targeted for applications with long travels,” Czajkowski said. “After 5 or 6 meters, it is a lot more cost effective.”
Of course, the new design operates somewhat differently from conventional linear motors. Because the primary section contains both permanent magnets and electromagnets, it is more massive than conventional shuttles. While it accelerates smoothly, it is not as fast as other linear primaries. Nor does the air-cooled system generate as much force as other linear motors, although models in the series range from 900 to 8,080 newtons.
According to Czajkowski, potential applications include water jet cutting, especially of large composites, gantries, laser cutting, and aluminum machining. The system supports multiple primaries on a single secondary track moving in the same or opposite directions.
POWER ELECTRONICS GROWING RAPIDLY, SAYS STUDY
The market for power electronics products and systems, which approached $10 billion in 2007, will grow 11.6 percent annually to reach $17.7 billion in 2013, according to a new study by market researcher BCC Research of Wellesley, Mass.
Classic power electronics devices convert electrical power from one form to another, such as alternating to direct current, low to high frequency, and transmission to line voltages. This is generally done with such semiconductor switching devices as diodes, thyristors, and transistors.
But power electronics devices increasingly fulfill other roles. Because they control and condition electrical power, they have a key role in determining the performance and efficiency not only of electronics, but also of motors, pumps, and other electrical equipment. According to BCC, power electronics represents one of the few inexpensive tools that can reduce fuel consumption across the entire industrial spectrum. |
