LESSON 8 - TELEPRESENCE AND TELEROBOTICS
TELEPRESENCE AND TELEROBOTICS
One form of virtual reality is telepresence. A user of a telepresence system immerses himself in a virtual world which exists in reality at some other physical location. His world may be an exact representation or a simulation. Such a system allows the user to observe another section of the real world.
Telepresence normally does not include interaction with the distant world. If any user control is allowed at all, it is usually just over viewpoint or perhaps navigation. A security guard watching a TV monitor has a form of telepresence system. He can observe a distant location. In some such systems the guard can rotate the camera or zoom in on a particular area. This is point of view and navigation. If he observes something untoward, however, he can take no direct action.
An interesting example of telepresence is the collection of real time cameras available over the web. There are hundreds of these cameras which supply real time shots from locations around the world. A list of some of them may be found at http://citynight.com/cams.html and another at http://132.183.145.103/cam.nclk. For example, a real-time view of Niagara Falls is available at http://fallscam.niagara.com/
Telepresence
systems are often used with robots for remote control. Here the user can interact with the
distant real world through the robot. The robot may be thought of as a type of actuator
for the VR system. Such systems are used for controlling robots in undersea operations, in
nuclear power plants, or in contaminated chemical areas.
FEEDBACK
When providing control of the robot it is generally necessary to also include feedback to the operator. Feedback is information flowing in the opposite direction as control signals. The operator directs the robot to move right; feedback tells the operator how far the robot has moved.
For a better understanding we will examine a generalized feedback system. Such a system contains a plant. The plant is the operating portion of the system, for example the robot. Control inputs are applied to the plant and certain actions result. If this is the entire system, it is said to be an open loop system. In an open loop system, if some outside force affects the operation of the plant, the final output will be different than was predicted for the particular control signal used.
An example of
such a system would be an automobile with a fixed accelerator petal setting. Suppose on a
flat road a petal setting of one inch results in a speed of 55 mph. If the car comes to a
hill or encounters a headwind, the actual speed will drop because additional energy is
required to overcome these outside forces. If the car comes to a down grade its speed will
increase. This system is operating open loop.
Suppose now that a measurement is made of the final output of the plant. In the case of the car described above the final output is speed (mph). A speedometer indicates a value for this quantity. Now when the car approaches an uphill grade and begins to slow this information is communicated to the driver. This information is feedback and the speedometer comprises the feedback portion of the system. The driver can increase the petal deflection, causing the car to return to its desired 50 mph. In this case the system is operating closed loop with the driver closing the loop by his action of increasing the accelerator setting. If this is carried one step further and a cruise control added, hardware will perform the same function as the driver and the system itself is a closed loop system.
Note that in an open loop control system the control input remains constant and outside forces cause the output to vary. In a closed loop system outside forces cause the input control signal to vary and the output to remain constant.
One major characteristic of feedback system is their stability. Using the car and
cruise control system as an example, consider three possible cases of performance for
three different versions of the system.
Suppose that the car
is traveling at 40 mph and the driver suddenly changes the command input to 55 mph. (This
type of sudden change in input value is known as a step input.
The behavior of a control system is often described by its response to a step
input.) The cruise control will increase the accelerator setting to increase speed.
One of four things can happen. First, the car can steadily increase in speed until it just
reaches 55 mph and settle to this speed. This is the optimal case.
Second, the car can increase in speed but overshoot by, say, 8
mph to 63 mph. The cruise control reacts to this and decreases the accelerator setting and
the car slows past 55 mph. By the time the cruise control has reacted, the speed has
dropped to 51 mph. Next the car increases speed and gets up to 57 before the cruise
control begins slowing. This series of over- and undershoots continues until eventually
the car settles at 55 mph. This system is stable, but takes longer to reach its final
goal.
Thirdly the car can increase to, say, 60 mph before the system begins to slow
it. It then drops to 50 mph. The cruise control commands higher speed and the car
increases again to 60 mph and then once again drops to 50 mph. This cycle can continue
indefinitely. This type of system is unstable and is referred to as oscillating.
Finally,
the car can increase speed and over-shoot to 55 mph. The cruise control reacts, but so
slowly that the speed drops to 45 mph. During the next cycle the speed increases to 62 mph
before the control system starts slowing it; eventually it drops to 37 mph. The next cycle
takes the speed to 71 mph and back to 26 mph. As you can see, the swing in speed is
increasing. The car will never stabilize at 50 mph or any other value. This system is
unstable and will likely self-destruct.
The stability of the control system is determined by a number of characteristics of the system, but one of the largest factors is the gain, or amplification, in the feedback loop. Usually the difference between actual plant output and commanded plant output determines the size of the change in the control input. If the gain in the feedback loop is sufficiently large, the change in control signal will result in an increase or decrease in plant output which is too much for the system to follow.
Another thing which can make a system unstable is too long a delay in getting the feedback information back to the control input. We assumed that the speedometer provided the cruise control with instant information. Suppose there is a large lag in time in getting this information. When the speed of the car passes 55 mph there is a delay in sending the information to the cruise control so that the speed increases more before the system tries to slow. When the speed again drops below 55 mph the delay allows the car to slow too much before responding. This results in an oscillating or unstable system.
TELEROBOTICS
A telerobotic system contains the VR telepresence display system, a communications path between the user and the remote location, and the robot (actuator) at the remote location. The communication system carries commands from the user to the remote site and feedback information from the remote location to the user. The VR system allows the user to send the control signals and receive the feedback information.
In a simple system, feedback can consist entirely of visual information from a single TV camera mounted on or near the robot. In a more complex system the feedback could have stereo imaging from the point of view of the robot, 3-D audio, and force and tactile feedback information from the robot itself. The user could not only control the robot’s actions, but also the point of view of the imaging system, its focus and zoom.
The primary advantage of a telerobotic system is that it is unnecessary to support an operator in the location of the actual operations. If the location is hazardous or restrictive in physical size or environment the user can control operations from a distance and a safe location. Such systems are now used to explore the ocean floor, clean toxic waste, and allow a surgeon to operate deep within the body without a major opening.
While it might be possible to construct an autonomous robot with enough built-in intelligence to perform a particular task, if the task has any complexity, it is usually easier to let a human operator control the robot. Telerobotic systems help make this practical.
The degree of realism required in the telepresence system is determined by the task required. Because we obtain the majority of our information visually, a real-time TV picture is often sufficient, even if it is only two dimensional. However some tasks are extremely difficult using only vision as a guide. Anyone who has tried to start a nut onto a bolt with numb fingers knows that without the tactile information normally provided by our fingertips such a task is almost impossible. Lacking effective tactile feedback from a robot has made this relatively simple job one which cannot yet be accomplished remotely.
Another consideration in a telerobotic system is the time lag required, both the lag for processing by the telepresence VR system and also the lag due to the communication process in each direction. If the robot and operator are located relatively close to each other and electrical or optical means are used for communication, this lag can usually be ignored. But if the operator and robot are separated by a long distance, the travel time of the communication signals can be significant. For example, NASA wants to control from the earth the progress of a lunar explorer on the moon. The one-way travel time for communication signals traveling at the speed of light is about one and a half seconds. Thus the period between the time the operator sends a command and the time he sees the result is three seconds. If the lunar rover camera shows that it is approaching a cliff, it is necessary that the operator give the command to turn at least three seconds before the rover would reach the cliff. This was demonstrated at the UW Engineering Expo a few years ago with a display using radio-controlled model cars operating on a model of the lunar landscape. Users could operate the cars in a normal fashion or a three second delay could be switched in so that commands were actually sent to the car three seconds after the user made them. The difference in controlling the car was astounding.
An even more extreme example is the Mars Rover vehicle which was used to explore the surface of Mars in the summer of 1997.
Like the time lag of a communication system itself - which may be due to distance or to delays in signal processing - the lag in response of the telepresence system is also important and can effect the operation in the same way. If a telerobotic system is used to control a vehicle located only 100 meters away, the lag of communication is effectively non-existent. But suppose it required three seconds for the VR system to render the image transmitted from the vehicle. Although unlikely, images might still be rendered at a rate to 30 per second, giving realistic motion; it would just be that the image was three seconds old by the time the user saw it. As you can see, this would have the same effect as a three second lag in the communication system itself.
While such lag times as three seconds in rendering an image are unlikely in any real-time system, lags of even a few or tens of milliseconds can have an effect on the overall system performance. Many headtrackers, for example, have a lag of from one to one hundred milliseconds. The user of such a system will notice this lag when he moves his head and the scene does not change immediately.
Another method of telerobotic application is a cross between the independent robot and the normal telerobotic operator-robot system. This type of system can be applied to a task which will be repeated exactly over and over. For example, to machine a precision part a trained machinist is required. Robotic milling machines and lathes exist but they must be programmed. Once programmed, however, they can repeat the operations exactly and turn out 10 or 100 or 10,000 identical parts. If a telerobotic system is used to control the machines (robots), an experienced machinist can perform the operations the first time. Because the control signals for the machines are being transmitted over the communication link, they can also be recorded and played back over and over, causing the machines to exactly duplicate the first operation. Once the system has "learned" its task, it is not necessary for the operator to be involved.
Telerobotic systems require the same devices - sensors, computer, and actuators - as any other VR system. A difference lies in the fact that the sensors must also supply information on the performance of the actuators. If the system is to operate closed loop - necessary for precise operations - there must be a method for the user to determine if the actuators did what they were commanded. In a non-robotic system an actuator, such as a motion simulator or an audio reproduction system, operates directly upon the user. Feedback as to how well the actuator did its job is provided by the user’s own senses. With a telerobotic system, however, the user does not experience the final action of the actuator. If the user commands the lunar rover to turn left, he must have some feedback - visual or otherwise - to determine if this action actually took place. This does not just mean that he is trying to see if the command system worked correctly; outside influences could have prevented the desired action even if all systems were operating properly. For example, the command is given to turn the lunar rover left. The system functions correctly but at the site of the rover there is a sheer rock wall to the left of the vehicle. The rover can not make the left turn even though everything in the system is working correctly.
The above discussion demonstrates a point about feedback systems. It is generally best to provide feedback sensors which measure the final output of the system. For example, the speedometer in a car actually measures the speed of revolution of a shaft in the transmission. This means that even though the speedometer indicates a particular speed, the car itself may not be traveling at that speed. If a shaft is broken or a wheel slips on ice, a faster than actual speed will show on the speedometer. In this case the speed of the transmission shaft was measured instead of the actual speed of the car because the instrumentation is much easier to devise. There is often such a tradeoff: ease of measurement vs. distance from the actual measurement which is desired. The more removed the measurement, the easier it is for errors in system performance to be masked.