DAVID@PENNDRLS.BITNET (03/10/89)
Some more fuel for the 'practicality of robot hands' debate: The February 18th Science News has an interesting article on a joint project between Columbia University and the Steinway Piano Company. The project's principle investigator is Moshe Shoham, a mechanical engineer at Columbia. His objective is to build a robot capable of taking over the job of finishing a Steinway: "By assembling existing robotic components in new ways and using novel computer control tactics, Shoham hopes to build a tireless, boredom-immune robot that finishes piano cases as finely as an experienced human worker but in far less time." Sounds to me like a true leading edge *application* (as opposed to research project). A somewhat different problem from a tele-operated rig, but the 'hand' technology involved is parallel: "...the robot must adapt to the subtle variations of each hand fashioned piano. It must be dexterous enough to reach every part of each piano case, and sensitive enough to 'know' how much pressure it is applying to the flat, curved, and oddly shaped surfaces." Dexterity and feedback are also primary requirements for a tele-operation rig, I would think. Building this robot is not going to be a trivial problem. It is certainly not 'off the shelf': "[Shoham notes that] good, reliable force control is a rare robotic skill. [...] Most current robots cannot respond fast enough to the force sensor's flow of information." It sounds like he is using a more-or-less standard robot of some sort, since it is described as coming with its own controller. To this he adds another control computer to deal with the force sensor data and to feed back information to the control computer. Aside from the feedback/control problem, other problems are things like how to deal with changing the sandpaper on a sanding block. For these, Shoham expects he may have to design 'robot friendly' finishing tools. Presumably changing sandpaper is a hard problem because it is a task with many complex variables; a problem in control that would not arise if a human were directing the hand via tele-operation, assuming the hand was dexterous enough. A problem intrinsic to a force-feedback hand is that "the rubbing machine's vibrations could 'confuse' the force sensor, causing the robot to apply improper forces to the piano case." What this says to me is that force-feedback is by no means a 'solved problem'; if you want to deal with a wide variety of applications of the hand there is a lot of work to be done on making the force sensing reliable. Overall I am encouraged by the article: a complex application of a force-feedback robot is reasonably close to commercial application, which puts us that much closer to 'off the shelf' force-feedback tele-operation. But it definitely looks like Henry is right as far as the short term goes. We need a lot more research and development before tele-operated experiments in orbit can become routine. Funny, that sounds like a statement I've read recently about manned operations in space . . . We should be doing both. -- R. David Murray (DAVID@PENNDRLS.BITNET, DAVID@PENNDRLS.UPENN.EDU) P.S.: I agree that unmanned space exploration is ridiculously under- funded compared to manned space (insofar as you can separate the two cleanly). It is criminal that experiments get dropped and maybe (if we're lucky) replaced with something less capable as a 'cost savings measure'. When you've already agreed to send the probe, you bloomin' well ought to send along the best instruments you can get your hands on!
rachiele@NADC.ARPA (J. Rachiele) (03/11/89)
> Let's take a specific case. I want the robot to open a screwed on access >cover, remove a circuit card from a card case, put it into a test fixture, >run test points on the traces (let's say .03" design rules), and finally >remove and replace a soldered chip and reassemble the unit. Can you point >to any of this being done even in a controlled ground environment? For maintenence of military computers in the field, it is not assumed that the human can do even this much. The fault localazation software on our system is required to print out a list of cards, in order, to replace. And the doors on the cabinet has handles. This would make it a little easier. A more difficult task might be screwing in a connector without bending any of the pins (although you may have trouble finding a human who can do this reliably.:-) > ++PLS Jim rachiele@nadc.arpa