gm26@prism.gatech.EDU (MCMURRAY,GARY V) (12/07/90)
I am interested in finding out information on the development of 6 DOF joysticks by various people. In particular , I am interested in the manner that the motions of the joystick are converted into a motion for the robot. I know that JPL has been doing work in this area for many years but I have not been able to locate any papers that explicitily define this mapping process. There is also a company called Kraft Teleoperation (I believe that is the name) that has a commercially available joystick. Also,
are the devices mentioned above generic in nature such that they can be used to control any robot, or are they restricted to robots of similiar kinematic structure?
Thanks for any ans all of your input!!!
Gary McMurray
Home of the "Number 1" Football Team???
--
MCMURRAY,GARY V
Georgia Institute of Technology, Atlanta Georgia, 30332
uucp: ...!{decvax,hplabs,ncar,purdue,rutgers}!gatech!prism!gm26
Internet: gm26@prism.gatech.edusmith@sctc.com (Rick Smith) (12/07/90)
gm26@prism.gatech.EDU (MCMURRAY,GARY V) writes: >I am interested in finding out information on the development of 6 DOF >joysticks by various people. In particular , I am interested in the >manner that the motions of the joystick are converted into a motion for >the robot. I spent a year or so working on a project invloving this, and you connect the components together like this: off-the-shelf 6 DOF joystick ==> LOTS OF WORK ==> off-the-shelf robot The LOTS OF WORK section is the implementation of your teleoperation control scheme. We looked at several approaches, and it all depends on your joystick and your robot's geometry. I don't think there's much out there in the way of off-the-shelf teleoperation software that supports a range of joysticks and robots. Usually the best you can do is buy a robot with a handheld programming control that uses something approximating a joystick. I assume that by "joystick" you mean anything that will map operator movements into robot motions, and not just handles like they use on video games... In the project I worked on, we planned to do teleoperation controlled by DataGloves -- that's those bizarre gauntlets that measure finger flexes and hand position/orientation. We were going to use hand displacement (scaled by a gain signal from a foot pedal) to specify end effector displacement. The project was axed before the robot was built, though we had lots of fun with the DataGloves. On the other hand (sorry!) we had another project evaluating 6 DOF hand controllers... I got the impression that the classic technique there was to use displacement to specify a velocity vector for the robot's end effector motion. Let go of the controller and motion stopped; exert some effort and motion followed the direction/orientation you pushed. Ideally, you want the kinematics of the joystick (well, hand controller) to match that of the robot. Thus, the DataGlove is really best with a cartesian robot, as are the 6 DOF generalizations of 2 DOF joysticks. With articulated robots like Pumas, however, you have to worry about singularities in your work envelope (e.g. places you can't quite reach). If you look around, though, there IS some company that builds an articulated hand controller designed to match the kinematic configuration of things like Pumas. Sorry, but I don't remember the company name. I do remember that they cost lots, though you save something in software development by avoiding the singularity issue. Rick. smith@sctc.com Arden Hills, Minnesota
gerry@frc2.frc.ri.cmu.edu (Gerry Roston) (12/07/90)
In article <1990Dec6.232210.2638@sctc.com> smith@sctc.com (Rick Smith) writes: >I am interested in finding out information on the development of 6 DOF >joysticks by various people. In particular , I am interested in the >manner that the motions of the joystick are converted into a motion for >the robot. The work that was done at JPL was headed up by Dr. Tony Bejczy, the last number I have for him is (818) 354-4568. Or, try writing to him at: Jet Propulsion Laboratory, ms 198-330 4800 Oak Grove Drive Pasadena, California, 91109 Another person you might try to track down is Bill Townsend, who recently completed a PhD at MIT. He has done extensive work in this area and has produced some interested "joystick" designs. Ideally, you want the kinematics of the joystick (well, hand controller) to match that of the robot. Actually, this statement is very far from the truth. You want your master arm, i.e. the joy stick, to be constructed in such a fashion that it is easily operable by the human; and you want the slave arm to be designed to achieve the required task in the best possible fashion. Connecting the two is a computer which performs the kinemtic translation required. Furthermore, tests done at JPL and elsewhere have shown that to perform meaningful tasks, force reflection is required and that time delay from your sensors (cameras, force sensors, etc) will seriously degrade performance. gerry -- gerry roston, field robotics center robotics institute, carnegie mellon university pittsburgh, pennsylvania, 15213 (412) 268-6557 gerry@cs.cmu.edu
minsky@media-lab.MEDIA.MIT.EDU (Marvin Minsky) (12/08/90)
In article <GERRY.90Dec7094126@onion.frc.ri.cmu.edu> gerry@frc2.frc.ri.cmu.edu (Gerry Roston) writes: >In article <1990Dec6.232210.2638@sctc.com> smith@sctc.com (Rick Smith) writes: > Ideally, you want the kinematics of the joystick (well, hand controller) to > match that of the robot. > ... tests done at JPL and elsewhere >have shown that to perform meaningful tasks, force reflection is >required and that time delay from your sensors (cameras, force >sensors, etc) will seriously degrade performance. Terms like "seriously degraded" seriously degrade our appreciation of some problems. I mention this because I'm convninced that earth-based remote control of, for example, a space station, would yiled a huge advantage in performance/cost payoff. So the question is, what did "tests done at JPL and elsewhere" really demonstrate? Consider that the internal sensor-brain-muscle roudtrip time of a human is of the order of 1/5 second -- so that when your brain tries to do anythig in the outer world, you have a delay time of this order. Now, suppose that the sensory-motor loop time of a remote control system were, say, 1 second. Then you could expect the human performance time to increase six-fold, so that it would be "degraded" by that much. My question: has JPL or anyone else shown that delays cause substantially more degradation than this?
smith@sndpit.dec.com (Willie Smith) (12/08/90)
In article [...], minsky@media-lab.MEDIA.MIT.EDU (Marvin Minsky) writes... >In article [...] gerry@frc2.frc.ri.cmu.edu (Gerry Roston) writes: >>In article [...] smith@sctc.com (Rick Smith) writes: >> Ideally, you want the kinematics of the joystick (well, hand controller) to >> match that of the robot. > >> ... tests done at JPL and elsewhere >>have shown that to perform meaningful tasks, force reflection is >>required and that time delay from your sensors (cameras, force >>sensors, etc) will seriously degrade performance. > >Terms like "seriously degraded" seriously degrade our appreciation of >some problems. I mention this because I'm convninced that earth-based >remote control of, for example, a space station, would yiled a huge >advantage in performance/cost payoff. So the question is, what did >"tests done at JPL and elsewhere" really demonstrate? > >Consider that the internal sensor-brain-muscle roudtrip time of a >human is of the order of 1/5 second [...] Another consideration is that humans can learn to predict and anticipate control inputs to systems with long delays. After about 1/2 hour driving my (simulated) lunar teleoperated vehicle I've found I can do significantly better than when I started, and the 'training time' gets shorter with repeated 'missions'. While I don't doubt that force-feedback and other tightly-coupled systems can get unstable when the delay approximates the human reaction time, more loosely-coupled systems (joysticks and video feedback with Heads-Up-Display to show the operator where he's pointing his controls) allow the operator to compensate for the delays. Yes, precision tasks take longer, but nothing is quite as bad as the expected "move, wait to see what happened, move again..." case, even when running with a full 3-second lunar teleoperations delay. In case anyone is interested, the vehicle in question is a modified RC truck with TV camera and transmitter, controlled by a Z80-based S-100 machine, with the received TV signal routed through an Amiga with genlock for HUD. You can indeed do lunar teleoperations research in your basement! The phase II vehicle is based on lawn-tractor wheels and cordless drill motors, with a couple of onboard computers. Lots more documentation available on request. Willie Smith smith@sndpit.enet.dec.com smith%sndpit.enet.dec.com@decwrl.dec.com {Usenet!Backbone}!decwrl!sndpit.enet.dec.com!smith
nagle@well.sf.ca.us (John Nagle) (12/09/90)
minsky@media-lab.MEDIA.MIT.EDU (Marvin Minsky) writes: >Consider that the internal sensor-brain-muscle roudtrip time of a >human is of the order of 1/5 second -- so that when your brain tries >to do anythig in the outer world, you have a delay time of this order. Eye-hand control loops are of that order, but many purely tactile control loops in the body are much faster. The grasping reflex, which maintains finger contact forces at a level sufficient to prevent slip, operates in about 20ms. Flight simulator designers have discovered that update rates as high as 500Hz are required to make control forces "feel right". ("Flight Simulation", Rolfe and Staples, Cambridge University Press, 1986, section 4.10). A 1 sec delay in a tactile control loop thus represents roughly a 2 order of magnitude performance degradation. John Nagle
lance@motcsd.csd.mot.com (lance.norskog) (12/11/90)
smith@sndpit.dec.com (Willie Smith) writes: >Another consideration is that humans can learn to predict and anticipate >control inputs to systems with long delays. After about 1/2 hour driving >my (simulated) lunar teleoperated vehicle I've found I can do significantly >better than when I started, and the 'training time' gets shorter with >repeated 'missions'. Indeed. If the delay is constant, you can learn to live it. It's variable delays that kill you. You can check this out using a PC v.s. a time-shared mini. You can train to very long rhythms. Lance