Keyboard Shortcuts
ctrl + shift + ? :
Show all keyboard shortcuts
ctrl + g :
Navigate to a group
ctrl + shift + f :
Find
ctrl + / :
Quick actions
esc to dismiss
Likes
Search
stewart platform
|
I'm curious about the Stewart platform, and what kind of actuators might be
salvaged to create a small one - like meter or less in dia? I suppose just some salvage steppers with threaded rod, but if others might have better ideas, I'd love to hear. One of these "sandtables" would be cool to build. . <> com/art_machines.htm . <> com/art_machines/sisyphus/bssis1_640.jpg Chuck Merja www.3rivers.net/~chuckm/index.htm Lemelson-MIT Inventeams - US FIRST Robotics - lat/long 47.52383 -111.67912 211 Adams Rd Sun River, MT 59483 vx 406.264.5955 cell 406.799.5955 fax 406.264.5830 |
|
did you check out
toggle quoted message
Show quoted text
one of the few DIY hexapods I have seen on the net. click screenshots and download a video at the bottom of the page. there's a cnczone thread about the machine at: If you just find the kinematics interesting and want to play around, you could suspend the moving platform from strings. I remembers seeing pictures of a couple of these kind of machines but I can't find the links now... If someone can come up with a smart way of calibrating home made parallell robots, then I'm very interested too - until then I'm sticking with a simple cartesian mill... I'm curious about the Stewart platform, and what kind of actuators might be |
|
Carl Mikkelsen
I've had what seems like good success calibrating a home-brew parallel robot.
toggle quoted message
Show quoted text
The general method is: 1) add a bunch of error variables into your kinematics equations to represent the major error contributors. This are typically the x, y, &z positions of the base and platform joints, the minimum leg length (at "0" position reference), and possible the actuator scale error. 2) move the platform to a series of known orientations, including all 6 axes being varied. I'll do something like a gird measuring every 6 to 12 inches, with Pitch, Roll, and Yaw at -30, 0, +30 for many of the points. 3) use simulated annealing to adjust (hill climb) the error variables for the closest fit. 4) look for measurements whose final error falls 3 standard deviations away from the mean, and either remeasure those points or discard them from the data set. 5) repeat steps 3 & 4 as necessary until the process converges. Steps 4 & 5 are necessary only if you (as I do) sometimes mess up the measurements. -- Carl At 01:20 PM 8/29/2006, Anders Wallin wrote:
did you check out |
|
Graham Stabler
--- In CAD_CAM_EDM_DRO@..., Carl Mikkelsen
<c.mikkelsen@...> wrote: 2) move the platform to a series of known orientations, includingall >6 axes being varied. I'll do something like a gird measuring every 6 >to 12 inches, with Pitch, Roll, and Yaw at -30, 0, +30 for many of the >points. How do you go about measuring the error? Say you move 100mm in X, how do you measure the error created in one of the rotational axes? Graham |
|
Carl Mikkelsen
Graham,
toggle quoted message
Show quoted text
I don't measure the error. I manually "fly" the platform with a joystick to the position where it should be, and then record the length of the actuators that puts it at that place. To set the Pitch and Roll, I've used an electronic level with 0.01 (I think, maybe 0.1) degree resolution. To set the Yaw (my W axis), I attach a straight indicator to the platform, and align it visually with the base coordinate system. I then use simulated annealing to estimate the error variable values that are needed to make the actuator lengths work out to be correct. The simulated annealing is used to minimize the total error contribution, ie., how much the recorder actuator length differ from the calculated lengths. When all is right, the measured and calculated lengths match. Other approaches would work, but this has worked for me. I had some problems attempting to use simulated annealing on ALL error variables at once, but by controlling all six axes, the problem decomposes into 6 separate estimation problems -- one for each leg. My original plan was to estimate all errors at the same time, and not control all six carteasian axes. I intended to have a precision flat surface, and use a surface probe to record the leg lengths at which the probe touched. The partial-control of the carteasian coordinates coupled the actuator errors together (necessitating the all-variables estimation), so I changed to the strategy I outlined. -- Carl At 08:56 PM 8/29/2006, you wrote:
--- In |
to navigate to use esc to dismiss