PDA

View Full Version : [DEMO] walking on stairs!



guoshihui
12-15-2015, 02:39 AM
We did two motions, which are used to climb on stairs. Hope you guys like them, =)

https://youtu.be/E74hslqVCCQ

https://youtu.be/py80COyHACM

The motions are attached here and you can use the WinRME project from r3n33 to transfer the motion to Jimmy.
6360

However we would like to improve it further and have some suggestions from you guys. One thing is that we want to compute the Zero-Moment-Point for Jimmy so that he can maintain stable. To do this, we need the mass for each part (servos, plates etc), anyone can advise what would be the best solution to acquire this data? Or other options to ZMP?

Thanks for your advice in advance.

Shihui

r3n33
12-16-2015, 03:37 PM
Oh wow! Those are impressive videos; especially climbing with only legs. So much so I've been putting my HR-OS1 together again so I can try the motions myself.

Thank you for sharing :veryhappy:

wadecore
12-16-2015, 05:10 PM
Looks great! I see that these videos are set to unlisted on youtube. Would you be okay with us doing a write up on this and showing the videos off in the Trossen Robotics blog?

guoshihui
12-16-2015, 07:15 PM
Looks great! I see that these videos are set to unlisted on youtube. Would you be okay with us doing a write up on this and showing the videos off in the Trossen Robotics blog?
Thanks wadecore,

No problem. I have changed the videos to public, so you won't have any problems in sharing with others.

Again, anyone could share their advice on getting the Zero Moment Point of Jimmy, or any other way to make it more stable? Thx again.

Shihui

jwatte
12-16-2015, 10:50 PM
There are about three parts of stability:

1) There is flex in the brackets and servos. You can make stiffer brackets, and pre-load joints with torsion springs (which will also be good for battery life) to make that better, but this is a constant problem even for big, expensive robots.

2) You could add a balance sensor (acceleration sensor) and make it adjust the orientation of the pelvis based on how it's leaning (compared to how it's supposed to be leaning.)

3) You could make the movements run faster and more fluidly. As it is, it seems to take certain poses on a specific schedule. If you run them much closer together, the motion will be more fluid. However, you may also need to adjust the motions to compensate for overshoot and flex that is the same each time you move.

Those are some ideas. This is not a "totally solved engineering problem" though, so expect to need to try a bunch of different things, and iterate a lot to get it good :-)

guoshihui
12-17-2015, 03:16 AM
There are about three parts of stability:

1) There is flex in the brackets and servos. You can make stiffer brackets, and pre-load joints with torsion springs (which will also be good for battery life) to make that better, but this is a constant problem even for big, expensive robots.

2) You could add a balance sensor (acceleration sensor) and make it adjust the orientation of the pelvis based on how it's leaning (compared to how it's supposed to be leaning.)

3) You could make the movements run faster and more fluidly. As it is, it seems to take certain poses on a specific schedule. If you run them much closer together, the motion will be more fluid. However, you may also need to adjust the motions to compensate for overshoot and flex that is the same each time you move.

Those are some ideas. This is not a "totally solved engineering problem" though, so expect to need to try a bunch of different things, and iterate a lot to get it good :-)

Thanks jwatte, I think you are right. Sometimes it takes a lot of hands-on experience to make something look great.

Meanwhile we also go for the hard way, to dissemble the robot and get the weight for each part (or most parts in our case). The data is provided here only for reference. Since we are using a weight scale with resolution of 1gram, so the accuracy of the data is a bit questionable. But this weight scale is what we have at this moment, so =(.

If Trossen Robotics can reveal their data, that would be great. Thanks.

Shihui

6365

jwatte
12-17-2015, 01:18 PM
The good news is that, once you have that data, you can calculate the center of inertia for any configuration!

Another (quicker) way is to balance the robot on a pin from each of the three major directions, which will give you the center of inertia, but only for that particular pose/configuration of the bot. You could then pose it in various representative poses, and build a model that gives you an approximate ZMP, but that's not as good as having the weights/armature procedurally described!

wadecore
12-17-2015, 02:59 PM
I hope that this blog post shows Qian that her work is appreciated and well worth attention in the robotics community!
http://blog.trossenrobotics.com/2015/12/17/hr-os1-tackles-the-stairs/ (http://blog.trossenrobotics.com/2015/12/17/hr-os1-tackles-the-stairs/)

As for the weights of the parts, I'm sad to say that our scales are calibrated roughly the same. Our measurements are as accurate as yours.

LloydF
12-17-2015, 03:33 PM
Thank you so much I put them in my winrme folder and all is supper. Merry Christmas .:cool:

tician
12-17-2015, 03:57 PM
If your CAD/CAM program also does FEM/FEA, then you may be able to simply import the HROS stp file to get the volume, various moments of inertia, and center of mass of each piece. Can then estimate mass from the volume and material density. The moments of inertia would be useful for really accurate dynamics analysis, but each piece's mass and center of mass should be enough for most uses.

guoshihui
12-17-2015, 08:37 PM
If your CAD/CAM program also does FEM/FEA, then you may be able to simply import the HROS stp file to get the volume, various moments of inertia, and center of mass of each piece. Can then estimate mass from the volume and material density. The moments of inertia would be useful for really accurate dynamics analysis, but each piece's mass and center of mass should be enough for most uses.
Hi Tician,

Yeah, I think it's a good idea, and this may be possibly how the Darwin guys work this out. I found the dynamics properties for Darwin on the sourceforge:
http://sourceforge.net/projects/darwinop/files/Hardware/Mechanics/Physical%20Information/DARwIn-OP_Dynamics.zip/download

r3n33
12-19-2015, 02:09 PM
This is so cool! Thanks again :) I had fun getting to try this out on my own.


http://www.youtube.com/watch?v=Tcgd-3KhDnk