A way to automatically detect tendon skipping?

[studywolf]

Intro

Hi!

I am a researcher at EPFL working on mouse arm modeling.

My setup

Mujoco 2.3.1
Ubuntu 22.04

My question

I'm trying to devise a way to detect when tendons skip from one side of their deformation object to the other.
I tried recording the tendon lengths and then looking at the derivative, thinking that sudden changes in tendon length would correlate with skipping from one side to the other, but unfortunately this isn't the case.
Is there another tendon variable that I could record during movements and analyze after to alert me when there is a tendon skip? Identifying tendon skips visually is far from ideal!
Thanks!

Minimal model and/or code that explain my question

No response

Confirmations

• I searched the latest documentation thoroughly before posting.
• I searched previous Issues and Discussions, I am certain this has not been raised before.

[yuvaltassa]

You want to look at mjData.wrap_xpos. wrapadr and wrapnum tell you where your tendon starts and ends.

[studywolf]

Thank you very much! I've been digging into this, and I'm a bit confused still. I see that the description of mjData.wrap_xpos is nwrap x 6, and unfortunately for me the description // Cartesian 3D points in all paths doesn't give me much of an idea of what this is exactly.

I was guessing that the 6 dimensions of each row are the start and end (x, y, z) of that segment of the tendon, but the last 3 values of a given row only match the first 3 values of the next row occasionally. I've been plotting it out trying to figure out what's going on, and I'm suspecting it's not the tendon segments start and end being stored ... but I was hoping that you might be able to shed some light on how to interpret the data in wrap_xpos!

[yuvaltassa]

IIRC that's almost right, but not quite. The segment is between two subsequent rows i.e. from last 3 elements of row i to the first 3 elements of row i+1. This is required because in the case of geom wrapping there is a mini-segment at the geom. Search engine_vis_visualize.c for wrap_xpos to see the actual rendering code. Hope this helps.

[studywolf]

Thank you for the follow up! I played around with it a bit and the following code lets me access the tendon segment 3D positions of interest (in case anyone else needs this at some point):

ten_wrapxpos = {key: [] for key in list_of_tendons}
wrap_xpos_stack = np.hstack(data.wrap_xpos)

for key in ten_wrapxpos.keys():
    tendon = data.tendon(key)
    start_idx = tendon.wrapadr[0]
    end_idx = start_idx + tendon.wrapnum[0]
    wrapxpos = wrap_xpos_stack[start_idx * 3 : end_idx * 3]
    ten_wrapxpos[key].append(np.copy(wrapxpos))

the data.wrap_xpos is set as a list of 6 element lists, which is what was throwing me off. by reworking it to a flat list and then indexing in the same way as the engine_vis_visualize.c code you reference above i got it going.

and then to plot there's a condition required to handle tendons with pulleys (where simply if the start or end connection is listed as [0, 0, 0] then don't plot)

for (name, xx) in ten_wrapxpos.items():
    for ii in range(int(len(xx[t]) / 3) - 1):
        # they use all zero entries when there's a space due to
        # pulley connections, don't plot if start or end is all zeros
        if not (
                np.sum(xx[t][ii * 3 : ii * 3 + 3]) == 0 or
                np.sum(xx[t][ii * 3 + 3: ii * 3 + 5]) == 0
                ):
            ax.plot(
                [xx[t][ii * 3 + 0], xx[t][ii * 3 + 3]],
                [xx[t][ii * 3 + 1], xx[t][ii * 3 + 4]],
                [xx[t][ii * 3 + 2], xx[t][ii * 3 + 5]],
                'k-',
                mew=3
            )

thank you again!