Closed kevinzakka closed 1 year ago
Hi Kevin!
This is about interpretability. The F/T sensors between welded bodies are easier to understand, since there is no relative motion. The F/T reading in the joint space does read the effective F/T in that subspace, but it cannot read the forces that create acceleration. See example below.
No idea, maybe @alaurens knows?
Can you write a minimal test that shows unexpected behaviour? Ideally building on the model below.
This is related to what I explained above. Some of the torque is causing acceleration and some isn't. The sensor can only report forces that do not contribute to acceleration.
Here is my minimal model. Note that as the pendulum swing freely under gravity the torque sensor is measuring nothing, since %100 of the torque goes to creating acceleration in the joint space. The force sensor is always measuring the full force since there are no linear motion DoFs.
<mujoco>
<option gravity="0 0 -10"/>
<worldbody>
<geom type="plane" size="1.5 1.5 .01"/>
<light pos="0 0 2.5"/>
<body pos="0 0 1.2">
<site name="sensor1"/>
<joint name="hinge" axis="0 1 0" limited="true" range="-180 180"/>
<geom type="capsule" size=".03" fromto="0 0 0 1 0 0" mass="0"/>
<geom size=".1" pos="1 0 0" mass="1"/>
</body>
</worldbody>
<sensor>
<force site="sensor1"/>
<torque site="sensor1"/>
</sensor>
</mujoco>
Note that I made gravity and all the lengths and masses nice round numbers so that it will be very obvious what the expected measurements should be.
Hi!
I've been playing around with F/T sensors in MuJoCo over the last few days and was wondering if I could get some advice / explanations for some things I am observing.
0 0 1
)?Just to make things easy, I made a very simple 2-link arm and wrote a minimal example to unit test force and torque if that helps clarify my questions! You can find them here: https://gist.github.com/kevinzakka/701a1b9dea30bc675a4cf40b2af01659
@shacklestone @alaurens Would really appreciate your help!!