Force sensors for measuring ground reaction forces

[sayantan]

Hello,

I have a custom robot model and wanted to add force sensors to the feet in order to measure the ground reaction forces. My objectives are:
(1) to detect when a foot makes contact with the ground, and
(2) to measure the location of the ZMP.
As I did not have any prior experience with VREP, I used the information in viewtopic.php?f=9&t=2207 to add the sensors. Visually things look more or less okay, but there are a few problems:

(1) The sensors connect the foot sole and a small respondable shape. This respondable shape (e.g. left_footsole_respondible_1 in the scene hierarchy in image 1) is of dimensions 1cmX1cmX0.3cm. This shape constantly wobbles when the simulation is switched on. Does this wobbling effect the force sensor readings? If yes, then how can I stop the wobbling?

(2) When I use ContactInfoDisplay, I can see that some sensors show readings even when the robot's feet are in the air (image 2)? My script for accessing the sensor values also show readings. One reason for this may be that the small respondable shapes are pushed inside the foot, but I cannot be sure, since I set the position of these shapes by visual inspection only. How can I fix the sensors and shapes so that they give readings only when ground contact is made (I handle the sensor noise separately)?

(3) Is what I did (image 1), the correct way of implementing foot pressure sensors/contact sensors or is there a better way?

Image 1: https://ibb.co/gdRQFQ
Image 2: https://ibb.co/hx9eaQ

Any help is greatly appreciated.

Best,
Sayantan

[coppelia]

Hello Sayantan,

from the pictures it seems that your cuboids that are linking the force sensors with the fee are constantly colliding with the feet.
You should either move your cuboids away from the feet, or set a different local respondable mask for the cuboids and the feet (e.g. 00000001 for the cuboids, and 11111110 for the feet).
The wobbling effect has probably two reasons:

• Your cuboids are colliding with the feet (what was said above)
• Your cuboids have a comparatively too small mass and inertia (when compared with the feet). For details, make sure to read the design considerations 6-8 here.

Try to also switch physics engines to see what effect could be linked to a specific engine, and what other effect is probably linked to a bad design.

Also, make sure to have a look how things were made for the robot located in Models/robots/mobile/NAO.ttm

Cheers

[sayantan]

Hello,

Thank you so much for the useful advice :) I increased the size of the cuboids and also set the local respondable masks properly, and the wobbling disappeared. I tried to make my force sensors' configurations exactly like the ones in the Nao scene (Models/robots/mobile/NAO.ttm). I can now see that the contact points look more or less accurate (image: https://ibb.co/dEacjk).

However, there's another problem which I haven't been able to get my head around. When I plot the forces in the z-direction of all the sensors, roughly half of them show positive values and the other half show negative values (image: https://ibb.co/n3usH5). When I plot the sensor readings for the Nao scene, all values are above zero, as they should be (image: https://ibb.co/joguVQ). I have verified that the force sensor orientations are correct.

Any suggestion regarding what I could be doing wrong? Thanks in advance.

Best,
Sayantan

[coppelia]

Are you sure all force sensors have the same orientation? I.e. the Z-arrow of the force sensor reference frame should point upwards.
At the same time, when the robot applies its weight on one foot, the Z-force direction should be opposite to the situation where the robot lifts the foot (in the first case the force sensor pushes the foot up, in the second case the force sensor pulls the foot down).

Cheers

[sayantan]

Thank you once again for replying.

Yes, I made sure that all sensors have the same orientation (Z-axis of force sensor ref. frame points upwards). As you said, when the robot puts one foot on the ground (or when the robot is standing still) the force sensors push the foot up. So the forces should be positive in the positive Z-direction. In my case, this is true for all sensors apart from a single sensor on each foot (left_sensor_3 and right_sensor_2), which show negative values in the Z-direction when the robot is standing still. The contact points look okay.

The funny thing is, after I give the robot a gentle push, the force sensor values go back to positive and stay there. The plots in this image https://ibb.co/i4WDS5 show that after the spikes caused due to the push, the forces stay positive. I have tried dropping the robot from a small height but the problem remains. Here is the scene: https://drive.google.com/open?id=0B0cTX ... UVYNUMyNjg. It would be very nice if you can take a look. Thanks for your help.

Best,
Sayantan

[coppelia]

You have sevral problems in your robot:

• Your robot hands/fingers are colliding, which causes instabilities with some of the other engines (e.g. ODE and Vortex)
• Your center of mass lies just above the feet (is way too low)

Finally, it seems that the reading of the sensors can be explained by the fact that the robot is hyperstatic (is touching the floor with more than 3 points, which introduce small vibrations, but probably also the effect you are wondering about. The total mass of the robot is 28kg, but each one of the pads that is touching the floor via a force sensor weights itself already 3kg...

Cheers

[sayantan]

Thank you very much for the quick response and for pointing out the issue with the low center of mass. I fixed the weights of the small pads below the feet and the results look marginally better now. The CoM now lies around the pelvis. Regarding the instability of the model in ODE/Vortex, I also noticed this, but could not figure out the reason, until now.

Best,
Sayantan