State/force/wrench data inconsistent

[hksaperstein]

Hey I'm getting inconsistent/inaccurate state/force/wrench readings. Whats expected is the collision force between the <sensor_name>Leg and <sensor_name>Tab. I have attached output data and images to illustrate the issue. As you can see when the sensorTab (green element) is not in contact with the collision of the sensorLeg (yellow element) I would expect wrench in y to be zero as nothing is acting upon the sensorTab as it is essentially free hanging. This is the same with z-axis, as the tab has mass of 1, with gravity, Id expect a force of -9.8N, however it ranges from -10 to +30/+40.

Also, I am getting collision clipping issues where elements will load in behind a collision element if they are in the same collision group. There are images included that show this issue.

The blender file is also included for reference.

Thanks,
Harry Saperstein
sensor images.zip

[adnanmunawar]

I briefly looked at the model and the way the "force measuring" sensors have been placed seems very complex and unnecessary. Maybe you can reproduce the issue with a simpler example and send over that .blend file.

A few points that I noticed:

1. The version of ambf_addon is outdated, better to update it.
2. Set all the bodies and joints that you do not need to be reported in the communication pipeline as passive using the newer version of ambf_addon. This would be at-least all fixed joints and maybe the limbs and head links since i am guessing that you are using the torso's handle to read all the joints of attached limbs and not reading the state data of the limbs themselves.
3. Try running the simulation with the following flags -t 1 -p 60. Fixed time stepping and Physics update rate of 60.
4. The reported wrench vector in expressed in the body frame rather than the world frame. I do plan on changing this such that the sensors readings are always in the world frame.