search

gazebosim / gz-physics

Abstract physics interface designed to support simulation and rapid development of robot applications.
https://gazebosim.org
Apache License 2.0
63 stars 38 forks source link

bullet-featherstone: F/T sensor reports incorrect values at fixed joint if joint forces are applied #565

Closed shameekganguly closed 2 months ago

shameekganguly commented 8 months ago

Repro steps:

  1. Create a model with a few revolute joints and links.
  2. At the last link, add another (payload) link with a fixed joint.
  3. Place an F/T sensor at the fixed joint.
  4. Add a joint position controller system to control the position of one of the revolute joints using joint force commands (PID).
  5. Start the simulation and wait for the model to reach steady state.
  6. Check published F/T sensor value.

Expected value of force: weight of the attached payload link in the sensor frame. Actual value of force: different from above, and magnitude is larger than the weight of the payload link.

You can test this with this sdf file: fixed_joint_ft_sensor_bullet.sdf.txt. Run gz sim and wait for steady state. Then check /ft_sensor/wrench topic for F/T values. Expected force is [0 0 -9.8], but the published value is [0 -0.34 -13.85].

Note that the bug only occurs when using bullet-featherstone and not for dartsim. Also, if a joint velocity command is used instead of joint force command to control the joint position (uncomment the second joint position controller plugin in the sdf), then the bug does not occur.

azeey commented 8 months ago

Is this related to https://github.com/bulletphysics/bullet3/issues/4451? Perhaps we can use the python file there to test without gz-physics to see if it's an issue in Bullet or gz-physics.

shameekganguly commented 8 months ago

Is this related to bulletphysics/bullet3#4451? Perhaps we can use the python file there to test without gz-physics to see if it's an issue in Bullet or gz-physics.

Its possible that this is a related issue in bullet. I'll try to test with the python file this week before I am OOO.

iche033 commented 8 months ago

looks like the bug could be related to https://github.com/bulletphysics/bullet3/issues/2966. I tested bullet (master branch) with the patch posted in https://github.com/bulletphysics/bullet3/issues/2966#issuecomment-688136359 and I now get reasonable values ([0 0 -9.8]) in the fixed_joint_ft_sensor_bullet.sdf world

scpeters commented 8 months ago

I added this to the tracking issue in #545

iche033 commented 2 months ago

fixed upstream with https://github.com/bulletphysics/bullet3/pull/4583