Robot stucks into the plane.urdf

[KH-HSU]

Happy New Year @erwincoumans and everyone

Problem:
When the robot walked from the board, sometimes the robot will stuck into the plane which are shown in pictures below. This problem also happened when I used Heightfield.

If the robot just walked on the plane without the red board in the pictures, the robot will sink a little bit but will not stuck into the plane. ( The red line is for RL experiment without collision, so it will not effect collision.)

What I have done

1. set timestep to 0.001
2. set numSolverIterations to 1000
3. Use the smallest resolution to generate mesh of the robot in Solidworks (also tried in STL and obj format)

but the problems still exists
Does anyone have met this problem before? Please give me some suggestions, thank you!

[KH-HSU]

I solved it by

1. putting the board base under the plane ( I put it just right on the plane before ),
2. set the numSolverIterations to 1000.
   I'm still curious if there are other solutions because I want to use Heightfield to do experiment.

[BluePrintRandom]

a plane has zero thickness - try a very short cube instead using the cube bound.

…

On Mon, Jan 4, 2021 at 6:09 PM KH-Hsu ***@***.***> wrote: I solved it by 1. putting the board base under the plane ( I put it just right on the plane before ), 2. set the numSolverIterations to 1000. I'm still curious if there are other solutions because I want to use Heightfield to do experiment. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#3214 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABW3SWKQ5RYUK7BASNUJMC3SYJYHJANCNFSM4VSNDZGA> .

[KH-HSU]

@BluePrintRandom
I'm a bit confused because the plane.urdf given by Pybullet has thickness which can be found in
plane.urdf, and I made the red cube by loading cube.obj into urdf

[BluePrintRandom]

you can use CCD to ensure that a object does not pop through another - but it's typically easier to simulate a more realistic environment (ie a thicker floor)

…

On Mon, Jan 4, 2021 at 9:00 PM KH-Hsu ***@***.***> wrote: @BluePrintRandom <https://github.com/BluePrintRandom> I'm a bit confused because the plane.urdf given by Pybullet has thickness which can be found in plane.urdf <https://github.com/bulletphysics/bullet3/blob/master/examples/pybullet/gym/pybullet_data/plane.urdf#L22> I made the red cube by loading cube.obj <https://github.com/bulletphysics/bullet3/blob/master/examples/pybullet/gym/pybullet_data/cube.obj> into urdf — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#3214 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABW3SWMLIXPQZV6WHGWEL5DSYKMIVANCNFSM4VSNDZGA> .

[KH-HSU]

@BluePrintRandom thank you, I will try CCD.
But the problem still exists when the floor thickness is 4 cm.

I put the green cube under the plane and fix it to solve the problem which is shown in the picture below
the red line without collision property is just right on the plane.

I wonder if I just put the floor right on the plane, there might be some small space that will make the robot stuck in the mesh.