Object unstable or gym inaccurate?

Hi there,

I’m trying to use BigBird dataset which some I’ve heard used successfully in Gazebo.

Some objects in the dataset seem unstable though? When I load an object from the dataset, they fall over.

My question is whether this behavior is a result of the object being unstable, e.g. its support surface is not flat enough or is there something on the simulator side? And of course, how to fix it?

I’m attaching a minimal reproducible example: code, urdf and .obj of the object.
unstable_mesh_min_example.py (2.1 KB)
object.urdf (283 Bytes)
mesh.obj (712.8 KB)

Here is video of what I observe: https://youtu.be/UQnz7-qIrj8

Ah, the joys of working with simulators like Gazebo. I’ve run into similar issues myself, and let me tell you, it can be a real mystery sometimes figuring out what’s causing the instability.

First off, double-check the properties of the object from the BigBird dataset. Ensure the mass and dimensions are accurate. If the object’s center of gravity is off or if its mass is incorrectly defined, that could make it unstable in the simulation.

Another thing you could look at is the physics engine settings in Gazebo. Try tweaking parameters like the time step or solver iterations to see if that has an impact.

Unfortunately, I can’t directly look at your attached files, but it would be crucial to investigate the URDF and .obj files for any inconsistencies. Sometimes the mesh file (.obj) might not perfectly align with what’s defined in the URDF, causing instability.

If you’ve covered all these bases and are still facing issues, it might be time to take a deeper dive into the Gazebo forums or even submit an issue to see if anyone else has faced similar problems with the BigBird dataset.

Good luck, and let me know if you find a solution!