Getting contact data using the contact_physx_view

[PerFrivik]

Hello everyone,

Thank you in advance for your help! I have a question regarding obtaining contact data using the RigidContactView class API within the file _isaac_sim/extsPhysics/omni.physics.tensors/omni/physics/tensors/impl/api.py. Specifically, I am uncertain about the def get_contact_data(self, dt) function, which I believe is based on the following documentation: get_contact_force_data.

Compared to the get_net_contact_forces(self, dt) function, I find get_contact_data(self, dt) harder to use because we need to predefine the size of our buffer. This would be manageable if I knew the exact number of contacts per robot. However, through trial and error, I have observed that each foot of the Anymal robot can have more than one contact. For instance, with an Anymal robot with 4 legs, I would expect to get 4 contacts between the feet and the ground using the following configuration:

filter_prim_paths_expr = [
    "{ENV_REGEX_NS}/Robot/LF_FOOT",
    "{ENV_REGEX_NS}/Robot/RF_FOOT",
    "{ENV_REGEX_NS}/Robot/LH_FOOT",
    "{ENV_REGEX_NS}/Robot/RH_FOOT",
    "/World/ground/GroundPlane/CollisionPlane"
]

However, I sometimes get up to 6 contacts, and I am unsure why this happens. This variability complicates calculations dependent on the friction location, as I am trying to implement the Pacejka Magic Tire Model. This model requires applying forces to the contact points to replace the "usual" friction calculated by the simulation.

Currently, I am testing with the Anymal to validate the pipeline, but the goal is to use it for the AoW.

A second issue I have is understanding what points in the point_buffer belong to what environment, I understand that the API provides the pair contacts count and start parid contacts indices, but since the buffer seems to just fill up, I am not sure how this is sorted as I need to differenciate between the different env_ids. Also the the contact count buffer and start indices buffer will explode in size, as with just 2 robots and 4 porential contacts (LF_FOOT with ground, RF_FOOT with ground, etc.) the tensor already has a size of ~30x1. So simulating 4096 in parallel would be problematic I guess.

I understand that this is a big question, but maybe someone has had any simmilar issues, or knows in what direction to point me. I will also keep posting my findings in this discussion if I find out more!

Thank you for your assistance and I would really appriciate any hints!

I guess this is also somewhat related to the following issue Issue 364

Best,

Per

[PerFrivik]

I have gained a better understanding after reviewing the test_cube_stack_contact_filtering(self) unit test in test_contact_sensor.py.

I now realize that the large tensor arises because it's not just the primitives specified in the filter_prim_paths_expr that get evaluated. Instead, the prim_path itself is cross evaluated using the prims in filter_prim_paths_expr. This means I was evaluating every possible prim in the robot with its own feet again + the ground plane.

Next I will try to figure out how to figure out what value in the buffer belongs to what env_id.