Hi - I was also making the transition from Isaac Gym to Omniverse Isaac Gym Envs and the documentation is not that clear - but there is access to tensors for rigid body contact force for multiple actors in rl learning.
here is the gist…using the OmbiIsaacGymEnvs task framework e.g. anymal.py but no need anymore for the anymalview import which just serves to confuse and is nor helpful if you are using your own model
from omni.isaac.core.articulations import ArticulationView #if you need an articulation view (likely)
from omni.isaac.core.prims import RigidPrimView
def set_up_scene(self, scene) → None:
…self._robots = ArticulationView(prim_paths_expr=“/World/envs/./biped", name=“biped_view”)
…self._lfoot = RigidPrimView(prim_paths_expr="/World/envs/./biped/LeftFoot”, name=“lfoot_view”, reset_xform_properties=False, track_contact_forces=True, prepare_contact_sensors=False)
…scene.add(self._robots)
…scene.add(self._lfoot)
def update_buffers(self):
…self.contacts_left = (self._lfoot.get_net_contact_forces(clone=False)[:,2] > 0.00)
…print(self.contacts_left.shape)
… print(self.contacts_left)
so this example takes the left foot prim from my biped.usd file for my 512 robot instances during training and detects if there is any contact force in the z plain (>0.00) that gives you a bool tensor to see if it is on the floor or not…
in the anymalview example in omniisaacgym look like they do combine the articulation view and the net force contact sensors but this version works for me.
Devs please look at documenting this better!! The two feature in the documentation that really are needed is explaining in general how views are handled, and how to obtain rigid body contact forces which is neede for pretty much all locomation tasks
Devs: does having fewer views/combining views make the rl more efficient?