search

isaac-sim / IsaacLab

Unified framework for robot learning built on NVIDIA Isaac Sim
https://isaac-sim.github.io/IsaacLab
Other
2.12k stars 859 forks source link

[Bug Report] filter_prim_paths_expr in contact sensor not working #828

Open supersglzc opened 2 months ago

supersglzc commented 2 months ago

Describe the bug

I recently updated my IsaacSim to 4.1.0 and pulled the newest IsaacLab. However, when I run add_sensors_on_robot.py, it gives the following warning: 

[Warning] [omni.usd] Warning (secondary thread): in GetInstanceIndices at line 2190 of /buildAgent/work/ac88d7d902b57417/USD/pxr/usdImaging/usdImaging/pointInstancerAdapter.cpp -- ProtoIndex 1 out of bounds (prototypes size = 1) for (/Visuals/ContactSensor, /Visuals/ContactSensor.proto1_mesh_id0)

[Warning] [omni.usd] Warning (secondary thread): in GetInstanceIndices at line 3256 of /buildAgent/work/ac88d7d902b57417/USD/pxr/usdImaging/usdImaging/delegate.cpp -- Empty InstanceIndices (/Visuals/ContactSensor, /Visuals/ContactSensor.proto1_mesh_id0)

I also tested filter_prim_paths_expr, but seems like the filtering does not work as expected. I modified the contact sensor cfg to 

contact_forces = ContactSensorCfg(
        prim_path="{ENV_REGEX_NS}/Robot/LF_FOOT", update_period=0.0, history_length=6, debug_vis=True, track_air_time=True, filter_prim_paths_expr=["{ENV_REGEX_NS}/Robot/RF_FOOT"]
    )

This only uses one prim and one filter prim. I suppose this cfg means that we only want to detect the contact between LF_Foot and RF_Foot. However, the sensor still detects the contact between the LF_Foot and the ground plane.

I am wondering if I misunderstand something or this is actually a bug.

Mayankm96 commented 2 months ago

Are you using the right attribute? The net_forces_w reports you still the total net normal force. The filtered forces are in a separate attribute force_matrix_w. Please check the documentation below.

https://isaac-sim.github.io/IsaacLab/source/api/lab/omni.isaac.lab.sensors.html#omni.isaac.lab.sensors.ContactSensorData.force_matrix_w

supersglzc commented 2 months ago

I used the current_air_time to detect if there is a contact. I found that this was computed by net_forces_w so that's why it didn't work. Now I changed to manually compute the norm using force_matrix_w.

Also, do you have any idea why the warning pops up? The warning is not there if I don't use filter_prim_paths_expr.

supersglzc commented 2 months ago

Follow-up: I occasionally encountered this error on some usd objects. Basically the net_forceis not zero but the force_matrix is always zero. I converted the usd from urdfs. Thanks! @Mayankm96 [Warning] [omni.physx.tensors.plugin] GPU contact filter for collider '/World/envs/env_3853/Object_0' is not supported