[Question] Missing Contact because of bouncing

[george-nehma]

Question

I have a contact sensor set up between two Convex Decomposition meshes, with a dt 1/60s and decimation of 6, and the contact sensor defined as

contact_sensor: ContactSensorCfg = ContactSensorCfg(
       prim_path="/World/envs/env_.*/Robot/MainBody", 
       update_period=0.0, 
       track_air_time = True,
       debug_vis=True,
       history_length=5,
       filter_prim_paths_expr=["/World/ground"]

I check for contact with contact = self._contact_sensor.data.current_contact_time.squeeze(1) > 0.1.

Yet when i have a collision at around 5m/s sometimes it misses the contact and will not flip the flag on the contact variable. How should I go about fixing this to capture all contacts?

[RandomOakForest]

Thank you for posting this. Here is a summary to review that may be of help, still under review, though. You are missing some contacts because your contact-time check does not line up with the actual sensor update timing and contact duration at high impact speeds.¹

Why contacts are missed

• current_contact_time is incremented only when the bodies are currently in contact and reset to 0 when they separate.¹
• The helper logic in Isaac Lab treats “in contact within the last $dt$” as 0.0 < current_contact_time < dt + abs_tol.¹
• With dt = 1/60 and a 5 m/s impact, contacts may be very short (one or two physics substeps). If your check is > 0.1, you can easily get a situation where:
  • At frame k, contact has just started and current_contact_time is still below 0.1.
  • At frame k+1, the contact has already ended and current_contact_time is reset to 0.
  • Result: your > 0.1 condition is never satisfied even though a contact occurred.¹

How to fix it

1. Use the provided helper instead of a raw threshold
   Isaac Lab’s contact sensor exposes utilities like compute_first_contact(dt) / compute_in_contact(dt) that directly encode the “contact within last dt” logic:¹

# dt here should match your RL/control step (e.g. 1/60 or decimation * 1/60)
in_contact = self._contact_sensor.compute_in_contact(dt=self.step_dt)
# or for edges:
first_contact = self._contact_sensor.compute_first_contact(dt=self.step_dt)
first_air = self._contact_sensor.compute_first_air(dt=self.step_dt)

Then use:

contact_flag = in_contact.squeeze(1) > 0

2. Align dt with your control/decimation
   • If physics runs at $1/60$ and your RL/control step is every 6 physics steps, then your effective step for “did a contact happen since last action?” is $dt_{\text{RL}} = 6/60 = 0.1$ s.
   • Pass that RL step into compute_in_contact(dt=0.1) so you capture any contact that occurred between RL steps.¹
3. Stop using > 0.1 on current_contact_time
   • Instead, treat any positive current_contact_time as “currently in contact”:

contact_now = self._contact_sensor.data.current_contact_time.squeeze(1) > 0.0

If you really want a duration threshold, make it < dt (i.e., “has been in contact for less than one step”), or use the helper methods which already implement that robustly.¹

4. Ensure the sensor period is not coarser than physics
   • With update_period=0.0, the sensor should update at the physics rate, which is good.²³
   • If you ever change this, do not set the sensor period larger than your desired detection resolution; otherwise fast contacts can be skipped.⁴⁵
5. Consider physics tuning if contacts are extremely brief
   • For very fast collisions and thin geometry, you can increase PhysX substeps or adjust contact_offset / rest_offset so contacts are generated earlier and last a bit longer.⁶
   • This makes “real” contact last more than one substep, which improves detection robustness.

Minimal practical change

For your current setup, the simplest robust solution is:

# Assume RL step = decimation * sim dt
self.rl_dt = self.decimation * (1.0 / 60.0)

# At each RL step:
in_contact = self._contact_sensor.compute_in_contact(dt=self.rl_dt)
contact_flag = in_contact.squeeze(1).any(dim=-1)  # per-env flag

Footnotes

1. https://github.com/isaac-sim/IsaacLab/discussions/2026 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

2. https://isaac-sim.github.io/IsaacLab/main/source/overview/core-concepts/sensors/contact_sensor.html ↩

3. https://docs.robotsfan.com/isaaclab_official/v2.0.1/source/overview/sensors/contact_sensor.html ↩

4. https://docs.isaacsim.omniverse.nvidia.com/4.2.0/advanced_tutorials/tutorial_advanced_isaac_contact_sensor.html ↩

5. https://docs.isaacsim.omniverse.nvidia.com/5.1.0/sensors/isaacsim_sensors_physics_contact.html ↩

6. https://docs.isaacsim.omniverse.nvidia.com/6.0.0/reference_material/sim_performance_optimization_handbook.html ↩

[george-nehma]

Hi @RandomOakForest,

Thank you for that very detailed answer, it makes a lot of sense. I tried to implement the .compute_in_contact method but returned an error

AttributeError: 'ContactSensor' object has no attribute 'compute_in_contact'

I had a look in the links you provided and think that maybe I implemented my Contact Sensor differently. I get the contact sensor from from isaaclab.sensors import ContactSensor, ContactSensorCfg but in https://docs.isaacsim.omniverse.nvidia.com/5.1.0/sensors/isaacsim_sensors_physics_contact.html I see it is implemented from from isaacsim.sensors.physics import ContactSensor.

I will try to implement it from IsaacSim not IsaacLab and see if this is able to solve my issue.

[RandomOakForest]

Yes, I'm reviewing the methods and will be updating the post as time permits. I hope the idea posted can get you unblocked. Thank you for your interest in Isaac Lab.

[george-nehma]

i create the sensor with

self.contact_sensor = ContactSensor(
prim_path="/World/envs/env.*/Robot/MainBody/Contact_Sensor",
name="Contact_Sensor",
frequency=60,
translation=np.array([0, 0, 0]),
min_threshold=0,
max_threshold=10000000,
radius=-1
)

yet I receive the error

File "/isaac-sim/extscache/omni.usd-1.13.10+69cbf6ad.lx64.r.cp311/omni/usd/_impl/utils.py", line 1257, in get_stage_next_free_path
raise ValueError(f"{path} is not a valid path")

<class 'ValueError'> /World/envs/env_.*/Robot/MainBody/Contact_Sensor is not a valid path
2026-02-06T19:26:26Z [22,493ms] [Error] [isaacsim.sensors.physics.impl.commands] Could not create contact sensor prim

Which I believe is because of the .* in the prim path. This sensor I believe only is built for IsaacSim and not compatible with IsaacLab.

[RandomOakForest]

I will move this to our Discussions section for follow up. Let us know if you need further help.