[Bug Report] Stiffness domain randomization does not affect the behavior

[youwyu]

If you are submitting a bug report, please fill in the following details and use the tag [bug].

Describe the bug

In Ant environment, setting the joint stiffness domain randomization with the same action, all robots have the same states.

Steps to reproduce

Ant-v0 environment.

randomize_actuator_gains:
      mode: 'reset'
      params:
        stiffness_distribution_params: [0.1, 2.0]
        operation: 'add'
        distribution: 'uniform'

Setting the same random action for all robots.

actions = self.alg.act(obs, privileged_obs)
N, A = actions.shape
single_random = torch.rand(1, A, device=actions.device, dtype=actions.dtype)
actions = single_random.expand(N, -1)

Then they have the same behavior. But they will have different behavior if I set domain randomization in mass, which means the domain randomization works for mass but not joint stiffness.

The following is the video for different joint stiffness with same mass.

rl-video-step-0.mp4

The following is the video for different mass with same joint stiffness.

randomize_rigid_body_mass:
      mode: 'reset'
      params:
        mass_distribution_params: [0.1, 5.0]
        operation: 'scale'
        distribution: 'uniform'

rl-video-step-0.mp4

.

System Info

Describe the characteristic of your environment:

• Commit: [e.g. 8f3b9ca]
• Isaac Sim Version: 4.5
• OS: Ubuntu 22.04
• GPU: RTX 4090
• CUDA: 12.8
• GPU Driver: 570.158.01

Checklist

• I have checked that there is no similar issue in the repo (required)
• I have checked that the issue is not in running Isaac Sim itself and is related to the repo

Acceptance Criteria

• [] Different subsequent states with different joint stiffness and same actions.

[RandomOakForest]

Thank you for posting this. The team will review.

[AntoineRichard]

Thanks @youwyu — "randomization runs but nothing changes" is exactly the kind of silent bug worth reporting. randomize_actuator_gains has since been rewritten as part of the actuator-ownership work (#6799, #6839): it now resolves per-group gain ownership and writes through backend-specific paths, with regression tests on both backends asserting the randomized kp/kd actually reach the solver — including test_implicit_storage_reuses_randomized_payload, which covers the implicit-actuator path Ant uses.

One aside: operation="add" with [0.1, 2.0] is a tiny perturbation next to Ant's default stiffness, so expect subtle effects. Reopen if it still looks inert.

---

🤖 This issue was reviewed as part of an AI-assisted triage of the Isaac Lab backlog, with automated verification against the current codebase. If we got this wrong, please reopen — we'd rather hear about it!