Zero the Newton shape margin for ANYmal-D physX-Newton sim2sim

[Double7sBurger]

Description

ANYmal-D fails on rough terrain under the Newton MJWarp backend: the robot is
terminated by the 1 N base_contact check while it is still upright and walking
normally.

The cause is a collision-geometry interaction. ANYmal-D's base collider is a
rotationally-symmetric capsule of radius 0.12 m, which leaves only a few millimetres
of clearance to the thigh colliders during normal gait. Newton's shape margin extends
the collision surface outward and the margins of both shapes are summed, so the shared
1 cm default turns that clearance into near-constant base↔thigh overlap.

Measured under Newton MJWarp with a PhysX-trained checkpoint (4096 envs, 3000 steps,
~12k episodes):

shape margin	success rate
0.01 (current default)	28.9%
0.005	71.2%
0.0 (this PR)	93.1%

PhysX reference on the same checkpoint: 96.2%.

The override is scoped to AnymalDRoughEnvCfg. Other robots keep the shared 1 cm
default, which they need for stable contact on triangle-mesh terrain. Verified at
runtime that ANYmal-D resolves to margin 0.0 while Go2, H1 and Cassie still resolve to
0.01, including when several tasks are constructed in the same process.

The root cause is really the asset — a rotationally-symmetric capsule is a poor
approximation of a flat chassis. Shrinking the capsule to r = 0.10 m recovers the same
policy to 95.6%, and margin 0.005 on top of that reaches 96.1%. That is the better
long-term fix, but it changes a shipped asset and would affect existing checkpoints, so
this PR takes the minimal config-only mitigation.

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (existing functionality will not work without user modification)
• Documentation update

Checklist

• I have read and understood the contribution guidelines
• I have run the pre-commit checks with ./isaaclab.sh --format
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I have added a changelog fragment under source/<pkg>/changelog.d/ for every touched package (do not edit CHANGELOG.rst
  or bump extension.toml — CI handles that)
• I have added my name to the CONTRIBUTORS.md or my name already exists there

[greptile-apps]

Greptile Summary

This PR scopes Newton’s zero shape margin to ANYmal-D locomotion configurations.

• Overrides the ANYmal-D Newton MJWarp default shape margin to prevent spurious base-contact termination.
• Adds an isaaclab_tasks changelog fragment documenting the fix and its task-specific scope.

Confidence Score: 5/5

The PR appears safe to merge, with no actionable correctness or security issues identified.

The override reaches the Newton MJWarp configuration selected for ANYmal-D, preserves the inherited shape configuration defaults other than margin, and does not alter other robot configurations.

Important Files Changed

Filename	Overview
source/isaaclab_tasks/isaaclab_tasks/core/velocity/config/anymal_d/rough_env_cfg.py	Applies the zero-margin override to the mutable ANYmal-D Newton preset before backend resolution, while leaving other robots unchanged.
source/isaaclab_tasks/changelog.d/henry-anymald-newton-margin.rst	Accurately documents the scoped ANYmal-D Newton collision-margin fix.

_{Reviews (1): Last reviewed commit: "Zero the Newton shape margin for ANYmal-..." | Re-trigger Greptile}

[isaaclab-review-bot]

Isaac Lab Review Bot

The PR scopes the Newton MJWarp shape-margin override to AnymalDRoughEnvCfg and adds the required isaaclab_tasks changelog fragment. The task-local configuration path and post-initialization ordering are consistent with the intended fix.

• Design and architecture: The shared Newton default remains unchanged for other robots, while ANYmal-D receives the zero-margin override after its inherited simulation configuration is initialized. The direct Newton import introduces backend coupling, but the supplied context does not establish that isaaclab_newton is optional or unavailable in supported task installations, so the reported import failure is not directly evidenced.
• API: No public symbols or signatures change. The behavioral change is limited to the Newton MJWarp configuration for the ANYmal-D rough task and its derived configurations. The changelog fragment is correctly located and describes the fix in the appropriate category.
• Implementation: The override executes after super().__post_init__(), ensuring it replaces the inherited shape configuration when newton_mjwarp is present. Constructing a fresh NewtonShapeCfg also restores any other shape fields to that class's defaults; this is a non-blocking tradeoff with the current code but should be kept in mind if additional per-task shape settings are introduced later.

No blocking issues. No inline issue met the actionable-evidence threshold; the assessment above records the review feedback.

Automated review; human maintainers own approval decisions.

[AntoineRichard]

I'm not sure this should be merged if the better fix is to update the asset. We can update the asset so I would favor that option over changing this setting. If you need help on how to update the asset let us know!

[AntoineRichard]

Blocking for now.

[Double7sBurger]

Hi! I think change on both assets or repo code would work, could you please send me some tutorials about how to modify the asset? Thank you!

…

---- 回复的原邮件 ---- | 发件人 | Antoine ***@***.***> | | 发送日期 | 2026年08月01日 00:16 | | 收件人 | isaac-sim/IsaacLab ***@***.***> | | 抄送人 | Henry ***@***.***>, Author ***@***.***> | | 主题 | Re: [isaac-sim/IsaacLab] Zero the Newton shape margin for ANYmal-D physX-Newton sim2sim (PR #6828) | @AntoineRichard requested changes on this pull request. Blocking for now. — Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS and Android. Download it today! You are receiving this because you authored the thread.Message ID: ***@***.***>