Significant non-determinism in unified collision pipeline for anymal_c_walking example #1505 by vreutskyy · Pull Request #1588 · newton-physics/newton

vreutskyy · 2026-02-09T15:44:55Z

Description

The ANYmal C walking example was intermittently failing because the robot would stumble
and fall on the procedural terrain. This was originally attributed to non-deterministic
contact ordering in the unified collision pipeline (wp.atomic_add on GPU), but the
root cause turned out to be the walking policy not being robust enough for the terrain
features combined with the small URDF foot collision spheres.

Note: this is not specific to the unified collision pipeline. The previous workaround
(PR #1574) switched to MuJoCo contacts, which masked the problem because MuJoCo's
collision does not support triangle mesh collision — it replaces the terrain mesh with
its bounding box, so the robot effectively walks on a flat surface.

Changes:

Double the radius of foot collision spheres after URDF loading, giving the contact
solver a larger support area on uneven ground
Reduce wave terrain amplitude from 0.15 to 0.1, which was still too aggressive even
with bigger feet
Remove the forced fallback to MuJoCo contacts, allowing Newton's collision pipeline
to be used with the terrain mesh as intended

Newton Migration Guide

Please ensure the migration guide for warp.sim users is up-to-date with the changes made in this PR.

The migration guide in docs/migration.rst is up-to date

Before your PR is "Ready for review"

Necessary tests have been added and new examples are tested (see newton/tests/test_examples.py)
Documentation is up-to-date
Code passes formatting and linting checks with pre-commit run -a

Summary by CodeRabbit

New Features
- Added geometry type support to the robot example configuration.
Bug Fixes
- Improved robot stability on uneven terrain by reducing terrain wave amplitude.
- Enhanced foot collision detection by enlarging foot collision geometry for more robust contact.
Refactor
- Collision-handling now respects the configured collision-pipeline flag, simplifying and clarifying runtime behavior.

…c_walking example newton-physics#1505 Signed-off-by: Viktor Reutskyy <vreutskyy@nvidia.com>

coderabbitai · 2026-02-09T15:45:11Z

📝 Walkthrough

Walkthrough

Imported GeoType; enlarged foot collision sphere radii; reduced wave terrain amplitude to 0.1; removed a runtime override so collision-pipeline creation now respects the provided use_mujoco_contacts flag.

Changes

Cohort / File(s)	Summary
Example configuration & collision logic `newton/examples/robot/example_robot_anymal_c_walk.py`	Added `GeoType` to imports. Doubled radii for foot collision spheres (SPHERE). Reduced wave terrain amplitude from 0.15 to 0.1. Removed runtime override that forced Mujoco collision pipeline; pipeline creation now follows the `use_mujoco_contacts` flag.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

Fix example_robot_anymal_c_walk.py #1574: Modifies collision/contact selection and initialization logic in the same example file (changes around use_mujoco_contacts and collision pipeline creation).

Suggested reviewers

adenzler-nvidia
eric-heiden

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title references issue `#1505` and describes the main problem (non-determinism), but omits the actual solution implemented (increased foot sphere radius, reduced terrain amplitude, unified pipeline enabled). The title identifies the problem rather than the primary change.
Linked Issues check	✅ Passed	The PR successfully addresses all coding requirements from issue `#1505`: doubles foot collision sphere radius for stability, reduces terrain wave amplitude, and enables the unified collision pipeline by removing the forced MuJoCo fallback, directly fixing the non-determinism and flakiness issues.
Out of Scope Changes check	✅ Passed	All changes are directly scoped to fixing the anymal_c_walking example stability issues: foot sphere enlargement, terrain amplitude reduction, and collision pipeline control. The addition of GeoType to imports is a minimal, necessary change to support the modifications.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

📝 Generate docstrings

🧪 Generate unit tests (beta)

Create PR with unit tests
Post copyable unit tests in a comment

No actionable comments were generated in the recent review. 🎉

🧹 Recent nitpick comments

newton/examples/robot/example_robot_anymal_c_walk.py (1)
107-113: Consider filtering by body name to future-proof the sphere enlargement.

The loop doubles every sphere in the model, relying on the assumption that only the four foot shanks carry sphere colliders. If the URDF is ever updated with additional sphere shapes (e.g., for the base or hip), they would be silently enlarged too.

A body-name filter would make the intent explicit and more robust:
♻️ Optional: scope to shank links only
-        for i in range(len(builder.shape_type)):
-            if builder.shape_type[i] == GeoType.SPHERE:
-                r = builder.shape_scale[i][0]
-                builder.shape_scale[i] = (r * 2.0, 0.0, 0.0)
+        shank_bodies = {builder.body_key.index(n) for n in ("LF_SHANK", "RF_SHANK", "LH_SHANK", "RH_SHANK")}
+        for i in range(len(builder.shape_type)):
+            if builder.shape_type[i] == GeoType.SPHERE and builder.shape_body[i] in shank_bodies:
+                r = builder.shape_scale[i][0]
+                builder.shape_scale[i] = (r * 2.0, 0.0, 0.0)

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

…collision-pipeline-for-anymal_c_walking-example

codecov · 2026-02-09T15:56:58Z

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!

coderabbitai · 2026-02-09T16:15:59Z

Caution