newton/_src/geometry/kernels.py (2)

1442-1455: Avoid magic threshold; use finiteness check for invalid contacts.

Comparing to 1.0e5 is brittle and unit-dependent. The primitive returns +inf for invalid contacts; prefer a finiteness check.

Apply this diff:

-    # Check if this contact point is valid (primitive function returns wp.inf for invalid contacts)
-    valid = distance < 1.0e5  # Use a reasonable threshold instead of exact wp.inf comparison
+    # Check if this contact point is valid (primitive returns +inf for invalid contacts)
+    valid = wp.isfinite(distance)

If wp.isfinite is unavailable here, fall back to an explicit comparison:
valid = distance < wp.inf

---

2163-2170: Early-return before doing work when contact is invalid.

Micro-optimization and clearer flow: check valid before flipping the normal.

Apply this diff:

-        # Flip the normal since we flipped the arguments
-        normal = -neg_normal
-        if not valid:
-            return
+        if not valid:
+            return
+        # Flip the normal since we flipped the arguments
+        normal = -neg_normal

Configuration used: Path: .coderabbit.yml

Review profile: CHILL

Plan: Pro

newton/tests/test_rigid_contact.py (1)

140-147: Add a rotated cylinder case to also validate radius-based offset

To catch regressions when the cylinder axis is horizontal (rest height should equal radius), add a second cylinder rotated from Z to Y (similar to the capsule). Example:

         expected_end_positions.append(wp.vec3(5.0, y_pos, 0.3 * scale))
 
+        # Rotated cylinder (axis along Y): rest height equals radius
+        b = builder.add_body(xform=wp.transform(wp.vec3(6.0, y_pos, 1.0), wp.quat_identity()))
+        builder.add_joint_free(b)
+        xform_cyl = wp.transform(wp.vec3(), quat_between_axes(newton.Axis.Z, newton.Axis.Y))
+        builder.add_shape_cylinder(
+            body=b,
+            xform=xform_cyl,
+            radius=0.1 * scale,
+            half_height=0.3 * scale,
+        )
+        expected_end_positions.append(wp.vec3(6.0, y_pos, 0.1 * scale))

Configuration used: Path: .coderabbit.yml

Review profile: CHILL

Plan: Pro

newton/tests/test_rigid_contact.py (4)

140-147: Cylinder addition looks correct; consider also testing rotated orientation.

Upright cylinder’s expected z=half_height is correct. To fully cover the cylinder–plane bug, add a rotated cylinder (axis along X or Y) and expect z=radius for that case. This catches orientation-related regressions.

---

149-156: Mesh shift and expectations updated — add a quick length check.

The move to x=7.0 and expected z=0.3·scale are consistent. Add a guard to ensure the number of expected positions matches the number of dynamic bodies to catch future divergences early.

Apply this minimal check near Line 183:

   body_q = state_0.body_q.numpy()
-  expected_end_positions = np.array(expected_end_positions)
+  expected_end_positions = np.array(expected_end_positions)
+  assert expected_end_positions.shape[0] == model.body_count

---

44-47: Remove redundant builder setup.

builder is created and configured at Lines 44–47 but immediately replaced at Lines 99–104; the first instance (and add_ground_plane() there) is unused.

Apply:

-    builder = newton.ModelBuilder()
-    builder.default_shape_cfg.ke = 1e4
-    builder.default_shape_cfg.kd = 500.0
-    builder.add_ground_plane()

Also applies to: 99-104

---

194-194: Drop explicit njmax=150 or document rationale. MuJoCo defaults njmax to nefc when unspecified (and uses the larger of the default or provided value); remove the override for model-agnostic stability or add a brief note explaining why 150 is required here.

Configuration used: Path: .coderabbit.yml

Review profile: CHILL

Plan: Pro

Learnt from: preist-nvidia
PR: newton-physics/newton#579
File: newton/examples/example_mujoco.py:350-354
Timestamp: 2025-08-19T12:27:30.629Z
Learning: In Newton examples, Example.use_mujoco is a high-level attribute that controls whether to use MuJoCo solver vs other solvers (like XPBD), while SolverMuJoCo.use_mujoco_cpu controls the backend within MuJoCo (CPU vs Warp). These are separate concepts serving different purposes - the PR rename only applies to the solver parameter, not the Example class attributes.