Fix Kamino contact capacity allocation by rubengrandia · Pull Request #403 · disneyresearch/newton

rubengrandia · 2026-07-30T15:19:00Z

Description

Fixes several Kamino contact-capacity bugs uncovered during the investigation of #323. Also touches warning behavior discussed in #115.

Changes

Plane-pair capacity accounting: max_contacts_for_shape_pair() now returns meaningful estimates for canonical (PLANE, X) pairs instead of (0, 0). A bug in the implementation missed the values we had assigned to those pairs.
External collision path (use_collision_detector=False): SolverKamino now sizes ContactsKamino from the Newton CollisionPipeline estimate when model.rigid_contact_max == 0, making allocation order-independent with the pipeline. This also removes the warnings that resulted from initializing Kamino before the contact pipeline.
Newton → Kamino conversion: scan all Newton contact slots (rigid_contact_max) so a saturated world cannot prevent later worlds from filling their per-world caps. This was a subtle bug that only happens when the external contact capacity is higher than the internal one.
CollisionDetector capacity resolver: replace equal model-total split with per-world geometry/pair-based budgets, proportional capping via max_contacts, and explicit max_contacts_per_world override.
Docs/warnings: clarify that max_contacts caps (not floors) the geometry estimate; soften conversion overflow warnings and restrict Kamino→Newton export warnings to only happen when Kamino's internal detector is used.

Checklist

New or existing tests cover these changes
The documentation is up to date with these changes
CHANGELOG.md has been updated (if user-facing change)

Co-authored-by: Cursor <cursoragent@cursor.com>

…ics#3305)

…s#3599)

…s#3455) Co-authored-by: camevor <camevor@nvidia.com>

Signed-off-by: Alain Denzler <adenzler@nvidia.com>

…cs#3601)

…sics#3488) Co-authored-by: linyaodong <yaodong.lin@lightwheel.ai> Co-authored-by: Dylan Turpin <dturpin@nvidia.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

…ics#3122) Signed-off-by: JC <jumyungc@nvidia.com> Signed-off-by: JC Chang <jumyungc@nvidia.com> Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

…ton-physics#3351)

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: Milad Rakhsha <mrakhsha@nvidia.com>

Co-authored-by: Cursor <cursoragent@cursor.com>

…3610) Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: Milad Rakhsha <mrakhsha@nvidia.com>

Co-authored-by: Cursor <cursoragent@cursor.com>

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…-physics#3657)

Signed-off-by: JC Chang <jumyungc@nvidia.com>

…nd filter template (newton-physics#3656)

… when approximating (newton-physics#3712) Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-authored-by: Eric Heiden <eric-heiden@users.noreply.github.com>

…s#3711) Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

Co-authored-by: Dex Smither <dsmither@nvidia.com>

…import (newton-physics#3654) Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

…n-physics#3707)

Co-authored-by: Viktor Reutskyy <vreutskyy@nvidia.com>

…ysics#3580) Co-authored-by: Eric Heiden <eric-heiden@outlook.com>

…-physics#3728) Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

…els (newton-physics#3664)

…ty estimate.

nvtw and others added 30 commits July 21, 2026 11:37

Teleop benchmark (newton-physics#3269)

50c6564

[Kamino] notify shape property updates (newton-physics#3577)

90bc2a7

Co-authored-by: Cursor <cursoragent@cursor.com>

Remove redundant math functions in Kamino (newton-physics#3583)

3fe35e6

Use analytic primitive SDFs for hydro texture generation (newton-phys…

709da0c

…ics#3305)

Update PADMM test tolerances (newton-physics#3586)

200f195

Require docstrings for test methods (newton-physics#3581)

ce603b3

Add rebuildable sparse implicit MPM grids (newton-physics#3412)

288fc9a

Add NVTX Scope for SensorTiledCamera rendering (newton-physics#3590)

3e06925

Fix Featherstone ball joint damping (newton-physics#3299)

0b45703

Create optional xforms argument for replicate function (newton-physic…

215c2ca

…s#3599)

Fix spurious Kamino floating-base reset warning (newton-physics#3563)

38429fc

Validate ArticulationView masks before kernel launches (newton-physic…

391c0f8

…s#3455) Co-authored-by: camevor <camevor@nvidia.com>

Update Menagerie pin for Robotiq masses (newton-physics#3555)

ac6d562

Surface additional simulation benchmark metrics (newton-physics#3566)

a051c39

Fix batched MuJoCo site poses (newton-physics#3389)

621275c

Signed-off-by: Alain Denzler <adenzler@nvidia.com>

Add compiled regex support to label selection (newton-physics#3584)

3177670

Enforce output checks for more examples (newton-physics#3527)

1ea8c1c

Dedupe USD single-DOF joint resolution (newton-physics#3323)

add2c97

Fix explicit mass on MJCF mesh geoms (newton-physics#3595)

1daa772

Fix USD axial visual scaling (newton-physics#3368)

2928f5a

Signed-off-by: Alain Denzler <adenzler@nvidia.com>

Reject inconsistent batched particle arrays (newton-physics#3458)

84c0ec8

Gate benchmark CI on runtime metrics (newton-physics#3604)

72559ba

Update vulnerable locked dependencies (newton-physics#3603)

7fedcaa

Load solver backends lazily to speed up import newton (newton-physi…

4a6e036

…cs#3601)

Add a joint_dof_mask argument to newton.ik.IKSolver. (newton-phy…

e8d9327

…sics#3488) Co-authored-by: linyaodong <yaodong.lin@lightwheel.ai> Co-authored-by: Dylan Turpin <dturpin@nvidia.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

Fix USD site import traversal (newton-physics#3367)

b36798a

Split VBD cable stretch/shear and bend/twist constraints (newton-phys…

c139dc4

…ics#3122) Signed-off-by: JC <jumyungc@nvidia.com> Signed-off-by: JC Chang <jumyungc@nvidia.com> Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

Split mesh subsets on binding structure, not material vocabulary (new…

fda8c48

…ton-physics#3351)

DVI solver for Kamino (newton-physics#3570)

f562c39

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: Milad Rakhsha <mrakhsha@nvidia.com>

Add Kamino restitution verification (newton-physics#3588)

e20fca2

Co-authored-by: Cursor <cursoragent@cursor.com>

rubengrandia and others added 29 commits July 27, 2026 07:28

[Kamino] Deduplicate FK constraint kernels (newton-physics#3614)

c487bc8

Co-authored-by: Cursor <cursoragent@cursor.com>

Performance improvements for Kamino's direct solvers (newton-physics#…

cabea00

…3610) Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: Milad Rakhsha <mrakhsha@nvidia.com>

Fix contact residual metric (newton-physics#3666)

4d6ca9b

[Kamino] Skip unused material updates (newton-physics#3677)

bfe8c31

Co-authored-by: Cursor <cursoragent@cursor.com>

Disable implicit VBD Dahl-friction defaults (newton-physics#3636)

c455e15

Honor MJCF geom inertia selection (newton-physics#3596)

de2d06f

Add Show Ground visualization toggle to the viewer (newton-physics#3626)

bd44951

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Remove redundant wp.static wrappers and fix a uint32 overflow (newton…

d124e5b

…-physics#3657)

Move VBD contact geometry to pipeline (newton-physics#3652)

9b0fcbf

Signed-off-by: JC Chang <jumyungc@nvidia.com>

Fix USD collider mass aggregation (newton-physics#3594)

0827c76

Startup Optimization: hydro-elastics, mesh creation, builder merge, a…

6f1453b

…nd filter template (newton-physics#3656)

Fix small-angle revolute joint reconstruction (newton-physics#3706)

f094b85

Load static USD visual shapes by default (newton-physics#3694)

e9aabe7

Decide USD collider visibility from purpose, and keep the render mesh…

92ae945

… when approximating (newton-physics#3712) Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-authored-by: Eric Heiden <eric-heiden@users.noreply.github.com>

Break the viewer shape count into visual and collision (newton-physic…

9166491

…s#3711) Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

Speedup material USD loading with clustering alg (newton-physics#3710)

831076f

Fix PR author auto-assignment (newton-physics#3708)

70765c1

Co-authored-by: Dex Smither <dsmither@nvidia.com>

Warn on mirrored (negative-determinant) rigid body transforms in USD …

825f0c9

…import (newton-physics#3654) Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

Update Kamino Jacobians to Newton free joint wrench convention (newto…

0a3e9e1

…n-physics#3707)

Bump mujoco-warp to 3.10.0.3 (newton-physics#3635)

1c681ac

Co-authored-by: Viktor Reutskyy <vreutskyy@nvidia.com>

Map contact kf to MuJoCo solreffriction for elliptic cones (newton-ph…

1f675f2

…ysics#3580) Co-authored-by: Eric Heiden <eric-heiden@outlook.com>

Restrict release build permissions (newton-physics#3648)

67313ba

Read newton:mimicCoef0 as degrees for angular mimic followers (newton…

930ef2d

…-physics#3728) Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

Fix SolverVBD over-allocating soft-contact buffers in multi-world mod…

36b2ddf

…els (newton-physics#3664)

Resolve initialization order by using Newton CollisionPipeline capaci…

8edb225

…ty estimate.

Fix kernel launch dim for contact conversion

30cdd17

Fix capacity estimation for pairs with planes.

dd362a1

fix doc string around max_contacts_per_world

d6e1e37

Fix contact capacity computation in the Kamino detector

f2748ca

rubengrandia closed this Jul 30, 2026

Navigation Menu

Uh oh!

Fix Kamino contact capacity allocation - #403

Fix Kamino contact capacity allocation#403
rubengrandia wants to merge 83 commits into
disneyresearch:mainfrom
rubengrandia:rgrandia/fix_kamino_contact_capacity

rubengrandia commented Jul 30, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants

Uh oh!

Conversation

rubengrandia commented Jul 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Changes

Checklist

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants

rubengrandia commented Jul 30, 2026 •

edited

Loading