[FEATURE] Enhance SAP coupler for coupling between Rigid body and Fem object.

[Libero0809]

Description

• Add Rigid body - Floor contact in the SAP solver.
• Add Rigid body - Fem Object contact in the SAP solver.
• Allow query of LBVH to return overflow flag.

Related Issue

Resolves #1450

Motivation and Context

Making SAP solver more general.

How Has This Been / Can This Be Tested?

Tests are added.

Screenshots (if appropriate):

test_fem_articulated.mp4

test_dragon_rigid.mp4

test_fem_rigid.mp4

Checklist:

• I read the CONTRIBUTING document.
• I followed the Submitting Code Changes section of CONTRIBUTING document.
• I tagged the title correctly (including BUG FIX/FEATURE/MISC/BREAKING)
• I updated the documentation accordingly or no change is needed.
• I tested my changes and added instructions on how to test it for reviewers.

• I have added tests to cover my changes.
• All new and existing tests passed.

[hughperkins]

(Note: I'm unsubscribing myself, so please re-tag me if needed)

[Libero0809]

@duburcqa First round fix finished.

[Libero0809]

Please do so. Thank you!

…

On Tue, Aug 12, 2025 at 8:38 PM Alexis DUBURCQ ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In genesis/engine/couplers/sap_coupler.py <#1458 (comment)> : > + # Step 1: Solve w st. L^T @ w = y + for i_b, i_e in ti.ndrange(self._B, self.rigid_solver.n_entities): + if not self.batch_pcg_active[i_b]: + continue + entity_dof_start = self.rigid_solver.entities_info.dof_start[i_e] + entity_dof_end = self.rigid_solver.entities_info.dof_end[i_e] + n_dofs = self.rigid_solver.entities_info.n_dofs[i_e] + for i_d_ in range(n_dofs): + i_d = entity_dof_end - i_d_ - 1 + out[i_b, i_d] = vec[i_b, i_d] + for j_d in range(i_d + 1, entity_dof_end): + out[i_b, i_d] -= self.rigid_solver.mass_mat_L[j_d, i_d, i_b] * out[i_b, j_d] + + # Step 2: z = D^{-1} w + for i_b, i_d in ti.ndrange(self._B, self.rigid_solver.n_dofs): + if not self.batch_pcg_active[i_b]: + continue + out[i_b, i_d] *= self.rigid_solver.mass_mat_D_inv[i_d, i_b] + + # Step 3: Solve x st. L @ x = z + for i_b, i_e in ti.ndrange(self._B, self.rigid_solver.n_entities): + if not self.batch_pcg_active[i_b]: + continue + entity_dof_start = self.rigid_solver.entities_info.dof_start[i_e] + entity_dof_end = self.rigid_solver.entities_info.dof_end[i_e] + n_dofs = self.rigid_solver.entities_info.n_dofs[i_e] + for i_d in range(entity_dof_start, entity_dof_end): + for j_d in range(entity_dof_start, i_d): + out[i_b, i_d] -= self.rigid_solver.mass_mat_L[i_d, j_d, i_b] * out[i_b, j_d] The only way may be to have some very generic ti.func that does LTDL decomposition and pass everything that it necessary in arguments, without any simulation related data structure. I can do this for you tomorrow if necessary. — Reply to this email directly, view it on GitHub <#1458 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AM3M7IJQTC23TYMLEQOG7ZD3NKXJVAVCNFSM6AAAAACCKJ23VOVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZTCMJTHAZTGNBYGE> . You are receiving this because you authored the thread.Message ID: ***@***.***>