For example, `compile mujoco_menagerie/robotis_op3/scene.xml` now outputs
```
Compile 1 (cold cache):
total: 317.2 ms
assets: 284.8 ms (wall clock)
load: 616.1 ms
hull: 26.4 ms
poly: 137.8 ms
inert: 177.8 ms
bvh: 568.2 ms
octr: 1.6 ms
tex: 25.3 ms
other: 32.4 ms
Compile 2 (warm cache):
total: 79.9 ms
assets: 54.5 ms (wall clock)
load: 888.5 ms
hull: 0.0 ms
poly: 0.0 ms
inert: 0.0 ms
bvh: 0.0 ms
octr: 0.0 ms
tex: 21.4 ms
other: 25.3 ms
```
PiperOrigin-RevId: 917850214
Change-Id: Iaec86230bec0faf2e47820e20cbff61de5b2621e
mju_boxQP documents that only the lower triangle of the Hessian H is
read, but the gradient and search-direction updates inside
mju_boxQPoption still called the dense mju_mulMatVec, which reads the
upper triangle as well. This violated the documented contract and
prevented callers from safely leaving the upper triangle uninitialized.
Add a file-local mulMatVecSym helper that computes res = H*vec while
reading only the lower triangle of H (mirroring the convention of the
existing mulVecMatVecSym quadratic-form helper), and use it in place of
mju_mulMatVec in both call sites. Extend the BoxQP test suite with
UpperTrianglePoisoned, which fills the strict upper triangle of H with
NaN and verifies that the solver produces the same result as on the
clean symmetric input.
Reported by @lshdlut.
Fixes#3275
This gives a 3x speedup in implicitfast.
Also cleanup old code that was used in the dense factorization of the stiffness matrix before we switched to CG.
PiperOrigin-RevId: 915900315
Change-Id: Id6973c4bfd7d371a43ec6db982703969b23a3550
The flex interp stiffness matrix is Negative Semi-Definite (NSD). When forming the RHS for implicit integration, the term involving the velocity and stiffness should be added, not subtracted. A new test is added to ensure energy stability for flex interp stretch stiffness with the implicitfast integrator.
PiperOrigin-RevId: 914845245
Change-Id: Iaaf0914909128e64e195f17cc5f2f344a8a43bc2
Standard flex (flex_interp=0) with thin-plate bending treated bending forces purely explicitly. This caused contact-induced vertex vibrations and non-physical energy injection for flat resting sheets, because the solver treated each vertex as an independent mass during contact and contact normals are orthogonal to stretch constraints.
Fix: extend the existing preconditioned CG solver to include the constant bending stiffness K_bend in the implicit operator via matrix-free mat-vec.
PiperOrigin-RevId: 914774020
Change-Id: I45e0d6749abb6f873566203bccae956514b2576b
with a preconditioned Conjugate Gradient (CG) solver that operates
on the full system matrix.
The previous approach extracted flex DOFs into a reduced banded system,
factored it separately, and overwrote the global solve. This required
precomputed bandwidth (makeFlexBandwidth), parent-joint detection,
coupling corrections, and a FlexInterpContext struct — and only worked
for standalone flex trees without parent joints.
The new CG solver uses the already-factored global system (M - h*qDeriv)
as a preconditioner and adds the flex stiffness contribution via
matrix-free products (mjd_flexInterp_mulKD/mulK). This handles any
kinematic configuration — including flexes attached to articulated
chains or with parent joints — without sparsity pattern restrictions.
Before (`bunny_multicell`):
```
Simulation time : 50.80 s
Steps per second : 197
Realtime factor : 0.20 x
Time per step : 5080.3 µs
CG iters / step : 3.16
Contacts / step : 31.04
Constraints / step : 124.15
Degrees of freedom : 178
Dynamic memory usage : 0.4% of 100M
```
After:
```
Simulation time : 9.52 s
Steps per second : 1051
Realtime factor : 1.05 x
Time per step : 951.7 µs
CG iters / step : 3.21
Contacts / step : 30.90
Constraints / step : 123.61
Degrees of freedom : 178
Dynamic memory usage : 0.3% of 100M
```
PiperOrigin-RevId: 913758038
Change-Id: If5aa617b2d535c86aec9bd71c9e0003a2b38bdd7
This shell test was extremely slow (taking up to 26 minutes under ASAN). In any case the compilation pipeline takes one kick-the-tires step, so the RecompileCompare test effectively does the same thing.
PiperOrigin-RevId: 913646840
Change-Id: If659705b88d0ac64722f88ddf2ae377932e5ef0f
The normal jump residual in flex elasticity calculations now rotates the rest-frame normal jump into the current frame using the face's corotational quaternion before subtracting the jump from the current normal difference.
PiperOrigin-RevId: 913242127
Change-Id: Ia62b28ecccd59e79225737d8225ef4131b333c96
The maximum depth of the octree used for SDF generation can now be specified in the mesh definition via `mjsMesh::octree_maxdepth`. The default value is 6.
PiperOrigin-RevId: 912494999
Change-Id: I6d828d1d3999b99d1210a6829a05b04fac591e1f
Since the mjz format is just an archive for MJCF assets this seems like a sensible place. The upcoming mjz encoder will also need to make use of the full precision XML utility that is local to src/xml.
PiperOrigin-RevId: 912044746
Change-Id: I6a9fef24b1c3fec5b5edc3a8d4c8e22584273264
The new function mjs_getOriginSpec returns the mjSpec that was used to define a given mjsElement. Unlike mjs_getSpec, this value remains constant even after the element has been attached to a different model.
PiperOrigin-RevId: 911930951
Change-Id: Ia9cd79d9dfabc513d6121eadcffc5d41075224c9
This change adds a new passive force computation for flexes with elastic2d="bend" and dof="trilinear". The bending energy is based on the squared difference of normals between adjacent face elements at their shared edge midpoint. The edge data is precomputed during model compilation and stored in flex_bending.
PiperOrigin-RevId: 910772638
Change-Id: I3b12c7b7f1ba6ac1875df495d89e8cfec921ca80
Modify EigendecomposeStiffness to project out rigid body translations and rotations from the stiffness matrix eigenvectors. This prevents "ghost damping" from rigid body modes being incorrectly constrained. Update hollow_vs_solid.xml with solimp parameters for edge constraints. Adjust engine_core_constraint_test.cc to reflect the reduced number of equality constraints due to the projection.
PiperOrigin-RevId: 910631207
Change-Id: Ibaaa59bc030cfc6ae657d8f0d1b2a51002cc8d4e
Total testspeed runtime for `2humanoids100.xml` reduced by 19.6% (49.5 -> 39.8s).
PiperOrigin-RevId: 910608140
Change-Id: Ided0ae5bdb8e4e8196f84e08354a9ae8cfa5b626
mjSENS_INSIDESITE uses xipos to test containment, but for massless flex
parent bodies xipos equals the static body frame origin and does not
track the actual flex position.
Fix: in the INSIDESITE sensor case, when the object is a massless body
with positive subtree mass (i.e., a flex parent), use subtree_com
instead of xipos. This correctly reflects the mass-weighted centroid of
the flex child bodies without changing the global semantics of xipos.
PiperOrigin-RevId: 908745144
Change-Id: I527f0efb7b419345188411c9da883505156ebeea
The `vert0` array, used for parametric vertex coordinates in flexes, is now computed in the flex's local, unrotated frame when using interpolation. This ensures that the parametric coordinates are independent of the flex's initial orientation. A new test confirms that `flex_vert0` is identical for an unrotated and a rotated flex grid.
PiperOrigin-RevId: 908196464
Change-Id: I47d6bcc2bc5df581479d485480e0e947ec6d3ffd
When elastic2d="stretch" is set on an interpolated flexcomp, treat the bounding box boundary as membrane elements rather than volumetric cells. This computes plane-stress stiffness over the boundary faces and updates the runtime force/derivative kernels accordingly.
Interior vertex tracking (moving vertices that follow the deforming shell) is not yet implemented so all mesh vertices need to be on the bounding box surface or the background grid should have no interior nodes (i.e. cellcount should be 1 on at least one axis).
PiperOrigin-RevId: 907654080
Change-Id: I51b90e2f6a1d1b036f9604e42de20e377dc5d3f9
Total testspeed runtime for `2humanoids100.xml` reduced by 27.6% (63.4 -> 49.5s) due to early termination on small islands
PiperOrigin-RevId: 906910915
Change-Id: If55ad468c3680ef44eda7000455a77f8003b3122
Do not allow a mix of `elastic2d != none` with `dof = trilinear` since the latter assumes 3d elasticity.
Also, do not assume that `flex_interp > 0` in the engine. This will enable to use, e.g., `flex_interp = -1` to mean a linear surface finite element instead of a 3d finite element which is currently identified with `flex_interp = 1`.
PiperOrigin-RevId: 903852035
Change-Id: Ia6290b4a05e9e510ffb7f36d141cd525b40d3110
The reference node positions and the positions used for computing stiffness eigenvectors were previously stored in world frame. However, the runtime expects these quantities in the unrotated local frame. This caused non-zero constraint residuals and simulation instability when the grid was rotated — either by the parent body's initial orientation, or by the flexcomp's own frame attributes.
Rather than tracking each rotation source individually, this change extracts the total grid rotation directly from the cell geometry. All node positions are then un-rotated before computing the stiffness matrix.
PiperOrigin-RevId: 903232388
Change-Id: If877af89025ce1e61a76b38c29403d593d892749
Previously specs that were attached to some parent spec would resolve its asset paths relative to the modelfiledir of the parent spec. This means path resolution would change depending on the source of the parent spec. Instead, this change makes asset file path resolution relative to the "owning spec" i.e. the spec where the asset was created.
This enables workflows such as loading a parent spec via resource provider, then loading a child spec via `from_zip` or in memory providing `spec.assets` and resolution will work as intended.
PiperOrigin-RevId: 903207910
Change-Id: Ia58020ab372a3ceadf31e804d145e2ae53d8e5f9
The previous implementation only checked whether grid cells contained mesh vertices to determine occupancy. For coarse meshes with large faces, most cells were incorrectly marked empty and pruned—even cells fully inside the object volume. This change improves the algorithm by:
- Marking cells that overlap with any mesh element's AABB as non-empty.
- For surface meshes, running a flood-fill from the grid boundary through non-overlapping cells to identify truly exterior cells. This preserves empty interior cells, preventing incorrect pruning of the object's core.
- For volumetric meshes, defaulting to element-AABB overlap detection directly.
Limitations for non-watertight meshes: If the mesh contains holes larger than the grid cell size, the flood-fill will leak into the interior. In this case, all non-element cells (including interior ones) will be marked as empty.
PiperOrigin-RevId: 902675331
Change-Id: I5a84303a33d5ca7436213e7ce9aca3806c9f5a0f
This change introduces an optimization for flexcomp objects defined by a mesh. It identifies grid cells that do not contain any mesh vertices and marks them as empty. Nodes that are exclusively part of empty cells are pinned, preventing them from moving. Stiffness computations are skipped for empty cells, reducing computational cost. The total mass is now distributed only among the non-pinned nodes.
PiperOrigin-RevId: 902565735
Change-Id: Id0a9a685536d5e18a3e42124a25ab08ff3a918f2
Combine sparse vectors in-place by first counting total `nnz` and then working backwards from the end. This removes the need for temporary buffers in `mju_combineSparse` and its callers and speeds up the function by ~10%.
PiperOrigin-RevId: 902530210
Change-Id: I4f48c327103552ab968d3915399c6067367bec9f
Yuval Tassa