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
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 flex interpolation stiffness matrix within the implicit/implicitfast solvers is now built and factorized in a banded format instead of a dense one. This involves:
- Calculating the bandwidth based on the sparsity of the mass/damping matrix and the connectivity within flex cells.
- Allocating and populating a banded matrix `H`.
- Using `mju_cholFactorBand` and `mju_cholSolveBand` for factorization and solving.
This change improves performance for flexes with many DOFs but local coupling.
PiperOrigin-RevId: 901297952
Change-Id: I3efe06353d1903ea65ab30dc49685cede228bb68
These flexes use only 24 DOFs (3 per vertex of the bounding box), while colliding with the full high resolution mesh.
On an 8x8x8 cube, the performance using DOFs at all vertices is
```
Simulation time : 18.74 s
Steps per second : 533
Realtime factor : 0.53 x
Time per step : 1874.4 µs
Contacts per step : 114.88
Constraints per step : 3322.51
Degrees of freedom : 1536
```
With the new implementation, it is the following:
```
Simulation time : 1.82 s
Steps per second : 5507
Realtime factor : 5.51 x
Time per step : 181.6 µs
Contacts per step : 38.84
Constraints per step : 155.36
Degrees of freedom : 24
```
PiperOrigin-RevId: 721008829
Change-Id: I833df027527db578d86667cc4b24295bcf6f7d22
`flex_xvert0` stored the Cartesian position of the vertices in `qpos0`. This array was unused, so it is now replaced with `flex_vert0`, which still contains the positions of the vertices, but normalized on the bounding box of the flex so that the coordinates are in `[0,1]^m->flex_dim`. This will be useful in the future for using different interpolation methods for computing flex deformations.
PiperOrigin-RevId: 691779361
Change-Id: I5c4223103cd4558f4e268fe3e8d8177541e4754f
These functions names and argument ordering are more consistent with the rest of the API.
PiperOrigin-RevId: 643788290
Change-Id: I783eda8021b80b82098e23ed95669b102bb82508
- remove mj_crbSkip since this was only utilized during compile time to compute m->dof_m0, replaced with mj_crbDiag to compute only the diagonals for dof_m0,
- change row backward pass to forward pass to optimize processing diagonal blocks,
- and various code refactoring.
PiperOrigin-RevId: 556799924
Change-Id: I0a18d7a45862860fdddd7150540ddc8d64b01b29
Yuval Tassa