mj_makeModel returns NULL if any of the array sizes are negative. Callers of the function need to handle that case.
PiperOrigin-RevId: 457948558
Change-Id: I9a5c8c0d3293ed0a92c5791f00d3ed34c848e024
- Add `qH` and `qHDiagInv` to `mjData` to save factorized modified inertia.
- Add `mj_EulerSkip`, `mj_implicitSkip`, to `engine_forward.c`.
- Using the above functions, implement `mj_stepSkip` in `engine_derivative.c`.
- Add `mjd_stepFD` and `mjd_transitionFD` to `engine_derivative.c` to compute `mj_step` Jacobians.
- Exploit "Skip" functionality for speed.
- Correctly handle quaternion derivatives.
- Handle warmstarts and control limits.
PiperOrigin-RevId: 456584811
Change-Id: Iee8541f11e7b66feb8f431cb102d9bbe65461f79
- Introduced private function `mj_advance()` as single point of state & time advancement.
PiperOrigin-RevId: 456525969
Change-Id: Iae17217e305c7baf07cb5ecd1e1b84b936421097
When using MSAN, mjData->buffer is marked as poisoned, despite being initialized with zeroes.
This is useful for catching computations that accidentally use uninitialized values.
However, when using engine_print, it's OK to assume the buffer is initialized to zeros.
PiperOrigin-RevId: 456085776
Change-Id: I9d85e37fa82958eb54ac413140543a6581aa46ad
This will allow us to add fields that can't be cast to int or double, without breaking print.
PiperOrigin-RevId: 456005913
Change-Id: I1fccfd11466f4745f69172af0ae61be07642927c
The C++ standard library `std::istringstream` is not guaranteed to parse "inf" and "nan" as valid floating point numbers. In our testing, libc++ does this, but libstdc++ and MSVCRT do not.
PiperOrigin-RevId: 455435431
Change-Id: I99c00303b08c2e4e62acfa3ca52a21a3008f4545
Somehow tests load these libs twice, and construct/destruct globals twice.
Can be preproduced even without sanitizers with logging from GlobalModel::~GlobalModel.
themodel does not need to be visible. If it's static each instance will have own a copy.
_mjMap does not need to use string and avoid lifecycle issues at all.
PiperOrigin-RevId: 454678814
Change-Id: I8fca5e1f06cbc2b413beabcece1219445482c0ed
Using fabs with int arguments broke the clang build on GitHub Actions, with a `-Wabsolute-value` error.
PiperOrigin-RevId: 453872474
Change-Id: I5c1278e9e628b0ea3d6f26c7ee7d6925c6276850
I am running through simulate.swift (my port of simulate.cc) with
address sanitizer to discover related bugs.
Besides ones in my port, there are two in MuJoCo:
1. in maketext, the logic to find . is not protected against j is less
than 0 (due to the decreasing logic above), creating out of bound
access.
2. in mj_printFormattedData, qfrc_applied should use length nv not nq,
otherwise out of bound access could be triggered.
Test Plan:
Run through the simulate.cc with asan. Before this fix, when presenting
profiler view, it will trigger bug #1. When print data, it will trigger
bug #2. Both are using model/humanoid/22_humanoids.xml.
- Applies to [orientation specifiers](https://mujoco.readthedocs.io/en/latest/modeling.html#frame-orientations) in `body`, `inertial`, `geom`, `site`, `camera`.
- Before this change the check was done only for multiple *alternative* specifiers, but not for `quat` and an alternative specifier.
- Moved the check from the compiler to the parser.
- Added tests.
PiperOrigin-RevId: 453201981
Change-Id: I20907361f211dae904e734cd083e9df0efe4f654
The introduction of additional mjData arrays for the implicit integrator caused in a shift in data alignment in memory. In certain combinations of machines and models, this change resulted in ~5% performance regression that appears to be attributable to L2 cache misses.
This change aligns all arrays in mjData to the boundary of a typical L2 cache line size. Benchmarking shows that this resolves the performance regression on a Broadwell machine.
PiperOrigin-RevId: 451652783
Change-Id: I4ed2c2d03453a83a0b4a9b8e224e2fd7c0dadea9
Inertias of moving bodies are enforced to be greater mjMINVAL at model compilation, there is no need to enforce inertias at mesh load time. Tiny meshes could be static or part of a larger geom assembly, in which case their tiny inertia is not a problem.
Also, make error message for small masses/inertias more precise.
PiperOrigin-RevId: 451625216
Change-Id: I4b48d7a881e620d6fbb591ec77fbdacc7c03496c
This allows us to detect reads and writes that are logically out of bounds.
PiperOrigin-RevId: 451520772
Change-Id: I614a8062c404d97148166bcd37db399420956d8b
Indiscriminate memset into d->buffer and m->buffer previously caused msan to not detect uninitialized reads.
Also fix tests with uninitialized read bugs that are detected by msan after this change.
PiperOrigin-RevId: 451508224
Change-Id: I1f4b080a8ef765c34ba7a0adc2c686419f6e5516
Before this change, xipos[0,1,2] was only set to 0 by mj_resetData zeroing the entire buffer.
When using mj_resetDataDebug, the value would be wrong.
PiperOrigin-RevId: 450939814
Change-Id: I887dd977161c7d132a80b766db7be23350ffb50b
Fixes a bug introduced in CL/432455486, which stopped saving explicit inertias if they were inferred from geoms. But if we don't save inertias then we need to save mass/density specifiers.
BEGIN_PUBLIC
xml_native_writer saves geom mass or density, if specified.
END_PUBLIC
PiperOrigin-RevId: 450667775
Change-Id: Ibb5b269b58f1143013091729642bb501243ed943
Added analytic derivatives of smooth (unconstrained) dynamics forces, with respect to velocities:
- Centripetal and Coriolis forces computed by the Recursive Newton-Euler algorithm.
- Damping and fluid-drag passive forces.
- Actuation forces.
A new implicit-in-velocity integrator is implemented using the analytic derivatives. This integrator lies between the Euler and Runge Kutta integrators in terms of both stability and computational cost.
PiperOrigin-RevId: 450377010
Change-Id: Ie192b441876c22e732fb749333926f296e0a09cc
Nimrod Gileadi