New/fixing derivs benchmarks - #110

Open

DVolpi256 wants to merge 192 commits into

mainfrom

new/fixing_derivs_benchmarks

Commits on Oct 23, 2025

Forward pass on first order derivatives
DVolpi256
committed
698d3d0

Commits on Oct 28, 2025

Added a backward pass and changed derivative matrix dimensions.
DVolpi256
committed
9bc3af0

Commits on Oct 31, 2025

Added swapped force product to first order derivative function and added some efficiencies after review.
DVolpi256
committed
b594f40

Commits on Nov 3, 2025

Tested firstOrderInverseDynamicsDeriviatives with additional lines in two unit test files and fixed the
DVolpi256
committed
64162ef

Commits on Nov 20, 2025

Added S_ring terms
DVolpi256
committed
7158c94

Commits on Dec 5, 2025

Commits on Dec 12, 2025

Debugged the rest of the ID derivatives function, and now the test passes.
DVolpi256
committed
49cd0e1

Commits on Dec 13, 2025

Added std::complex<double> support from the complex step branch
DVolpi256
committed
b87f990

Commits on Dec 16, 2025

Commits on Dec 17, 2025

Made separate tests for the inverse dynamics derivatives. One does a finite difference and the other does a complex step. They work well for the double pendulum case. The three link and four link s…
DVolpi256
committed
53901fd

Commits on Dec 18, 2025

Fixed an issue pertaining to gravity propagation. The inverse dynamics derivatives tests now pass.
DVolpi256
committed
4bf2ca1
Condensed the ID derivatives forward pass.
DVolpi256
committed
8416cfb
Added the Mini Cheetah to the finite difference code.
DVolpi256
committed
3555b8e
streamlined unit test
pwensing
committed
163b6d2
Merge branch 'anon_func' into first_order_derivatives
pwensing
committed
996775e
prototype code
pwensing
committed
c104c3f
prototype code
pwensing
committed
4767fbf

Commits on Dec 19, 2025

Commits on Dec 20, 2025

Added comments to ClusterTreeDynamics.cpp documenting necessary gradient terms
DVolpi256
committed
e1bf655

Commits on Dec 26, 2025

Commits on Dec 28, 2025

Needed to address how the complex model of the mini cheetah was building to get the complex step test of the Mini Cheetah to work. The Mini Cheetah should now successfully pass validation for the I…
DVolpi256
committed
97fb0ea

Commits on Dec 31, 2025

Made the first steps to add in derivative terms. Joints that have zero derivative terms still function, and there is architecture in place for non-zero terms.
DVolpi256
committed
7039079

Commits on Jan 1, 2026

Created a revolute pair absolute joint to test a non-zero derivative terms case. It appears to work well.
DVolpi256
committed
630e19f

Commits on Jan 2, 2026

Made major progress in getting the MIT Humanoid to pass the inverse dynamics derivatives tests. There was a sign error for rotation matrices that was a large source of error.
DVolpi256
committed
eca71ee

Commits on Jan 4, 2026

Implemented derivative term support for all but the generic joint. Will need to examine how accurate the unit tests can be made.
DVolpi256
committed
ab111ec

Commits on Jan 7, 2026

Commits on Jan 8, 2026

Commits on Jan 9, 2026

Commits on Jan 13, 2026

Fixed memory issues that now allow the Four Bar, Tello, and TelloWithArms robots to be verified according to the ImplicitConsntrain tests
DVolpi256
committed
1f6b4f7

Commits on Jan 14, 2026

Commits on Jan 16, 2026

Added benchmarks for firsOrderIDDerivatives() and made adjustments for speed.
DVolpi256
committed
0e2ae7e

Commits on Jan 17, 2026

Made various improvements to increase performance. Will need to do another pass on the CRBA refactoring.
DVolpi256
committed
8f56cc8

Commits on Jan 18, 2026

Added batched operations for additional speed gains and examined the possibility of precomputing terms in a shared frame. The additional overhead appears to be too costly to implement.
DVolpi256
committed
67c7843

Commits on Mar 19, 2026

Added robots and adjusted benchmark figure generation for paper, poster, and presentation
DVolpi256
committed
bdaf9f2

Commits on Mar 24, 2026

Added support for Tello complex step test
DVolpi256
committed
a7ae4a9
Finalized complex step test for closed loop constraints
DVolpi256
committed
c6bc8df

Commits on Apr 21, 2026

Improved getSdotqd_q
pwensing
committed
adc6310
make generic
pwensing
committed
6737d4f

Commits on Apr 22, 2026

Adding back in finite diff comparisson code
pwensing
committed
c85dcb2
Finite diff verification for getSdotqd_q
pwensing
committed
2702f5e
clean up
pwensing
committed
da1e0f4
whoops -- more cleanup
pwensing
committed
f48fda2
Changed how getSq and is handled using jacobian_S_times_b and jacobian_ST_time_F functions. Also added infrastructure for Cassie and Kangaroo
DVolpi256
committed
a36c741
fix GenericJoint?
pwensing
committed
e18b708
Attempt at a refactor of GenericJoint for S derivatives. Need to implement throughout library
DVolpi256
committed
32c5427

Commits on Apr 23, 2026

Included most recent Kangaroo and Cassie complex step tests.
DVolpi256
committed
1aa6a6a
Should restore genericjoint functionality to before yesterday's edits for the cleanest complex step tests.
DVolpi256
committed
6a362ec
small cleanup in lie group addition routine
pwensing
committed
7056102
Setting up ability to do complex step for states off the constriant manifold. (separate problem into simpler parts)
pwensing
committed
48b011f
Initial q perturbation switched to G method
pwensing
committed
372311b
Updated complex step second instance
pwensing
committed
8acac7c
tello derivatives now allow finite diff steps off the constraint manifold and initial states off the constraint manifold.
pwensing
committed
1e3064e
Streamline
pwensing
committed
632d1ae
Remove Unused
pwensing
committed
3e045e8
Streamline code
pwensing
committed
62f32c1
Cleanup
pwensing
committed
45fbcae
Errors persist for MIT humanoid, but builds. PlanarLegLinkage derivatives likely out due to same issue.
pwensing
committed
a40b29e
streamline
pwensing
committed
08c66ec
use joint constraint solving
pwensing
committed
6ba907d
helper for randomizing model state
pwensing
committed
2d3903e
many tests restored
pwensing
committed
cad4c8a
Remove direct tests
pwensing
committed
73c97ff
cleanup
pwensing
committed
b45833d
cleanup
pwensing
committed
f88a351
refactored out simple tests
pwensing
committed
bc284da

Commits on Apr 24, 2026

humanoid passes again
pwensing
committed
18faa2d
comment for PlanarLinkage Unit tests
pwensing
committed
c607afc
cleanup
pwensing
committed
f8e70cc
clarifying comment
pwensing
committed
50adb93
Allow random states off manifold -- revisit
pwensing
committed
c06717a
restore testRigidBodyDynamicsAlgos
pwensing
committed
f5aacbc
add native phi to fix tello constraint handling
pwensing
committed
c8c535f
native_phi issue alerts
pwensing
committed
580bb7a
Streamline further for commonality between simple and complex step tests
pwensing
committed
8a2795b
Working to simplify the Simple unit tests
pwensing
committed
265718f
Simplified the simple unit test
pwensing
committed
8755302
Cleaned up unit tests
pwensing
committed
5d7afa4
Oh boy. Four bars fixed.
pwensing
committed
f703f58
Fail
pwensing
committed
abf33d6
Fix fail -- make sure MITHumanoid samples a feasible state (now that we are relying on the GenericImplicit base for the revolutePair with rotors)
pwensing
committed
2b1a751
ComplexStepTests
pwensing
committed
a889491

Commits on Apr 26, 2026

This is the 80% complete code for switching getSq() to jacobian AD code.
DVolpi256
committed
ab5fca9

Commits on Apr 27, 2026

Commits on Apr 28, 2026

Commits on Apr 29, 2026

Updated the loop depth test to test the proper model shape.
DVolpi256
committed
f3bf98d
Fixed MIT Humanoid (-R) timing problem
DVolpi256
committed
d6d89bd

Commits on Apr 30, 2026

Commits on May 2, 2026

Updated benchmarking files to make newest figures. Also expanded the parallel chain depth test to include more data points for the ten link case. To orun the test, run benchmarkParallelChainDepth. …
DVolpi256
committed
e66ae09

Commits on May 5, 2026

Fix MIT Humanoid
pwensing
committed
aa2e19a
fixed testClusterJoints for compatibility with new nq nv with revolutePairWithRotor
pwensing
committed
10d694a
fix algosDerivatives test
pwensing
committed
6209de1
WIP -- attempt fix of forwardKinematics test
pwensing
committed
0846b44
Partial fix for reflected inertia algorithms -- something is still off for revolute pair with rotor
pwensing
committed
17ef582
Disable RevolutePairChainWithRotor in reflectedinertia tests for now.
pwensing
committed
37ec573
Fix forward kinematics test
pwensing
committed
7c0e1a4
Bye bye mr kangaroo
pwensing
committed
6ae0637
Missed one kangaroo test -- removed
pwensing
committed
22814ea
Remove benchmarkIDDerivativesAccuracy
pwensing
committed
3c46077
relax cassie tolerance for non-deterministic unit test
pwensing
committed
9a57649
deterministic seed for unit tests so tests can catch true regressions, not just bad dice rolls
pwensing
committed
69411ee
Fix tello constraint functions (errant integer division)
pwensing
committed
b1b9779
Remove debug for native_phi
pwensing
committed
706e1d6
Restore kinematics check vs. Revolute Pair Cluster
pwensing
committed
beb4e61
Patch fix for reflected inertia tests
pwensing
committed
6b2fdbd
Restore Revolute Pair Chain Reflected inertia test
pwensing
committed
0955e32
Updated the Cassie model to utilize the proper inertias from its MJCF and mirror components correctly across legs.
DVolpi256
committed
c314926
Fix silent undefined behavior in makeRPWRBodies/makeRPWRJoints sort
pwensing
committed
d81d585

Commits on May 6, 2026

streamline generic joint (remove unused functions)
pwensing
committed
1018277
candidate jit options
pwensing
committed
5a0e15b
Updated Cassie to match the library's handling of four bar mechanisms
DVolpi256
committed
f43543c

Commits on May 7, 2026

Commits on May 20, 2026

Commits on May 21, 2026

Commits on May 22, 2026

Remove benchmarking divergences from main (profiling instrumentation, benchmark files)
DVolpi256
committed
7c755bb

Commits on May 26, 2026

Removed more dead references to getSq() and caching behavior
DVolpi256
committed
22647b5
Removed dead references to Sdotqd_qd_
DVolpi256
committed
0fe636d
Update GenericJoint.h
DVolpi256
authored
c2b2a01
q_cache_ is now q_spanning_ to prevent confusion with caching behavior
DVolpi256
committed
6a4b44f
Rewrite comments on quaternion normalization
DVolpi256
authored
d34ecc3
Removed dead references to transformBlockDiagonalInertiaToWorld
DVolpi256
committed
7911966
Fix comment about return dimensions in getSdotqd_q method
DVolpi256
authored
0e34277
Fixed comment about tolerancing
DVolpi256
authored
86f6b22
Reworded comments for clarity in GenericJoint.cpp
DVolpi256
authored
2c36c2d
Took out outdated debugging references
DVolpi256
committed
af3a93d
Reword comments for clarity on complex conjugation
DVolpi256
authored
819f7b6
Clean up inverse dynamics derivative tests
DVolpi256
authored
e097d55

Commits on May 27, 2026

Commits on Jun 1, 2026

Commits on Jun 2, 2026

Took out KUKA LWR header in favor of cloning the URDF double model to std::complex
DVolpi256
committed
70c8e2a

Commits on Jun 3, 2026

Commits on Jun 10, 2026

Commits on Jun 15, 2026

Took out remaining TeleopArm tests until it is fixed
DVolpi256
committed
0a79315
Add benchmark executables from fixing_derivs branch

DVolpi256
and
claude
committed
f6cc254
Remove benchmarkIDDerivativesBreakdown — requires profiling API

DVolpi256
and
claude
committed
963485d
Switch CasADi JIT compiler from clang to gcc

DVolpi256
and
claude
committed
bf90812
Fix benchmark compilation errors from removed profiling API and RevoluteTriple

DVolpi256
and
claude
committed
abc7791
Fix remaining t6 reference in benchmarkComplexJointChains per-cluster output

DVolpi256
and
claude
committed
7752572
Disable TeleopArm in benchmarkIDDerivativesComparison

DVolpi256
and
claude
committed
82a861a
Restore profiling API and re-add benchmarkIDDerivativesBreakdown

DVolpi256
and
claude
committed
466798c

Commits on Jun 16, 2026

Re-enable RevoluteTriple and TeleopArm benchmark sections

DVolpi256
and
claude
committed
0ee8618
Updated RevoluteTriple to Generic here too
DVolpi256
committed
6e9326a

Commits on Jun 18, 2026