Files
Mujoco_WASM/test/engine/engine_inverse_test.cc
T
Yuval Tassa cd6db9ebe2 Improve CG solver precision and stability under float32 via line search refactor
Fixes #2313

This CL combines two complementary improvements to the Conjugate Gradient (CG) solver, significantly improving numerical precision, stability, and efficiency, particularly in single precision (float32).

1. Line Search Cost Evaluation Refactor: Previously, solver improvement was calculated by subtracting absolute costs: cost(alpha) - cost(0). In highly converged states or single precision, this is susceptible to catastrophic cancellation. We refactor PrimalSearch to compute the cost delta directly, dramatically improving precision.

2. Improved Solver Termination Condition: Near the float32 precision limit, line search deltas can occasionally be slightly negative due to numerical noise. Previously, any value below m->opt.tolerance (including negative values) triggered termination, halting the solver and locking in destabilizing steps. We update the termination condition to require positive improvement (0 < improvement < m->opt.tolerance), allowing the solver to continue iterating and recover stability.

Together, these changes yield substantial improvements, see reduced tolerances herein.

PiperOrigin-RevId: 924229669
Change-Id: Ic0bbefaed090f3a8b1e79ab8d45422c3e86fb56c
2026-05-31 04:00:09 -07:00

159 lines
4.8 KiB
C++

// Copyright 2023 DeepMind Technologies Limited
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// Tests for engine/engine_inverse.c.
#include "src/engine/engine_inverse.h"
#include <string>
#include <gmock/gmock.h>
#include <gtest/gtest.h>
#include <mujoco/mjmodel.h>
#include <mujoco/mujoco.h>
#include "test/fixture.h"
namespace mujoco {
namespace {
using ::testing::NotNull;
using InverseTest = MujocoTest;
const int kSteps = 70;
static const char* const kModelPath = "testdata/model.xml";
// test standard continuous-time inverse dynamics
TEST_F(InverseTest, ForwardInverseMatch) {
const std::string xml_path = GetTestDataFilePath(kModelPath);
char error[1024];
mjModel* model = mj_loadXML(xml_path.c_str(), nullptr, error, sizeof(error));
ASSERT_THAT(model, NotNull()) << error;
mjData* data = mj_makeData(model);
// set small tolerance and enough iterations for all solvers to converge
model->opt.iterations = 500;
model->opt.tolerance = 0;
// solver names for diagnostics
const char* solver_name[] = {"PGS", "CG", "Newton"};
for (int diagexact = 0; diagexact < 2; diagexact++) {
if (diagexact) {
model->opt.enableflags |= mjENBL_DIAGEXACT;
} else {
model->opt.enableflags &= ~mjENBL_DIAGEXACT;
}
for (mjtSolver solver : {mjSOL_PGS, mjSOL_CG, mjSOL_NEWTON}) {
model->opt.solver = solver;
mj_resetData(model, data);
// simulate, call mj_forward
for (int i = 0; i < kSteps; ++i) {
mj_step(model, data);
}
mj_forward(model, data);
// call built-in testing function
mj_compareFwdInv(model, data);
// per-solver tolerances
mjtNum epsilon;
switch (solver) {
case mjSOL_PGS: epsilon = MjTol(1e-6, 1e-2); break;
case mjSOL_CG: epsilon = MjTol(1e-9, 1e-1); break;
case mjSOL_NEWTON: epsilon = MjTol(1e-10, 1e-2); break;
}
EXPECT_LT(data->solver_fwdinv[0], epsilon)
<< solver_name[solver] << " diagexact=" << diagexact;
EXPECT_LT(data->solver_fwdinv[1], epsilon)
<< solver_name[solver] << " diagexact=" << diagexact;
}
}
mj_deleteData(data);
mj_deleteModel(model);
}
// test discrete-time inverse dynamics
TEST_F(InverseTest, DiscreteInverseMatch) {
// load and allocate
const std::string xml_path = GetTestDataFilePath(kModelPath);
char error[1024];
mjModel* model = mj_loadXML(xml_path.c_str(), nullptr, error, sizeof(error));
ASSERT_THAT(model, NotNull()) << error;
int nv = model->nv;
mjData* data = mj_makeData(model);
int nstate = mj_stateSize(model, mjSTATE_INTEGRATION);
mjtNum* state = (mjtNum*)mju_malloc(nstate * sizeof(mjtNum));
mjtNum* qvel_next = (mjtNum*)mju_malloc(nv * sizeof(mjtNum));
mjtNum* qacc_fd = (mjtNum*)mju_malloc(nv * sizeof(mjtNum));
for (auto integrator : {mjINT_EULER, mjINT_IMPLICIT}) {
model->opt.integrator = integrator;
for (bool invdiscrete : {false, true}) {
// set/unset mjENBL_INVDISCRETE flag (affects both forward and inverse)
if (invdiscrete) {
model->opt.enableflags |= mjENBL_INVDISCRETE;
} else {
model->opt.enableflags &= ~mjENBL_INVDISCRETE;
}
// simulate
mj_resetData(model, data);
for (int i = 0; i < kSteps; ++i) {
mj_step(model, data);
}
// save state
mj_getState(model, data, state, mjSTATE_INTEGRATION);
// call step, save new qvel
mj_step(model, data);
mju_copy(qvel_next, data->qvel, nv);
// reset the state, compute discrete-time (finite-differenced) qacc
mj_setState(model, data, state, mjSTATE_INTEGRATION);
mju_sub(qacc_fd, qvel_next, data->qvel, nv);
mju_scl(qacc_fd, qacc_fd, 1/model->opt.timestep, nv);
// call mj_forward, overwrite qacc with qacc_fd
mj_forward(model, data);
mju_copy(data->qacc, qacc_fd, nv);
// call built-in testing function
mj_compareFwdInv(model, data);
if (invdiscrete) {
mjtNum epsilon = MjTol(1e-9, 0.05);
EXPECT_LT(data->solver_fwdinv[0], epsilon);
EXPECT_LT(data->solver_fwdinv[1], epsilon);
} else {
EXPECT_GT(data->solver_fwdinv[0], 1.0);
EXPECT_GT(data->solver_fwdinv[1], 1.0);
}
}
}
// deallocate
mju_free(qacc_fd);
mju_free(qvel_next);
mju_free(state);
mj_deleteData(data);
mj_deleteModel(model);
}
} // namespace
} // namespace mujoco