Objectives And Clarabel

COPP solvers optimize convex objectives in addition to satisfying TOPP-style constraints. Clarabel-backed solvers also expose raw solver settings and expert diagnostics.

Include

#include <copp/objective.hpp>
#include <copp/clarabel.hpp>

or:

#include <copp/copp.hpp>

Objective Descriptors

Objective factories live in copp::objective:

std::vector<copp::Objective> objectives{
    copp::objective::Time(1.0),
    copp::objective::ThermalEnergy(0.1, {1.0, 1.0}),
};

The built-in objective support is:

Objective	COPP2-SOCP	COPP3-SOCP
Time	yes	yes
ThermalEnergy	yes	yes
Linear	yes	yes
TotalVariationTorque	yes	yes

For Linear, expected vector lengths depend on solver order:

• COPP2: alpha.size() == s_len, beta.size() == s_len - 1
• COPP3: alpha.size() == s_len, beta.size() == s_len

Torque-related objectives use the robot's inverse-dynamics callback when one is installed. Otherwise they use point-mass dynamics tau = ddq.

Clarabel Options

copp::clarabel::Options combines COPP's acceptance policy with raw Clarabel settings:

copp::clarabel::Options options;
options.allow_almost_solved = true;
options.clarabel_settings.max_iter = 200;
options.clarabel_settings.verbose = false;

copp::clarabel::Settings mirrors the current raw Clarabel setting surface, so advanced users can tune tolerances, iteration limits, linear solver choices, and equilibration behavior without dropping down to Rust.

Expert Results

solve returns only an accepted profile:

auto a = copp::solver::copp2_socp::solve(problem, options);

solve_expert returns diagnostics even when no profile is accepted:

auto expert = copp::solver::copp2_socp::solve_expert(problem, options);
 
if (expert.a) {
    std::cout << "accepted profile length = " << expert.a->size() << "\n";
} else {
    std::cout << "Clarabel status = "
              << static_cast<int>(expert.solver_status)
              << ", iterations = "
              << expert.iterations
              << "\n";
}

Expert results include raw Clarabel vectors, residuals, status, objective value where applicable, per-objective terms, and linear-solver metadata.

Complete COPP2-SOCP Example

std::vector<double> s{0.0, 0.5, 1.0};
std::vector<double> amax{1.0, 1.0, 1.0};
 
copp::Robot robot(1, s.size());
robot.append_s(s);
robot.constraints().add_constraint_1st(amax, 0);
 
namespace copp2 = copp::solver::copp2_socp;
 
copp2::Problem problem{
    robot,
    {copp::objective::Time(1.0)},
    copp::IndexInterval{0, s.size() - 1},
    copp::Boundary2{0.0, 0.0},
};
 
copp::clarabel::Options options;
options.allow_almost_solved = true;
 
auto expert = copp2::solve_expert(problem, options);
if (expert.a) {
    auto time = copp::interpolation::s_to_t_topp2(s, *expert.a);
}

Tutorial Sources

• bindings/cpp/examples/copp2_socp.cpp
• bindings/cpp/examples/copp3_socp.cpp
• bindings/cpp/examples/robot_inverse_dynamics.cpp