meco-group/fatrop

What it solves

Fatrop is a fast and numerically robust solver for constrained nonlinear optimal control problems (OCP). It allows users to find the optimal trajectory and control inputs for a dynamical system, while respecting constraints on states and inputs. It also provides interfaces for general nonlinear programs (NLP), including dense problems (like inverse kinematics) and block-sparse problems (like SLAM).

How it works

Fatrop uses an interior-point algorithm inspired by Ipopt, employing advanced numerical optimization techniques to ensure robustness. To achieve high speed, it leverages a specialized linear solver based on generalized Riccati recursion and integrates BLASFEO for high-performance linear algebra. It handles path equality and inequality constraints without using penalty methods and can incorporate exact Lagrangian Hessian information. It also supports optimization over manifolds (Lie groups), where primal variables live on a manifold and search directions are calculated in the tangent space.

Who it’s for

Robotics researchers and engineers who need to perform trajectory optimization and control for complex dynamical systems, as well as those working on SLAM, bundle adjustment, and inverse kinematics.

Highlights

• High numerical robustness and speed through specialized linear solvers and BLASFEO integration.
• Support for optimization over manifolds (Lie groups) via user-supplied retractions.
• Effective handling of path constraints without penalty methods.
• Multiple interfaces: OCP, Dense NLP, and Graph NLP for different problem scales and small-scale dense problems.
• Integration with CasADi for easier problem definition.