This repository contains a simple header-only implementation of the RTI scheme in a multiple-shooting fashion using the sparse OSQP solver and autodiff for algorithmic differentiation.
The goal is to provide a fast - yet simple - C++ NMPC library. While care has been taken to optimize performance, surely much more could be done.
Everything is still experimental and breaking changes might occur.
It can be used by including it as an INTERFACE library, see examples.
- Eigen3
- OsqpEigen
and their dependencies.
The autodiff header-only library is bundled in the include/ directory.
The user is required to define the implementation of the system dynamics, constraints and cost function
via the two classes vurtis::Model and vurtis::Cost which inherit the additional necessary functionality
from the abstract classes vurtis::ModelBase and vurtis::CostBase respectively.
The numerical integration is performed using 4-th order Runge-Kutta method and all the sensitivities are automatically computed using AD.
It is possible to define step-wise parameters for the constraints.
The provided examples/simple_example.cpp performs trajectory tracking with a kinematic unicycle vehicle
while satisfying input constraints.
- include autodiff as a submodule or system dependency
- some interface to set OSQP settings during setup
- maybe rewrite the OSQP wrapper to avoid unnecessary checks
- interface with qpSWIFT