This paper gifts the design and implementation of an effective place controller for quadrotors in keeping with real-time nonlinear fashion predictive keep watch over (NMPC) with time-delay reimbursement and limits enforcement at the actuators. To take care of the restricted computational assets onboard, an offboard keep watch over structure is proposed. It’s applied the use of the high-performance tool package deal acados, which solves optimum keep watch over issues and implements a real-time iteration (RTI) variant of a sequential quadratic programming (SQP) scheme with Gauss-Newton Hessian approximation. The quadratic subproblems (QP) within the SQP scheme are solved with HPIPM, an interior-point approach solver, constructed on most sensible of the linear algebra library BLASFEO, finely tuned for more than one CPU architectures. We display that via the use of a Hessian condensing set of rules specifically well-suited for a structure-exploiting QP solver, a substantial speed-up may also be completed with appreciate to a state of the art NMPC solver.
The video presentations the experimental validation of our keep watch over structure at the Crazyflie 2.1 nanoquadrotor. We ran two experiments the use of a setup that has an estimated time-delay of 60 ms to be compensated. For every experiment, a reference trajectory is generated on a base pc after which handed to our NMPC ROS node at every sampling fast. When producing the trajectories, we explicitly deal with the feasibility factor within the design procedure, growing two references: one possible and one infeasible. In addressing this factor, we end up via experiments that the functionality of the proposed NMPC isn’t degraded even if the nanoquadrotor makes an attempt to trace an infeasible trajectory, which might, in concept, make it deviate considerably from the supposed trajectory and even crash.
Our ROS stack is freely to be had at: