“…This is done without modifying the NLP solution algorithms themselves nor their convergence properties. We explore task parallelism, i.e., multi-core parallelization, and data parallelism, i.e., single-instruction-multiple-data (SIMD), to efficiently evaluate multiple instances of the robot dynamics and kinematics in a tunnel-following MPC [6], [7] for a robot manipulator, where the underlying OCP is solved by using (i) the firstorder method PANOC with augmented Lagrangian, and (ii) the second-order method SCQP. The effect of parallelization of function evaluations on both methods is also compared.…”