For trajectory optimization I’m using Casadi.jl which is a wrapper on python’s casadi library which in turn is a wrapper around C++ library with ipopt as NLP solver. I found it the fastest for my tasks. But JuMP (with ipopt backend)can be used either with limitations I’ve mentioned in the post above. I’m transcribing my problem using collocation pseudospectral methods. For the system like quadcopter with 13 states and several waypoints I’m getting hundreds of decision variables and optimization takes about 1 second. For more complicated fixed wing aircraft with highly nonlinear complex aerodynamics it takes thousands of decision variables and dozens of seconds. In both cases it faster then realtime with realtime factor about 10x.