Hello, I'm trying to solve an optimisation problem with a Neural Network (similarly to the Reinforcement Learning concept) applied to a UAV:

The NN would have two state inputs (vehicle's velocity and path's shape), and must calculate a control parameter, let's say "L", that is then used in a path following controller in such a way that this controller becomes adaptive to the velocity and shape of the followed path.

The problem is that I don't know the optimal value of "L" for a given velocity and path shape. Thus, I can not give to the NN a pair of inputs-target to train it. Nevertheless, I can implement a functional that for a given velocity, path shape and value of "L" computes a value of performance.

The question is: how can I implement a NN that can be trained using a functional and not using set of inputs-target pair in MATLAB?

Thank you for your attention.

Similar questions and discussions