# stl-control-imitation

stl-control-imitation is a framework based on Breach Matlab toolbox to teach a neural network to imitate a given controller.

## Interfacing a new case study

### Define params, states, controls

This determines the fundamental caracteristics of the problem.

States have nominal values and initial ranges. They are declared in a structured form as follows:

% state named x
pb.states.x.nominal= 0; 
pb.states.x.range_init = [-1 1];

% state named y
pb.states.y.nominal= 3; 
pb.states.y.range_init = [2 4];

### Define a control function

Control function are expected to have the following form:

[u, cost] = mpc_control(x,opt)

### Define a simulation function

### Define a requirement

### Putting it together in a imitation_pb class

## Initial testing

- test nominal will run a simulation with nominal values for initial states and parameters and stuff.

## Misc

### Result structure

Call to `pb.algo2` returns a `all_results` cell array of structures of the form:

res.nn              % neural network
res.Btrain % Samples used for training of nn
res.Btrain_grid % grid samples covered by Btrain
res.Bcex_traces % (optional) counter examples found for nn