COTONN is a toolchain for converting deterministic correct-by-design controllers into neural networks. This toolchain was developed as part of the bachelors end project of Myla van Wegen, Rob Neelen, Hero Miltenburg and Ward van der Velden.
The tool allows for the loading and formatting of controllers generated with tools such as SCOTS and converting them into neural networks. In its current state the tool is only able to deal with deterministic controllers. Controllers which have multiple inputs per state will be interpreted in such a way as to only deal with the first input for that state. The tool was developed with the controller format in mind as generated by SCOTS and can deal with both look-up tables (called staticControllers in SCOTS) and binary decision diagrams. To use the binary decision diagrams the converter tool included in the toolchain must first be utilized to format the binary decision diagram into a look-up table so that it can be used by COTONN. To use the converter tool a installation of the CUDD library is required (https://github.com/ivmai/cudd).
COTONN is equiped with some standard functions included in the COTONN.py file. A basic user can simply provide the proper arguments to the standard functions and use the tool in that way.
In addition to the functionality to converter controllers to neural networks the toolchain is also equiped with a two simulation environments (MATLAB and C++).
We hope you will find COTONN a useful research/industrial tool. If you have questions or want to report a bug please feel free to email: W.vanderVelden@student.tudelft.nl