Introduction

Scott C. Livingston <slivingston@caltech.edu>

btsynth is an implementation of the "backtracking synthesis" algorithm, currently only applied to gridworlds. The dependencies include TuLiP and NumPy.

Some functionality remains undocumented, but a good start is printworld.py for viewing gridworld data files, and gridworld_example.py for running examples. dgridworld_example.py is the deterministic case, i.e., no adversarial environment. E.g., try:

$ tools/printworld.py examples/data/paper4{,_real}.world
$ examples/gridworld_example.py examples/data/paper4{,_real}.world

To generate a random gridworld problem of size 4 by 10, try:

$ tools/printworld.py -r 4 10

The output is a pretty visualization and a world code that can be saved to a plaintext file for later use (see below). The problem thus generated may not be feasible. A simple-minded but correct way to automatically test feasibility is to attempt global synthesis on it by calling gen_navobs_soln (see gridworld_example.py for example usage).

The code includes an extension to the Automaton class defined in TuLiP. Examples of new features are finite memory and switched transitions dependent on the state of this memory.

To test btsynth, from the root directory run:

$ nosetests

More detailed documentation is under doc directory, written in reStructuredText for Sphinx. A snapshot of this documentation will occasionally be posted at http://vehicles.caltech.edu/scott/btsynth/