James-Oswald/Flow-Free-PDDL

Using Computer Vision and Automated Planning for solving the mobile game "Flow Free"

Flow-Free-PDDL

This project looks at using automated planning to solve the mobile game flow-free. We also look at using computer vision (OpenCV) to convert images of flow levels into a json format that we can convert to PDDL problems. The majority of our problems come from json problem encodings in /problem-json from AndraCoteanu/Flow-Free-Solver.

Other Solvers

This solver isn't great at the moment, check out these optimized solvers

• mzucker/flow_solver (Custom heuristic search with fast forwarding)
• Torvaney/flow-solver (SAT Based)
• AndraCoteanu/Flow-Free-Solver (SAT Based)

Extraction of Problems using OpenCV

We use a lightly modified modified version of KananVyas/BoxDetection procedure to segment the game board. We check for the presence of a circle using HoughCircles and grab the circle color from the center pixel.

Problem Viewer

A Small JS app embedded in an HTML file can be found at problems/visualizeProblem.html to display the JS formatted problems.