A Geometric, Best-Improvement Heuristic for the Travelling Salesman Problem
I spent six months in 2021 building a TSP solver from geometric first principles, with no prior reading of the literature. Recently turned it into an interactive web app with step-by-step animation of every algorithmic move.
Interesting result: the algorithm independently rediscovered convex hull seeding and 2-opt uncrossing. It also produced a farthest-neighbour strategy.
On Berlin52: 7783 vs world optimal 7542 (3.2% above). 14ms on a single CPU core.
Demo: tsp.uncledroid.app
Paper: tsp.uncledroid.app/paper.pdf
Happy to discuss the algorithm — especially if anyone can point me to prior work on farthest-neighbour as a TSP construction heuristic.
video: youtu.be/-gAhARI2ZFI