What solvers is he using? Is it really just some sort of simulated annealing? I would have expected A* search at a first glance.