This paper examines the approach taken by team gitastrophe in the CG:SHOP 2021 challenge. The challenge was to find a sequence of simultaneous moves of square robots between two given configurations that minimized either total distance travelled or makespan (total time). Our winning approach has two main components: an initialization phase that finds a good initial solution, and a k-opt local search phase which optimizes this solution. This led to a first place finish in the distance category and a third place finish in the makespan category.

Our work on minimum convex decompositions is based on two key components: (1) different strategies for computing initial decompositions, partly adapted to the characteristics of the input data, and (2) local optimizations for reducing the number of convex faces of a decomposition. We discuss our main heuristics and show how they helped to reduce the face count.