Routing algorithms aim to find low-cost routines for vehicles. Local search algorithms are popular choices for quickly finding one single locally optimal solution. There can be many local optimums with different costs in a single problem. Researchers have investigated many methods to escape from local optimums so that local search algorithms are guided to explore other local optimums in hope for finding better solutions. This project will investigate learning-based route re-construction heuristics to support escaping from local optimums due to its\ potential to adjust itself based on the current state and optimization history of the solutions.