# POMO+: Leveraging starting nodes in POMO for solving Capacitated Vehicle Routing Problem

arXiv: https://arxiv.org/abs/2508.08493 · HTML: https://mlsanity.com/paper/2508.08493/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in RL for games and combinatorial optimization, one of 15 directions in Deep reinforcement learning methods.
- Builds on 7 of its 16 references that are in the map.
- First posted on arXiv 11 Aug 2025.

## Placed in

- Deep reinforcement learning methods / [RL for games and combinatorial optimization](https://mlsanity.com/deeprl/rl-for-games-and-combinatorial-optimization/)

## Builds on (most cited first, 7 of 7)

- [Pointer Networks](https://mlsanity.com/paper/1506.03134/) (2015)
- [Attention, Learn to Solve Routing Problems!](https://mlsanity.com/paper/1803.08475/) (2018)
- [Learning to Perform Local Rewriting for Combinatorial Optimization](https://mlsanity.com/paper/1810.00337/) (2018)
- [RL4CO: an Extensive Reinforcement Learning for Combinatorial Optimization Benchmark](https://mlsanity.com/paper/2306.17100/) (2023)
- [Towards Generalizable Neural Solvers for Vehicle Routing Problems via Ensemble with Transferrable Local Policy](https://mlsanity.com/paper/2308.14104/) (2023)
- [A concise guide to existing and emerging vehicle routing problem variants](https://mlsanity.com/paper/1906.06750/) (2019)
- [Leader Reward for POMO-Based Neural Combinatorial Optimization](https://mlsanity.com/paper/2405.13947/) (2024)

## Abstract (arXiv)

In recent years, reinforcement learning (RL) methods have emerged as a promising approach for solving combinatorial problems. Among RL-based models, POMO has demonstrated strong performance on a variety of tasks, including variants of the Vehicle Routing Problem (VRP). However, there is room for improvement for these tasks. In this work, we improved POMO, creating a method (POMO+) that leverages the initial nodes to find a solution in a more informed way. We ran experiments on our new model and observed that our solution converges faster and achieves better results. We validated our models on the CVRPLIB dataset and noticed improvements in problem instances with up to 100 customers. We hope that our research in this project can lead to further advancements in the field.