# Towards More Efficient Shared Autonomous Mobility: A Learning-Based Fleet Repositioning Approach

arXiv: https://arxiv.org/abs/2210.08659 · HTML: https://mlsanity.com/paper/2210.08659/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 9 of its 37 references that are in the map.
- First posted on arXiv 16 Oct 2022.

## Builds on (most cited first, 9 of 9)

- [Asynchronous Methods for Deep Reinforcement Learning](https://mlsanity.com/paper/1602.01783/) (2016)
- [Reinforcement Learning for Ridesharing: An Extended Survey](https://mlsanity.com/paper/2105.01099/) (2021)
- [Deep Reinforcement Learning for Multi-Driver Vehicle Dispatching and Repositioning Problem](https://mlsanity.com/paper/1911.11260/) (2019)
- [Graph Neural Network Reinforcement Learning for Autonomous Mobility-on-Demand Systems](https://mlsanity.com/paper/2104.11434/) (2021)
- [MOVI: A Model-Free Approach to Dynamic Fleet Management](https://mlsanity.com/paper/1804.04758/) (2018)
- [A Cooperative Multi-Agent Reinforcement Learning Framework for Resource Balancing in Complex Logistics Network](https://mlsanity.com/paper/1903.00714/) (2019)
- [Graph Meta-Reinforcement Learning for Transferable Autonomous Mobility-on-Demand](https://mlsanity.com/paper/2202.07147/) (2022)
- [Scalable Deep Reinforcement Learning for Ride-Hailing](https://mlsanity.com/paper/2009.14679/) (2020)
- [A Distributed Model-Free Algorithm for Multi-hop Ride-sharing using Deep Reinforcement Learning](https://mlsanity.com/paper/1910.14002/) (2019)

## Abstract (arXiv)

Shared-use autonomous mobility services (SAMS) present new opportunities for improving accessible and demand-responsive mobility. A fundamental challenge that SAMS face is appropriate positioning of idle fleet vehicles to meet future demand - a problem that strongly impacts service quality and efficiency. This paper formulates SAMS fleet repositioning as a Markov Decision Process and presents a reinforcement learning-based repositioning (RLR) approach called integrated system-agent repositioning (ISR). The ISR learns a scalable fleet repositioning strategy in an integrated manner: learning to respond to evolving demand patterns without explicit demand forecasting and to cooperate with optimization-based passenger-to-vehicle assignment. Numerical experiments are conducted using New York City taxi data and an agent-based simulation tool. The ISR is compared to an alternative RLR approach named externally guided repositioning (EGR) and a benchmark joint optimization (JO) for passenger-to-vehicle assignment and repositioning. The results demonstrate the RLR approaches' substantial reductions in passenger wait times, over 50%, relative to the JO approach. The ISR's ability to bypass demand forecasting is also demonstrated as it maintains comparable performance to EGR in terms of average metrics. The results also demonstrate the model's transferability to evolving conditions, including unseen demand patterns, extended operational periods, and changes in the assignment strategy.