# Fast Decision Support for Air Traffic Management at Urban Air Mobility Vertiports using Graph Learning

**Authors:** Prajit KrisshnaKumar, Jhoel Witter, Steve Paul, Hanvit Cho, Karthik Dantu, Souma Chowdhury  
**arXiv:** [2308.09075](https://arxiv.org/abs/2308.09075)  
**Date:** 2023-08-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2308.09075v1/

## Key points

- Urban Air Mobility (UAM) aims to revolutionize travel in urban and suburban areas
- UAM aircraft operate from small airports called vertiports
- Vertiports consist of multiple take-off/landing spots and battery-recharging stations
- Traditional air-traffic controllers face challenges in managing the schedule of vertiports due to high volume of aircraft movements
- A novel approach called Urban Air Mobility - Vertiport Schedule Management (UAM-VSM) is proposed
- UAM-VSM leverages graph reinforcement learning for efficient management of vertiport schedules
- Graph convolutional network (GCN) is employed to extract relevant features from graphs representing physical spots and vehicles
- Perceptron layers determine appropriate actions for each vehicle based on extracted features
- Performance evaluation metrics include delays, safety, and battery consumption
- Realistic simulations using AirSim validate the effectiveness of graph reinforcement learning for UAM-VSM problem solving
- This approach outperforms basic reinforcement learning with graph embeddings or random choice baselines

## Summary

Urban Air Mobility (UAM) is an emerging concept that aims to revolutionize travel in urban and suburban areas by providing decongested, safe, and fast transportation options. UAM aircraft operate from small airports called vertiports which consist of multiple take-off/landing spots and battery-recharging stations. To address the challenge of managing the schedule of these vertiports in real-time for traditional air-traffic controllers due to the high volume of aircraft movements, a novel approach called Urban Air Mobility - Vertiport Schedule Management (UAM-VSM) is proposed in this paper. The UAM-VSM leverages graph reinforcement learning to generate decision-support policies for efficient management of vertiport schedules. In this approach, the designated physical spots within the vertiport's airspace and the vehicles being managed are represented as separate graphs. To extract relevant features from these graphs, a graph convolutional network (GCN) is employed. The extracted features are then passed through perceptron layers to determine appropriate actions for each vehicle such as hovering or cruising, idling or taking off or landing on an allocated vertiport spot. The performance of the system is evaluated based on metrics like delays, safety (number of collisions), and battery consumption. Realistic simulations using AirSim with scaled-down multi-rotor vehicles were conducted to validate the effectiveness of using graph reinforcement learning for solving the UAM-VSM problem. The results demonstrate that this approach outperforms basic reinforcement learning with graph embeddings or random choice baselines. Overall, this research presents a promising solution for efficiently managing vertiport schedules in urban air mobility systems using graph reinforcement learning which automates decision-making processes enabling safer and more efficient operations in dense urban areas while reducing congestion and improving travel times.