Featured Research
Led by VRAC faculty Dr. Cody Fleming, Urban Air Mobility (UAM) envisions integrating the skyscape into the transportation network and encompasses services such as delivery drones, on-demand shared mobility by Vertical-Take Off and Landing aircraft for intra-city passenger trips. Unfortunately, research shows that the design of transportation systems has a long-lasting, often discriminatory effect that reinforces existing socio-economic inequality. Technology developed by VRAC researcher Cody Fleming helps overcome these issues.
With a novel, emerging paradigm like UAM, it is unclear how the air traffic should be designed and managed. In addition, there are questions regarding how many aircraft can be in the sky at any given time, how many aircraft can depart from or arrive at a given location, and how flights are sequenced. Such decisions have a profound impact not only on the UAM technology but also the community. The number of aircraft, where they fly, and how they fly have consequences on the privacy and fairness of passengers and/or receivers of goods in delivery systems, but also the general population below. Moreover, economic viability will require UAM systems to operate at high volumes, thereby imposing severe requirements in terms of air-traffic control. Navigating such a large and complex design space calls for novel design and optimization algorithms.
To address these challenges, we are creating machine learning techniques leveraging Graph Neural Networks and Reinforcement Learning to determine the optimal allocation of vertiports and the associated airspace design. Such vertiport assignment and airspace design recognizes the tradeoffs between the performance metrics and social indices provided by the UAM technology acceptance model derived by our collaborators. Through extensive evaluations, demos, and interactions with stakeholders from the city of Austin, TX, we are refining the technologies to ensure their acceptance by, and benefit for, the community. Our stakeholders include the City of Austin mayor’s office, several regional transportation authorities, a community council, and representative individuals solicited from the community.
We leverage Monte Carlo tree search based on recent success in reinforcement learning and graph-based design, which have found success in other domains. Reinforcement learning famously achieved expert level performance in chess and Go, while graph-rewriting techniques have been used to design complex electronic circuits. To enable stakeholders such as transportation planners to efficiently explore and control the design, we create a notion of graph rewriting for modifying the airspace architecture. Instances of graph rewriting include the addition/subtraction of so-called vertiports (think of airports in the sky for vertical takeoffs and landings), moving a vertiport (thus changing a node’s features), scaling of vertiport capacity, and addition/subtraction of intermediate waypoints, which is depicted below. Part of this research involves creating a taxonomy of graph rewriting rules that is appropriate for general UAM airspace design, in our case study but also across a broad class of urban areas.
The conclusion of this project will result in a Community-in-the-Loop Integrative Framework for Fair and Equitable UAM Infrastructure Design as well as an open-source Computer Aided Planning tool, called VertiCAP.