Data-driven Methods for Network-level Coordination of Autonomous Mobility-on-Demand Systems Across Scales - ITSC 2024

Relevant Links

Motivation and Objectives

Recent urbanization trends have resulted in increased travel and related externalities, with cities accounting for over 78% of the world’s energy consumption, and for over 60% of the global greenhouse gas emissions. Congestion levels worldwide are escalating, causing 8.8 billion hours of extra travel time in the USA alone. In this context, relying on private cars for personal mobility is becoming increasingly impractical and unsustainable. Ride-hailing services have emerged as an alternative, offering mobility on- demand (MoD) services, raising concerns related to the exploitation of public resources, equity, profitability, and, importantly, scalability. In particular, the travel demand for such services is spatio-temporally asymmetrically distributed (e.g., commuting toward downtown in the morning), making the overall operations imbalanced and extremely sensitive to disturbances. In this context, technological advances in the field of autonomous driving offer a new mobility paradigm: autonomous mobility-on-demand (AMoD). An AMoD system consists of a fleet of autonomous vehicles (AVs) that pick up passengers, and transport them to their destination. A manager controls the fleet by simultaneously assigning passengers to AVs, routing them, and rebalancing the fleet by relocating customer-free AVs to realign their geographical distribution with transportation demand (thereby solving the aforementioned issue). AMoD services promise two main benefits: they increase the supply of drivers to match increasing demand, and drastically reduce transportation costs. However, such services also entail controlling thousands of AVs in complex, congested networks, raising numerous challenges at the interface of computational scalability of control schemes and system performance.
This tutorial’s contribution is threefold. First, we will cover state-of-the art problems, techniques, and metrics of interest related to AMoD services. By joining the tutorial, researchers will gain a comprehensive perspective into the topic. Second, we will present a hierarchical decision-making framework to centrally control AMoD systems. Specifically, our framework blends optimizationbased methods and data-driven approaches by exploiting the main strengths of graph neural networks (GNNs), reinforcement learning (RL), and classical operations research tools. We will show that our framework exhibits a number of desirable properties, including computational tractability, generalizability, and close-to-optimal performance. Furthermore, we show the ability of the proposed approach to interface with policy evaluation tools of varying fidelity, starting from heuristic-based ones, all the way to microscopic traffic simulators. By joining the tutorial, users will be able to interact with the framework, reproduce results, and smoothly leverage such techniques to solve their problems of interest. Finally, motivated by the need to democratize research efforts in this area, we move the first steps toward the creation of publicly available benchmarks, datasets, and simulators for network level coordination of AMoD systems and will release code to (i) provide openly accessible simulation platforms, and (ii) create a common validation process and allow direct comparison between different methodologies.

Program

The half-day tutorial consists of a list of interactive talks, and an interactive activity. Specifically, we will introduce tutorial by providing a primer of problem definitions and data-driven techniques to solve them. Subsequently, we will provide an interactive “hands-on” session, covering benchmarks,metrics, simulation, and code-bases.

Time	Speaker	Title
13:30-14:00	Zardini	Introduction, logistics
14:00-14:30	Gammelli	Graph RL
14:30-15:00	Schiffer	Multi-Agent RL
15:00-15:30	Schmidt	Offline RL, Robustness, and Scalability
Coffee break
16:00-16:30	Tresca	Benchmarking
16:30-17:30	All	Interactive code-based activity