Thesis Defense: Information, Learning and Incentive Design for Urban Transportation Networks

Committee: Saurabh Amin (supervisor), Asuman E. Ozdaglar, Patrick Jaillet, John Tsitsiklis

Today, data-rich platforms are reshaping the operations of urban transportation networks by providing information and services to a large number of users. How can we model the behavior of human agents in response to services provided by these platforms and develop tools to improve the aggregate outcomes in a socially desirable manner? In this dissertation, we tackle this question from three aspects: 1) analyzing the impact of data-rich information platforms (specially navigation apps) on the strategic behavior and learning processes of travelers in uncertain networks; 2) designing incentive mechanisms to improve system efficiency in the presence of autonomy-enabled carpooling services; 3) developing operational strategies to improve network resiliency under random or adversarial disruptions.

Firstly, we present game-theoretic analysis to evaluate the impact of multiple heterogeneous information platforms (navigation apps) on travelers’ selfish routing decisions, and the resulting network congestion. We compare the value of information provided by multiple platforms (navigation apps) to their users, and capture the key trade-off between gain from information about uncertain network state and congestion externality resulting from other users. Then, we extend the static model to a dynamic setting that addresses the behavior of users who learn and strategically operate in an uncertain environment, while adapting their decisions to the information received from platforms. The resulting stochastic learning dynamics requires analyzing strategic and adaptive (hence endogenous and non i.i.d.) user data. We present new results for convergence and stability of such learning dynamics and develop conditions for convergence to complete information equilibrium.

Secondly, we design a market mechanism that enables efficient carpooling and optimal road pricing in an autonomous transportation market. In this market, the transportation authority sets toll prices on edges, and riders organize carpooled trips using driverless cars and split toll prices. Riders have heterogeneous preferences, with value of each trip depending on the travel time of chosen route and rider-specific parameters that capture their individual value of time and carpool disutilities. We identify sufficient conditions on the network topology and travelers’ preferences under which a market equilibrium exists, and carpooling trips are organized in a socially optimal manner. We also provide an algorithm that computes a set of equilibrium trips, toll prices and payments that achieve the maximum rider utilities.

Finally, we develop machine learning tools to predict travelers’ aggregate behavior response to the uncertain congestion delay in traffic networks. We also study resource allocation problem of defending traffic infrastructure facilities against an adversarial attacker. These results provide insights for mitigating transportation network disruptions and security risks.