Student Name: Cayetana Salinas Rodriguez

Advisor: Dr. Sarah Li

Milestone: MS Thesis Final Examination (Defense)

Degree Program: Aerospace Engineering

Title: Systematic Integration of Estimation and Control in Stackelberg Games

Abstract: Autonomous systems such as self-driving cars, delivery robots, or humanoids replacing household workers are increasingly deployed in shared environments, making their ability to interact effectively with others critical to overall performance. In that regard, dynamic game theory is a powerful tool to model these interactive scenarios; in particular, we study the two-player Stackelberg game. Existing solutions to the Stackelberg game assume that the leader has knowledge of the follower’s intent. However, when the nature of the interaction is unknown, the leader must be capable of learning the follower’s best response (BR) and optimizing its strategy simultaneously. Active research areas such as intent estimation and inverse learning implicitly assume that knowing the correct BR leads to better interaction outcomes. However, we re-examine this question: Can we decouple estimation and control in Stackelberg games? To capture the follower’s intent uncertainty, we introduce an update in the leader’s belief at some point in the game horizon. Under this setting, we prove that in general, assuming an incorrect follower’s BR can lead to more optimal leader costs over the entire game than knowing the true follower’s BR. In the case of open loop Stackelberg equilibria (OLSE) this can be traced to the time-inconsistent nature of the equilibrium. Thus, we introduce a novel “commitment-breaking” solution approach for these kinds of games that is time-consistent at the update time. This approach exploits the prior knowledge of the update to guarantee global optimality of the true BR belief and can be used to characterize the sub-optimality of the true belief OLSE solution. Further, we demonstrate that this “commitment-breaking” framework can be applied to a number of existing game-theoretic formulations to model the strategic nature of breaking pre-commitment in adversarial and uncertain environments. Finally, we demonstrate the theoretical findings of this thesis in various dynamic simulations including a collision-avoidance example. 

Date and time: 2026-04-16, 10:00am

Location: Price Gilbert 4222

Committee:
Dr. Sarah Li (advisor), School of Aerospace Engineering
Dr. Sarah H.Q. Li, School of Aerospace Engineering
Dr. Jonathan Rogers, School of Aerospace Engineering
Dr. Samuel Coogan, School of Electrical and Computer Engineering
Dr. Chih-Yuan Chiu, School of Electrical and Computer Engineering
,