Dissertation Talk: Learning and Optimization for Mixed Autonomy Systems - A Mobility Context
Seminar | August 20 | 1-2 p.m. | 521 Cory Hall
Cathy Wu, UC Berkeley
In this talk, we discuss machine learning and optimization critical for enabling mixed autonomy systems in the context of mobility. Mixed autonomy characterizes the problems surrounding the gradual and complex integration of automation and AI into existing systems. In the context of mobility, the question is: how will self-driving cars change urban mobility? We first explore and quantify the potential impact of a small fraction of automated vehicles on low-level traffic flow dynamics such as traffic congestion, using novel techniques in model-free deep reinforcement learning. Then, we present generic reinforcement learning techniques for improved variance reduction, developed for large-scale control systems such as traffic networks and other mixed autonomy systems. Next, we explore the sensing challenges and requirements to enable mixed autonomy systems. In particular, the coordination of automated vehicles relies on accurate traffic flow sensing. To this end, a new convex optimization method for cellular network measurements from AT&T for all of California is introduced to address a flow estimation problem previously believed to be intractable. Finally, mixed autonomy systems may be embedded within a larger dynamical system, one which dictates the progression of the integration or use of automation, and may induce substantial positive or negative effects on the system. This motivates the design of (mixed autonomy) systems to mitigate the negative effects. For example, automated vehicles are expected to increase transportation demand through a phenomenon called induced demand. To address this, joint work with Microsoft Research is presented, which provides empirical and theoretical justification for studying ridesharing systems and the optimization methods thereof as a design paradigm for the mixed autonomy mobility system. Together, these contributions demonstrate, through principled learning and optimization methods, that a small number of vehicles and sensors can be harnessed for significant impact on urban mobility.