Timezone: »

Of Moments and Matching: A Game-Theoretic Framework for Closing the Imitation Gap
Gokul Swamy · Sanjiban Choudhury · James Bagnell · Steven Wu

We provide a unifying view of a large family of previous imitation learning algorithms through the lens of moment matching. At its core, our classification scheme is based on whether the learner attempts to match (1) reward or (2) action-value moments of the expert's behavior, with each option leading to differing algorithmic approaches. By considering adversarially chosen divergences between learner and expert behavior, we are able to derive bounds on policy performance that apply for all algorithms in each of these classes, the first to our knowledge. We also introduce the notion of moment recoverability, implicit in many previous analyses of imitation learning, which allows us to cleanly delineate how well each algorithmic family is able to mitigate compounding errors. We derive three novel algorithm templates (AdVIL, AdRIL, and DAeQuIL) with strong guarantees, simple implementation, and competitive empirical performance.

Author Information

Gokul Swamy (Carnegie Mellon University)
Sanjiban Choudhury (Aurora)
James Bagnell (Aurora Innovation)

Drew has worked for two decades at the intersection of machine learning and robotics both as a faculty member at Carnegie Mellon University and in engagements with industry from self-driving haul trucks to perception architecture for Uber’s self-driving cars and in his current role as CTO of Aurora Innovation.

Steven Wu (Carnegie Mellon University)

More from the Same Authors