Latent Space Alignment Using Adversarially Guided Self-Play

Abstract

We envision a world in which robots serve as capable partners in heterogeneous teams composed of other robots or humans. A crucial step towards such a world is enabling robots to learn to use the same representations as their partners; with a shared representation scheme, information may be passed among teammates. We define the problem of learning a fixed partner’s representation scheme as that of latent space alignment and propose metrics for evaluating the quality of alignment. While techniques from prior art in other fields may be applied to the latent space alignment problem, they often require interaction with partners during training time or large amounts of training data. We developed a technique, Adversarially Guided Self-Play (ASP), that trains agents to solve the latent space alignment problem with little training data and no access to their pre-trained partners. Simulation results confirmed that, despite using less training data, agents trained by ASP aligned better with other agents than agents trained by other techniques. Subsequent human-participant studies involving hundreds of Amazon Mechanical Turk workers showed how laypeople understood our machines enough to perform well on team tasks and anticipate their machine partner’s successes or failures.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Additional information

Notes on contributors

Mycal Tucker

Mycal Tucker is a PhD student at MIT working on cognitively-inspired neural network models for human understanding. His research includes methods for designing interpretable neural network architectures, enabling human-understandable emergent communication, and uncovering underlying principles in large neural models.

Yilun Zhou

Yilun Zhou is a PhD student at MIT working on the interpretability and transparency of learned (and especially black-box) models. His research develops algorithms to improve human’s understanding of a model and methods to critically evaluate quantify existing claims of interpretability.

Julie A. Shah

Julie A. Shah is a Professor of Aeronautics and Astronautics at MIT, and directs the Interactive Robotics Group in the Computer Science and Artificial Intelligence Laboratory. Her lab aims to imagine the future of work by combining human cognitive models with artificial intelligence in collaborative machine teammates that enhance human capability.