Machine Learning in Human-Robot Collaboration:

ABSTRACT

While deep learning has achieved tremendous feats in virtual tasks such as competitive game-playing, these advances have not translated to collaborative and physical human-robot teaming. Unfortunately, most human-robot systems are still based on hand-crafted, heuristic policies and even Wizard-of-Oz-based controllers.

The reasons for the lack of portability of deep learning into human-robot interaction (HRI) are as significant as they are numerous, including: the sample complexity of these algorithms versus the finite-duration lifespan and patience of humans; unrealistic assumptions about human behavior (e.g., homogeneity, optimality, etc.); a lack of understanding of situated learning theory; and a focus on uni-modular, virtual interaction.

In this workshop, we aim to bring together researchers to:

• explore and identify ways in which human-robot collaboration can not only reap the benefits of deep learning,

• define for machine learning researchers and roboticists what abilities or milestones need to be accomplished to support human-robot teaming.

The objectives of this workshop are to construct a road map, specifying key milestones and research thrusts that will lead us towards rapid, revolutionary advances in physical human-robot teaming. This workshop will consist of invited speakers, presentations from researchers submitting papers of original research, poster sessions, and a debate on the role of the human in equitable machine learning going forward.