Adaptive Conventions for Trustworthy Human-Robot Interaction
Abstract
As robots leave factory floors and start collaborating with humans in different workspaces and missions, we need to study and analyze the consequences of repeated interactions between humans and robots. Humans collaborate well together even in complex tasks by adapting to each other through repeated interactions. What emerges from these repeated interactions is shared knowledge about the interaction history that enables them to trust each other. We intuitively refer to this shared knowledge as convention. Convention formation helps explain why teammates collaborate better than groups of strangers, and why friends develop lingo incomprehensible to outsiders. The emergence of conventions between agents directly influences how much they trust each other. In the context of human-robot interaction, building conventions with humans can enable robots to trust humans, collaborate with them, or even influence and guide their actions. Here, we propose developing a theory for human-robot systems that relies on incorporating adaptive conventions emerging between the agents through repeated interaction. Our goal is to characterize the importance of conventions in collaborative tasks. We will learn a non-stationary and low-dimensional representation that captures continually adapting conventions. We will then formalize how conventions relate to trustworthiness, specifically how it can help build trust with new partners without re-learning the full complexities of the task. Finally, leveraging the learned models of conventions, we will develop collaborative policies for robots that coordinate or even guide humans towards better long-term outcomes while maintaining trust.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 07, 2023
- Source ID
- FA95502110119
Entities
People
- Dorsa Sadigh
Organizations
- Air Force Office of Scientific Research
- Stanford University
- United States Air Force