TRUSTWORTHY AD HOC TEAMWORK
Abstract
The past few years have witnessed a profound transformation in intelligent technology in our daily lives. This technology is now a ubiquitous reality that provides us with personal assistants deployed on several platforms. For instance, the consumer market has several high-tech products that are now common in our lives, such as small devices (e.g., smartphones, smartwatches, and smart speakers) and domestic robots (e.g., Roomba). People will inhabit intelligent physical environments endowed with ambient intelligence and populated with intelligent devices and robots soon. As the number of robots continues to increase, we expect to see robots interacting more and more with various other robots and humans. In many of these interactions, the robots may share the same goal and cooperate to perform a task. However, many robots may not have coordination and standard communication protocols because they have been designed by different developers and at other times. Ad hoc teamwork is a research topic that aims to address the problem mentioned above. This research community focuses on building learning agents, such as softbots or robots, that engage in cooperative tasks with other unknown agents without relying on predefined coordination strategies and communication protocols. However, many ad hoc teamwork research works use simple scenarios that are not close to practical deployment situations. In particular, none of these research works are tailored for human-robot interactions due to several limitations. For example, many works rely on the following assumptions: (i) the robot knows its role and the target task in advance, (ii) the robot can observe the teammates actions, (iii) the robot can fully observe the states, and (iv) the environment provides a reward signal to the robot. These assumptions, however, are rarely valid in real-world scenarios with humans and robots. For instance, human-robot interaction settings may have (i) humans that cannot communicate the intended task to the robot, (ii) robots that are not capable of observing the teammates actions, (iii) robots that cannot fully observe the states, and (iv) interactions that do not generate an implicit/explicit reward signal. In addition, trust is an essential element for effective cooperation between humans and robots. In this work, we focus on the development of ad hoc teamwork algorithms in order to facilitate the trust development by humans towards the unknown robotic teammates. However, to the best of our knowledge, no research work has analyzed human-robot trust within ad hocv teamwork scenarios or created tailored algorithms for trustworthy ad hoc teamwork. We characterize this type of collaborative interactions as trustworthy if the human teammate is able to establish trust towards the robot, i.e., relies on the robot s actions. This project addresses these key issues by exploring cutting-edge research on trustworthy ad hoc teamwork in human-robot collaborations. We thus believe that trustworthy ad hoc teamwork is one of the most promising applications for this research topic; however, it largely remains to be explored. We envisage an exciting research work that explores the challenges of human-robot trust within ad hoc teamwork scenarios.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 20, 2023
- Source ID
- FA95502210475
Entities
People
- Jose Sardinha
Organizations
- Air Force Office of Scientific Research
- United States Air Force