Introduction to the special issue on the science behind embodied AI : The robots of the AAAI competition and exhibition

Abstract

Welcome to the special issue of Autonomous Robots devoted to the topic of The Science Behind Embodied AI : The Robots of the AAAI Competition and Exhibition. The robotics competition held at the annual conference of the American Association for Artificial Intelligence (AAAI) was first held in 1993 as a way to help promote the science of embodied AI as illustrated by the use of robotics. The competition has evolved over the years and has been a driving force to encourage researchers to push the state of the art in deployed AI. For a more detailed history of the competition, please see Balch and Yanco [1]. One of the greatest challenges that any robotics researcher faces is the disparity that occurs between developing and testing an algorithm in the relative safely of an software-only simulation system and then attempting to make that same algorithm work well on a deployed robotic system. Immediately when deployed on a robot, the researchers must contend with problems such as errors in sensing, inaccurate actuation, and unanticipated interactions between the robot and the environment. A common complaint by many who attempt such a challenging venture is something along the lines of “Well, it worked fine in the lab...” Even in the AAAI robot exhibition, where the robots on display do not have to compete but rather demonstrate an interesting aspect of embodied AI that is not part of the competition itself, the robots must operate in the environment provided to them at the venue. There are many lessons learned by the researchers who attend these events, not the least of which is the important balance and intermingling of both scientific and engineering efforts. The AAAI competition consists of two distinct parts. The first is the series of competitions themselves which are structured events with pre-defined rules and scoring mechanisms where teams compete against each other for the best performance in their particular categories. The second is the exhibition which is an open-ended venue that is set up to allow researchers to demonstrate interesting scientific endeavors to the AI community at large. In both categories of events, the robots are expected to operate in an environment that is largely unknown until the researchers arrive. These events serve both as interesting research and pedagogical experiences for faculty and students. Many students have had their first experiences with deploying a fully functional mobile robot at one of these events and in many cases, this leaves a lasting impression that stays until those students are faculty researchers in their own right and are encouraging their own students to participate. The efforts described within this issue provides a window into the kinds of research, both in terms of the science as well as the engineering, that is required to create embodied AI and deploy it in a physical environment. The first four papers describe a number of different research efforts that have participated in the various competitions that have occurred since the creation of the competition. David Miller et al.’s article, “Scarecrow: If I only had AI”, starts out the issue as a historical perspective on the very early days of the competition. This article describes a very low-cost specialized robotic entry in the competition which stood in contrast with the much more expensive general-purpose research robots that were entered by the other teams. This article serves as useful reminder that the research community must