A Multi Agent Approach to Vision Based Robot Scavenging

Abstract

This paper proposes a design for our entry into the 2006 AAAI Scavenger Hunt Competition and Robot Exhibition. We will be entering a scalable two agent system consisting of off-the-shelf laptop robots, capable of monocular vision. Each robot will demonstrate the ability to localize itself, recognize a set of objects, and communicate with peer robots to share location and coordinate exploration. Overview Our goal is to produce two fully autonomous robots which will compete as a team in the scavenger hunt and exhibition categories of the 2006 AAAI Robot Competition and Exhibition. In the scavenger hunt, the robots will attempt to look for the target items by bisecting the search space. To facilitate efficient search space coverage, the robots will update each other on their respective positions and findings via ad-hoc wireless connection. In the exhibition, we plan to display some of the advanced features of these robots in more detail. Each robot will operate independently and will contain at least four major subsystems. In order to cooperate and not overlap their searches, each robot will possess a framework for localization and mapping. A path planning component, incorporating avoidance of dynamic obstacles, such as people, will also be necessary to find a viable route through the mapped space. Finally, object recognition algorithms will be necessary for detecting and approaching target objects once the robots are in range. A peer-to-peer networking protocol will be used for communication. These areas, along with a brief description of the platform we will be using, will be discussed in more detail below. Platform Each robot will be based on an Evolution Robotics ER1 kit. The computing platform will be a 2GHz laptop with Windows XP. In addition to the kit-provided USB web cam and short-range infrared sensors, we will investigate