A World Wide Web for Robots •

H umans can use the Internet to share knowledge and to help each other accomplish complex tasks. Until now, robots have not taken advantage of this opportunity. Sharing knowledge between robots requires methods to effectively encode, exchange, and reuse data. In this article, we present the design and first implementation of a system for sharing knowledge between robots. In the manufacturing and logistics industries, robotic systems have brought significant sociological and economic benefits through improved human safety, increased equipment utilization, reduced maintenance costs, and increased production. In a world that is undergoing significant environmental and social change, there will be an increasing demand for these robots to leave the safety of their controlled environments and operate in the real world. Robots will be required to operate in homes and hospitals to service the health of a rapidly aging population, and they will be required to mine and farm in increasingly remote locations. In these environments, robots will need to reliably perform tasks beyond their explicitly preprogrammed behaviors and quickly adapt to the unstructured and variable nature of tasks. Although there has been much progress in task performance with well-defined sets of objects in structured environments, scaling current algorithms to real-world problems has proven difficult. Today's robots can only perform highly specialized tasks, and their operation is constrained to a narrow set of environments and objects. The majority of the world's 8 million service robots are toys or drive in preprogrammed patterns to clean floors or mow lawns, while most of the 1 million industrial robots repetitively perform preprogrammed behaviors to weld cars, spray paint parts, and pack cartons [1]. To date, the vast majority of academic and industrial efforts have tackled these challenges by focusing on increasing the performance and functionality of isolated robot systems. However, in a trend mirroring the developments of the personal computing (PC) industry [2], recent years have seen first successful examples of augmenting the computational power of individual robot systems with the shared memory of multiple robots. In an industrial context, Kiva Systems successfully uses systematic knowledge sharing among 1,000 individual robots to create a shared world model that allows autonomous navigation and rapid deployment in semistruc-tured environments with high reliability despite economic constraints [3], [4]. Other examples for shared world models include research on multiagent systems, such as RoboCup [5], where sharing sensor information has been shown to increase the success rate of tracking …