This paper dislills the past six years of experience with rescue robots in the field into a gaps analysis of what is needed for successful ground robots and why. It proposes a taxonomy of ground rescue robots as a function of void and platform size, and then discusses the requirements for ground rescue robots in terms of the general robot system, the operator control unit, and the payloads for the four missions identified to date (structural assessment, search, hazardous material detection, and medical support). It concludes that the biggest gaps that can be most readily addressed using current engineering practices and research studies are the user interfaces, reliability, and support for "plug and play" sensors. The paper is intended to foster both industrial development and applied research in academia.