Methods are needed to improve the capabilities of autonomous robots to perform tasks that are difficult for contemporary robots, and to identify those tasks that robots cannot perform. Additionally, in the realm of remote handling, methods are needed to assess which tasks and/or subtasks are candidates for automation. We are developing a new approach to understanding the capability of autonomous robotic systems. This approach uses formalized methods for determining theachievabilityof tasks for robots, that is, the likelihood that an autonomous robot or telerobot can successfully complete a particular task. Any autonomous system may be represented in achievability space by the volume describing that system’s capabilities within the 3-axis space delineated by perception, cognition, and action. This volume may be thought of as a probability density with achievability decreasing as the distance from the centroid of the volume increases. Similarly, any task may be represented within achievability space. However, as tasks have more finite requirements for perception, cognition, and action, each may be represented as a point (or, more accurately, as a small sphere) within achievability space. Analysis of achievability can serve to identify,a priori, the survivability of robotic systems and the likelihood of mission success; it can be used to plan a mission or portions of a mission; it can be used to modify a mission plan to accommodate unpredicted occurrences; it can also serve to identify needs for modifications to robotic systems or tasks to improve achievability. ©2001 American Institute of Physics.