Future NASA Robotic Science and Exploration Missions are vastly enabled by high-level autonomy software capable of making critical decisions independently of human operators. Current missions are accomplished using a combination of direct human control and preprogrammed automatic responses. This approach has worked well in the past for certain classes of missions, but represents a limit on what we may accomplish in the future due to communications delays, mission complexity, or cost. Future missions need to operate in regimes in which direct human control is either impossible, impractical, or too expensive. Employing autonomy for these missions present formidable problems for mission designers, however, and will require robust implementation of software providing high levels of independent decision making to maintain vehicle health and safety, accomplish complex science and mission goals, and adapt to changing circumstances or opportunities.