Robust and opportunistic planning for planetary exploration

Planning for rover operations involves a significant amount of uncertainty. With limited a priori knowledge of the area a rover will explore, it is difficult to predict the affects of actions including their duration and the amount of resources they will consume. In addition, the system may not even know ahead of time all of the goals it will be asked to achieve as new opportunities may be identified during the mission. We are developing the OASIS system to enable rovers to generate and execute high quality mission operations plans and to identify and exploit new science opportunities that may arise during the mission. OASIS combines planning and machine learning techniques to achieve these results. In this paper we discuss how OASIS handles these types of uncertainties and present results from testing the system in simulation and on rover hardware.