Fast Retargetable Goals Driven Approach to Deal with Plan Failures of Spacecraft

Plan repair is more preferred over replanning when the agent suffers from plan failures. Most of the existing research work on autonomous plan repair regards the planned action as an instantaneous point, preventing it from being directly applied to the field of spacecraft operation, where is full of concurrent actions with varying duration and resource consumption. In this chapter, a reactive rapid autonomous plan repair algorithm based on retargetable goals, Retargetable Goals Plan Repair Method (ReGPR), is proposed. In ReGPR, a mechanism for transforming a mission plan into a state queue and a method of determining the optimal states, i.e., retargetable goals, based on evaluation are proposed. For transformation, ReGPR discretizes the concurrent actions into two state nodes distributed at their beginnings and ends, which encapsulates much information such as logic, numeric, and duration. Then, the plan repair problem is transformed into the puzzle of state transition by mapping them into the same timeline, which forms the retargetable goals. To determine the optimal goal for recovery, an evaluation criterion including goal reachability and search time estimation is designed. With the help of evaluation, ReGPR gets rid of searching for the recovery plan by the try-error method and finds the possible solution quickly. Several experiments with either logic failures or energy shortage or both were done in the modified Satellite Complex domain to show the performance of ReGPR by comparing it with the corresponding results of the replanning methods and other plan repair methods. And the results desmonstrate that in the repair progress, ReGPR explores fewer state nodes in no more than one round, and its advantage is greater when the plan repair problem becomes more complex.