Introduction T HE new generation of microspacecraft currently under development for Earth orbiting missions will require a high degree of autonomytomeet stringentcost and performancegoals. The microspacecraftmay have to performmultiple, con icting tasks, such as pointing its solar array toward the sun or communicating with a ground station, to ensure it operates within limits and performs a useful function. The traditional, deliberativemethods of arti cial intelligence require complexworld models to reason, whereas neural network and fuzzy logic approaches are somewhat inscrutable and dif cult to validate to ensure spacecraft survival.2 In contrast, recent concepts for arti cial agents borrow heavily from ethology, where the agent responds directly to environmental stimuli.3 In this Note such an arti cial agent approach is proposed that providesa method for action selection that balances the demands of the satellite users (for example, gathering or communicating data, thus draining the spacecraft batteries) and the actions necessary to ensure spacecraft survival (for example, charging the batterieswith a solar array).We have adoptedthe cue-de cit actionselectionalgorithmofMcFarland and Spier because it is directlydevelopedfrom optimal control theory and is computationallysimple to implement on a microspacecraft. The spacecraft is modeled as a nonlinear dynamic system with a state space consisting of key internal variables (battery charge, memory state, temperature, transmitted data bits, and attitude). The state space has a set of limits that de nes the satellite’s useful operating domain.A nite repertoireof behaviors is then used to control the internal dynamics of the spacecraft.A cost function is provided that measures the deviation of the spacecraft from its normal statespace operating point. Application of optimal control theory yields the optimum action selection rules. The action selection rules must maintain a position close to this operating point in the presence of perturbations due to the spacecraft’s own behavior. For example, transmitting data may be necessary for the satellite to perform a useful function but will drain the batteries. The proposed action selection rules are found to display a degree of opportunism:For example,we showduringa simple satellitesimulation that the satellite charges its batteries when the opportunity arises (and not just when the batteries become low on charge). We demonstrate that in the event of major hardware failures the algorithm will resequence the spacecraft actions to ensure survival and to continue to achieve its goals, albeit in a degradedmanner.