Risks evaluation and failures diagnosis for autonomous tasks execution in space

We present a control system for autonomous manipulators based on a theory of actions integrated with a theory of perception and failures. The theory of actions, perception and failures is defined in the Situation Calculus, a logical language that allows the representation of dynamic domains. We assume that an au-tonomous agent is provided with a set of possible goals. The goals are suited to the domain and the agent abilities. Despite these abilities, an autonomous agent might fail, during the exe-cution of a task: failures can be caused by any unexpected event like loss of power, vibrations, noise, etc. These unexpected or exogenous events can cause the agent to loose an object is car-rying or to slip or to hit some payload or some other structural component. While it is impossible to avoid a failure it is still possible to minimize the cumulation of failures during the exe