Safe Agents in Space: Lessons from the Autonomous Sciencecraft Experiment

An Autonomous Science Agent is currently flying onboard the Earth Observing One Spacecraft. This software enables the spacecraft to autonomously detect and respond to science events occurring on the Earth. The package includes software systems that perform science data analysis, deliberative planning, and run-time robust execution. Because of the deployment to a remote spacecraft, this Autonomous Science Agent has stringent constraints of autonomy, reliability, and limited computing resources. We describe these constraints and how they are reflected in our agent architecture.