Image-Based Meta-Reinforcement Learning for Autonomous Guidance of an Asteroid Impactor

What is it about?

The terminal guidance of a kinetic impactor towards an asteroid requires high precision navigation and maneuvering capabilities, especially when a binary system is considered, because of the complex dynamics. We show that an image-based guidance algorithm based on deep meta-reinforcement learning can keep the spacecraft on the correct collision path also in presence of unmodeled dynamics and uncertainty.

Featured Image