Deep learning to enhance future Mars rover capabilities

NASA's Mars rovers have made significant scientific contributions over the past two decades, but their current capabilities are limited by the computing power available on board. Future rovers could benefit from deep learning and advanced computing to improve their performance and expand their reach on the Red Planet.

Masahiro (Hiro) Ono, a researcher at NASA's Jet Propulsion Laboratory, highlighted the need for rovers to cover greater distances to better understand Mars' diverse terrain. He compared the current exploration efforts to an alien landing on a few points on Earth and driving a short distance, which would not provide a comprehensive understanding of the planet.

The Perseverance rover uses RAD 750 computers, but future missions may use more powerful, radiation-hardened processors developed through the High Performance Spaceflight Computing project. These new chips could offer 100 times the computational power of current systems while using the same amount of energy.

Chris Mattmann, from JPL, noted that current rover autonomy is largely human-in-the-loop, requiring human interaction. The goal is to enable on-board deep learning and machine learning, similar to what is used on Earth. The MAARS program has been exploring how AI can be applied to rover systems, including new capabilities like Drive-By Science and Energy-Optimal Autonomous Navigation.

Ono and his team have been training machine learning models on supercomputers to develop more advanced pathfinding and terrain analysis for future rovers. They aim to give rovers a more human-like ability to assess and navigate Mars' complex and varied landscape, which requires a shift in how rovers are commanded and controlled.