Researchers at the University of Illinois Urbana-Champaign have developed a novel approach for robots on celestial bodies to decide on sample collection methods and locations autonomously.
Mars rovers rely on human guidance from Earth, whereas lander missions to Saturn or Jupiter’s moons lack real-time communication due to long distances.
Researchers in the Departments of Aerospace Engineering and Computer Science at the University of Illinois Urbana-Champaign have developed an innovative learning approach. This enables autonomous decision-making for robots on celestial bodies to determine optimal methods and locations for collecting terrain samples. The researchers have adopted a unique approach for landers to quickly learn to scoop new materials, adapting to changing landscapes and properties.
The researchers explained that with minimal attempts, robots using this method quickly learn how to scoop new materials. They adjust their strategy if they make unsuccessful attempts, moving to different areas for scooping. Understanding ocean worlds like Europa poses a challenge due to limited knowledge. The resolution of the available images needs to be improved to discern terrain features. The team have highlighted the uncertainty about Europa’s surface and what lies beneath the ice. In some trials, the team concealed material under another layer. The robot, perceiving the top layer as scoopable, learns otherwise upon encountering the non-scoopable bottom layer, prompting it to relocate.
NASA prioritizes sending battery-powered rovers to Europa to minimize contamination risks to ocean worlds from potentially dangerous substances, alongside other mission-specific factors. With nuclear power lasting months and batteries only enduring 20 days, exchanging messages daily becomes impractical. Thus, the robot’s autonomy becomes essential for efficient decision-making. The unique learning method uses vision and little experience for superior scooping. Illinois team’s robot collects data on various materials, amassing 6,700 points in a 100-point knowledge database for 67 terrains. The team’s model will be utilized at NASA’s Jet Propulsion Laboratory’s Ocean World Lander Autonomy Testbed.
The significance lies in transferring adaptable knowledge and methods from Earth to extraterrestrial bodies. Limited pre-landing information and short battery life necessitate swift autonomous learning and decision-making, as the lander’s lifespan may be brief.
Reference: Yifan Zhu et al, Few-shot Adaptation for Manipulating Granular Materials Under Domain Shift, Robotics: Science and Systems XIX (2023). DOI: 10.15607/RSS.2023.XIX.048 www.roboticsproceedings.org/rss19/p048.html
