Lunar ROADSTER: Autonomous Bulldozer for Lunar Terrain Manipulation

Bhaswanth Ayapilla, Boxiang Fu, Ankit Aggarwal, Deepam Ameria, Simson D'Souza

Keywords: autonomous robotics, lunar terrain manipulation, crater grooming, path planning, GPS-denied localization, fisheye odometry, regolith mechanics, field robotics

TL;DR: A terrestrial autonomous bulldozer prototype that identifies, navigates to, and grooms lunar craters into traversable trails using closed-loop perception, terrain validation, and GPS-free localization.

Abstract: We present Lunar ROADSTER, a terrestrial prototype of an autonomous bulldozer for lunar trail grooming in support of sun-synchronous circumnavigation missions. The system autonomously identifies gradable craters, navigates to each along a reference latitude ring, and executes a closed-loop grooming-validation cycle until the trail segment meets traversability criteria. The platform integrates a custom YOLOv8-nano crater detection pipeline, lattice-based A* path planning, and a depth-based terrain validation module. Localization is achieved without GPS by fusing fisheye ceiling-light tracking with IMU and wheel encoder data via an Extended Kalman Filter, directly analogous to star-tracker-based navigation on the Moon. The system was validated in a simulated lunar testbed against three mission performance requirements, satisfying all three: mean planned path deviation of 22.89\% ($\le$25\%), mean path-following deviation of 7.01\% ($\le$10\%), and wheel contact pressure of 1.37 kPa ($\le$1.5 kPa).

Submission Number: 28