eva‎luation of the position the lunar rover by implementing a machine vision algorithm

10/13/2025 12:00:00 AM

Exploration of the lunar polar regions, particularly permanently shadowed craters with a high potential for water ice deposits, has recently become a central focus of international space agencies. The absence of satellite-based positioning systems (GPS), extremely poor illumination, complex terrain, an‎d the unknown nature of the environment pose major challenges to precise rover navigation in these regions. Classical localization methods, such as wheel odometry an‎d inertial measurement units (IMU), although effective in controlled conditions, suffer from significant accuracy degradation an‎d loss of stability on the lunar surface due to factors such as wheel slip, penetration into soft regolith, an‎d angular drift. In this thesis, three complementary approaches were developed an‎d eva‎luated to enhance rover navigation accuracy: (1) fusion of wheel encoder an‎d gyroscope data within an Extended Kalman Filter (EKF) as the baseline algorithm, (2) a monocular visual odometry–encoder–gyroscope hybrid algorithm employing EKF with slip detection, scale learning, an‎d velocity blending mechanisms an‎d (3) an RGB-D–based visual odometry method with gyroscope angle fusion. The required datasets were generated in a simulated lunar environment using the Webots simulator, including RGB an‎d RGB-D image sequences, IMU data, an‎d wheel encoder signals, while simulated GPS coordinates were used exclusively as ground-truth references for performance eva‎luation. Quantitative an‎d qualitative results demonstrated that the monocular visual odometry–encoder–gyroscope hybrid algorithm provided the highest accuracy an‎d stability in rover localization. This approach significantly reduced position, angular, an‎d translational errors, offering robust performance even in the presence of cumulative errors. In comparison, the RGB-D–based algorithm with gyroscope fusion also showed notable improvements over the baseline method but remained less accurate an‎d stable than the monocular hybrid approach.

ادومتري بصري , ماه‌نورد , پيمايش سنجي بصري , سيستم‌هاي مكان‌يابي و نقشه‌ برداري همزمان , رباتيك

Visual Odometry , Lunar Rover , Computer Vision-Based Navigation , Simultaneous Localization an‎d Mapping (SLAM) , Robotics

seyed Javad Shoja Alsadati

Mahdi Nasiri Sarvi