Reconciling Simulation-to-Real Discrepancy and Digital Twins Divergence in Autonomous Rovers

This paper examines simulation-to-real discrepancies and digital twin state divergence in autonomous rovers operating in changing and unpredictable environments. It reviews existing localization and state-estimation technologies, including EKF/UKF sensor fusion, SLAM, visual-inertial odometry, and wheel-slip detection, with attention to their limitations in maintaining consistency between physical robot movement and its digital representation. The paper proposes a conceptual framework based on discrepancy-triggered parallel simulation. When a meaningful deviation occurs between predicted and observed rover movement, the framework preserves the previous predicted trajectory and creates parallel branches representing alternative explanations and possible future trajectories. These branches can be compared against subsequent sensor observations, while historical predictions and discrepancies are retained for analysis and potential model improvement. The proposed approach is theoretical and has not yet been experimentally validated. Its potential benefits, computational requirements, and limitations are discussed as directions for future research.