International Journal of Science, Engineering and TechnologyAuthors: Mirza Abrar Baig
Abstract: Autonomous rover operation in unstructured and uncertain environments requires control systems that are both adaptive and provably stable. Classical model-based approaches often exhibit limited robustness under terrain variability, sensor noise, and actuator uncertainty, while purely learning-based methods lack formal safety guarantees. This paper proposes a hierarchical autonomous rover control framework that integrates adaptive artificial intelligence with model predictive control and Bayesian risk awareness. The proposed architecture combines multi-sensor perception, reinforcement learning–based decision making, and adaptive model predictive control to enable real-time learning while ensuring bounded closed-loop behavior. A Lyapunov-based stability analysis establishes uniform ultimate boundedness of the system under bounded disturbances. Extensive experimental validation is conducted using both simulation and real-world rover platforms, including terrain disturbance injection, Monte Carlo trials, energy-aware evaluation, simulation-to-reality comparison, and ablation studies. Statistical reliability is demonstrated through 95% confidence intervals and significance testing, confirming that the proposed approach achieves faster convergence, lower steady-state error, and reduced energy consumption compared to baseline and single-method controllers. The results demonstrate that the proposed hybrid adaptive AI framework provides a robust, energy-efficient, and practically deployable solution for safety-critical autonomous rover applications.
DOI: http://doi.org/10.5281/zenodo.18631355
