Autonomous mobile robots operating in outdoor unstructured environments face major challenges due to uneven terrain, dynamic obstacles, sensor uncertainty, and GPS-denied conditions. Traditional navigation systems often fail to provide reliable performance in such environments. This paper proposes a robust hybrid navigation and control framework integrating semantic segmentation, traversability estimation, intelligent path planning, and adaptive Model Predictive Control with Reinforcement Learning (MPC-RL). The proposed framework combines multi-sensor perception using RGB-D cameras, IMU sensors, and wheel encoders to improve environmental understanding and navigation safety. A hybrid global-local planning strategy is implemented for energy-efficient path generation and dynamic obstacle avoidance. The adaptive MPC-RL controller improves path tracking accuracy and motion stability under terrain disturbances such as wheel slippage and uneven surfaces. The system is implemented using ROS 2 and evaluated through Gazebo simulation and real-world outdoor experiments. Experimental results demonstrate significant improvements in navigation success rate, obstacle avoidance capability, tracking accuracy, and energy efficiency compared to conventional navigation approaches. The proposed system is suitable for applications such as precision agriculture, environmental monitoring, inspection, and search-and-rescue operations.
G et al. (2026) studied this question.