In order to achieve efficient and reliable destination reaching for indoor navigation, mobile robots must accurately perceive their surroundings utilizing multi-modal sensors, dynamically avoid various obstacles, and plan optimal trajectories. In practice, there usually exist obstacles (e.g., walls, cabinets) blocking the sight of mobile robots when they reach the corner or the intersections. Sensors equipped on mobile robots like LiDAR and depth cameras often fail to detect occluded pedestrians and obstacles, potentially leading to suboptimal path planning. Therefore, enabling around-corner sensing can help prevent potential collision by triggering better path planning. For various sensors equipped on mobile robots, mmWave radar provides the opportunity to perceive conditions of around-corner obstacles via multi-bounce signal reflection through wall. In this paper, we achieve the non-line-of-sight (NLOS) sensing by utilizing mmWave Radar with LiDAR assistance. Furthermore, we propose an obstacle-aware corner navigation algorithm based on reinforcement learning (RL) methods, which leverages the NLOS sensing results for corner navigation. Besides, in order to address the challenge that the position and orientation of the mmWave radar significantly influence the performance of NLOS sensing, we design a specialized RL architecture where mmWave radar’s instant observations are embedded into the optimization objective for path planning. Experiments show that our proposed method enables more accurate NLOS sensing for mobile robot corner navigation, and significantly outperforms the baseline method with respect to the navigation success rate.