For collision and obstacle avoidance in path planning, robots usually rely on basic 2D cost maps lacking semantic information about detected obstacles. As a result, the robot’s path planning follows an arbitrarily large safety margin around obstacles.We present a risk-aware 2D cost map for robot navigation that effectively mitigates potential risks, enabling the robot to navigate more confidently and efficiently while maintaining a safe distance from obstacles. It uses commonly available RGBD sensors, making it a practical and accessible option for many applications.Our approach employs a CNN to segment object instances into three distinct safety classes based on common characteristics. Using the abstract representation, we generate semantic occupancy grids, which are then inflated based on their safety classification. These semantic occupancy grids are merged into a final risk-aware 2D cost map. We provide feasible real world results.A robot’s path planner can use the merged cost map as a drop-in replacement for standard 2D cost map, enhancing established robot navigation algorithms based on cost maps without altering their fundamental algorithms and enabling risk-aware navigation in the real world.