To fulfil their tasks, autonomous robots have to be able to independently and safely reach a target location from their current position. Path planning enables a robot to do exactly this. Usually, this is done with a map containing the static obstacles (e.g., walls) of the environment. Since dynamic obstacles (e.g., humans) can be present in the environment, the static map alone is not sufficient. Thus, the robot additionally needs sensors to continuously observe the environment to detect dynamic obstacles and to avoid them. This approach, to avoid dynamic obstacles when the robot "sees them", can lead to highly suboptimal behaviour, since the robot will have to start, stop and change the direction frequently. To remedy this problem, probabilistic maps are used in this thesis to enhance common path planning strategies. In contrast to static maps, a probabilistic map contains probabilistic information about the likelihood of encountering dynamic obstacles in certain areas of the environment. This information is used to avoid dynamic obstacles, such that the robot does not have to react to them. In this thesis, the navigation stack from the robot operation system (ROS) is used to allow a Pioneer 3-AT (P3AT) robot to navigate autonomously. The ROS navigation stack, responsible for path planning, is modified to use probabilistic maps to avoid dynamic obstacles. Additionally, further adjustments and software components are developed to incorporate the limitations of the Pioneer 3-AT robot. Since the Pioneer 3-AT robot only has sonar sensors the ROS navigation stack is modified to use these sensors instead of a laser scanner. With the help of experiments, suitable parameters for the new planning strategies have been found. Finally, the advantages of path planning using probabilistic maps compared to static maps are shown with further experiments.