The ROS (Robot Operating System) is a set of software libraries and tools used to build robotic systems. Recently, some researchers have been working on integrating planning systems into ROS. However, robotics domains are often probabilistically interesting. At present, only a few studies focused on integrating probabilistic planning into ROS. Moreover, they have limits on flexibility or scalability. Therefore, we propose a new framework, called PRobPlan ( PRob abilistic Rob ot Plan ning), to alleviate these problems. On one hand, our framework uses a series of generation programs to build a problem file instead of the knowledge base. This makes it more modifiable. On the other hand, our framework integrates a state-of-the-art planner, SOGBOFA, and enhances it with a new recommendation function. This makes it more scalable. We instantiate the proposed framework in a warehouse-robot domain where mobile robots are allocated tasks of fetching or packing goods. Both of the domain and problem instances are modeled with RDDL (Relational Dynamic Influence Diagram Language). Experimental results showed the effectiveness of the proposed framework. We also tested two top planners, SOGBOFA and PROST, for comparison. The enhanced version of SOGBOFA mildly outperformed its original version and vastly outperformed PROST. Our work promotes the integration of probabilistic planning with robotic systems.