Knowledge-Based Hierarchical POMDPs for Task Planning

The main goal in task planning is to build a sequence of actions that takes an agent from initial state state. In robotics, this particularly difficult because usually have several possible results, and sensors are prone produce measurements with error. Partially observable Markov decision processes (POMDPs) commonly employed, thanks their capacity model the uncertainty modify monitor system. However, since solving POMDP computationally expensive, usage becomes prohibitive for most robotic applications. paper, we propose architecture service robotics. context robot design, present scheme encode knowledge about its environment, promotes modularity reuse information. Also, introduce new recursive definition enables our autonomously hierarchy POMDPs, so it can be used generate execute plans solve at hand. Experimental results show that, comparison baseline methods, by following hierarchical approach able significantly reduce time, while maintaining (or even improving) robustness under scenarios vary size.