Task Planning Based On Htn For Robots With Memory

Recently, AI planning has been commonly applied to robot field. HTN planning has been much more applied than most other AI-planning research for its unique characteristics and good performance in the experiments. However, the conflict between simplified assumption for planning on the one hand and the complexity of the real world on the other occurs, which becomes a major challenge in robotics research. The current work environments for mobile robots tend to be complex and dynamic, which often causes difficulties in the execution of tasks due to incomplete and uncertain information. A number of experiments and applications have shown that it is an appropriate and efficient approach to handle this issue by integrating the AI method with plan-based robot control.In this paper, an HTN (Hierarchical Task Network) planner is integrated in the control architecture of mobile robot as the deliberative planner. Simultaneously, by combining a robot memory base which can probabilistic inference the robot's former experience, hence the lacking information for robot planning can be compensated. A cognitive architecture system which integrates enhanced HTN planning with memory database is proposed in this paper.Firstly, the related operators that intelligent household robot implements daily household tasks are modeled in planning language. Simultaneously, the objects in the environment that are associated with robot skills are classified into two kinds by object-oriented approach, i.e. symbolic properties and physical properties. And the daily tasks are hierarchically decomposed into primitive operators and modeled by HTN method. Then an enhanced plan-based control system integrated outside information is mainly described. The system is implemented by HTN method and combined with a robot memory base which can probabilitic inference the robot's former experience, hence the lacking information for robot planning can be compensated. A new approach of automated generating plan problem is also described. The execution monitoring module is used to judge whether the plan has a provision for the actual current situations or need to be revised. Replanning occurs when something goes wrong. These mechanisms can speed up the overall system efficiency and robot autonomy.Several experiments show that the robot can accomplish the task more efficiently under uncertain information in a dynamic environment by using this approach. The revised HTN planner is carried on two different platforms, i.e. clean-service robot and mobile robot. The success of experiment also shows the reconfiguration and portability of the system of the enhanced HTN planner.