STRIPS planning with modular behavior selection networks for smart home agents

A smart home has highly advanced automatic systems for lighting, temperature control, multi-media and many other functions. In the field of intelligent service agents in smart home, the service agent should collect the information using sensors at home such as cameras, temperature sensors and light sensors, and generate agent behaviors appropriate to the user's request. This paper presents a smart home system with the intelligent agent using modular behavior networks and STRIPS planning system. Behavior selection networks, one of the well-known behavior based methods suitable for goal oriented tasks, are designed in two types of modules: The networks with specific goal for the user's request, and the networks with common goal to achieve a sub-goal in the specific networks. This modular approach can facilitate reuse and reduce the computation of activation levels without changing the structure. In order to solve complex problems in situations that require sequential inference, we propose a hybrid system, a goal-oriented STRIPS planning system in behavior selection networks, to achieve global goals efficiently in a smart home. The proposed method is applied to the smart home implemented using Unity3D and verified the usefulness by various scenarios. Experimental results confirm the reduction in computation of activity on action nodes.