SLICE: self-learnable IoT common software engine

From the perspective of IoT, AmI is one of the promising application areas combining AI and IoT. Existing edge-based IoT platforms provide a centralized AmI intelligence such as perception, reasoning and learning. Recently, advances of technologies on intelligent IoT devices accelerate moving intelligence to the edge device, which enables edge devices to behave as agents. Though agent-based IoT is more suitable for developing AmI applications than edge-based IoT, application developers should design collaborations among individual agents in the agent-based IoT, which is not easy and cumbersome enough to be an obstacle to development. In order to solve the problem, in this paper, we propose a novel agent-based IoT platform called SLICE for developing and managing AmI applications. Our contribution is threefold. First, we propose a new system model for abstracting each individual agent and designing collaboration among those agents. Second, we design and implement a software framework to support developing AmI applications based on the proposed system model. Third, we design and implement a runtime engine and management tools to provide an efficient and robust execution environment for an agent participating in collaborations to perform AmI applications. As a proof of concept, we present complete lifecycles of developing a smart car based on the SLICE platform. From the experiment, we evaluate and verify how the proposed platform efficiently and effectively supports entire development phases, i.e. design, implement, deploy, and execute.