Distributed AI framework for low-power embedded devices

September 28, 2014   

Systems, Applications and Technology Conference, 2008 IEEE Long Island

Date of Conference: 2-2 May 2008
Publication Reference: Systems, Applications and Technology Conference, 2008 IEEE Long Island


Traditional middleware is typically designed to sit on top of established operating systems, which already provide rich abstractions such as task and memory management. In this paper, we propose a new middleware framework for external sensor systems to aid the development of a target-tracking algorithm. The middleware for our framework consists of three components: Target Server, Data Server, and Interface Server. The “Target Server” is used for receiving sensor data and performing data fusing, target tracking, and target prediction. The “Data Server” is used for maintaining a distributed database of TinyDB and providing target information for the target server. To improve service performance,the Data Server keeps snapshots (views) of current targets in main memory to avoid saving and retrieving data from the TinyDB database. Finally, the Interface Server is used for identifying the communication boundary between the data server and the vehicle’s interface. The Interface Server acts as the gateway between the clusterheads and the vehicle’s on-board computer. We demonstrate the utility of our framework by showing how it can be used to implement a simple, low-power, target-tracking algorithm.

A Distributed Reconfigurable Active SSD Platform for Data Intensive Applications
Refers to our implemented high performance indexing and adaptive algorithms based on available power.

comments powered by Disqus