Search

A time series (or equivalently a data stream) consists of data arriving in time order. Single or multiple data streams arise in fields including physics, finance, medicine, and music, to name a few. Often the data comes from sensors (in physics and medicine for example) whose data rates continue to improve dramatically as sensor technology improves and as the number of sensors increases. So fast algorithms become ever more critical in order to distill knowledge from the data. This paper presents our recent work regarding the incremental computation of various primitives: windowed correlation, matching pursuit, sparse null space discovery and elastic burst detection. The incremental idea reflects the fact that recent data is more important than older data. Our StatStream system contains an implementation of these algorithms, permitting us to do empirical studies on both simulated and real data.

Autonomous robot architectures typically fall within two camps, top-down symbolic AI systems, and physically grounded bottom-up reactive systems. In this paper, we present a novel behavior fusion based architecture for the control of an intelligent powered wheelchair, which takes advantage of the modularity presented by the reactive bottom-up design, the inference mechanism of symbolic AI systems, and the flexibility of the layered architecture. Thus, this control architecture of wheelchair can realize the multi-level information fusion and facilitates task and information sharing and trading between man and machine. In addition, the basic behaviors and their fusion algorithm are also described.