The sequencing of data flow tasks in SIGNAL: application to active vision in robotics

Abstract

Many applications of real-time systems feature a combination of "continuous" (possibly sampled) and discrete (sequencing and task control) behaviors, both reactive to the evolutions of the execution environment. In particular, robotics applications involve the programming of the control functions for each task, and the sequencing of such tasks at a higher level. In the framework of the real-time synchronized data-flow language SIGNAL, we propose extensions based on time intervals, enabling the specification of tasks, and of their suspension or abortion for their sequencing in reaction to discrete events. This paper illustrates the application of these constructs to active vision in robotics, where successive phases have to be sequenced, each associated with a particular control law, in order to recognise a complex environment.<>