Visual Observation
for
Hybrid Intelligent Control Implementation
Tarek M. Sobh
,
Ruzena Bajcsy, and John R. James
Abstract
We address the problem of design and implementation of a discrete event dynamic system (DEDS) observer for the execution of commands sent to a robotic arm during grasping and screwing tasks of an assembly operation. We discuss the resulting robot arm mechanism as a hybrid intelligent system. We argue that the non-intrusive observation mechanism offers reliability and robustness advantages over other sensor systems used to detect errors. Hybrid systems contain both continuous and discrete components. Hybrid systems analysis differs from conventional digital control systems analysis in that the discrete component is appropriately modeled using a finite-state machine (DEDS) to describe high-level machine dynamics and the continuous component contains both analog and digital models of the lower-level continuous variable dynamic system (CVDS) portion of the device. Another description of hybrid systems is that they are networks consisting of continuous physical devices controlled by discrete digital programs, requiring digital-analog transfer of information for their control (logical-linguistic techniques may be used to derive control laws for continuous systems and machines are structured as an interconnected network of continuous, nonlinear systems and finite-state automata). Our work has focused on the use of the DEDS linguistic approach for abstracting high-level knowledge concerning the current state of the machine in the presence of errors, mistakes and uncertainties in the manipulation system.