Next: State Modeling and Up: Hybrid and Discrete Previous: Discrete event dynamic

DEDS for Modeling Observers

DEDS can be considered as very suitable tools for modeling observers. In particular, in the manipulation observer domain, there is a need to recognize and report on distinct and discrete visual states, which might represent manipulation tasks and/or sub-tasks. The observer should have the ability to state a symbolic description of the current manipulation agent action. The coarse definition of DEDS states provide a means for such symbolic state descriptions.

The definition for observers and the observer construction process for discrete event systems are very coherent with the requirements for an autonomous robotic observer. The purpose of DEDS observers is to be able to reconstruct the system state, which is exactly the requirements for a visual observer, which needs to recognize, report and possibly act, depending on the visual manipulation state. The notions of controllable actions is easily mapped to some tracking and repositioning procedures that the robotic observer will have to undertake in order to ``see'' the scene from the ``best'' viewing position as the agent under observation moves over time. The actions which the observer robot might need to perform, depends on the sequence of ``observable'' events and the reconstructed state path.

Event description in a visual observer is possibly a combination of different 2-D and 3-D visual data. The visual primitives used in an observer domain could be motion primitives, matching measures, object identification processes, structure and shape parameters and/or a number of other visual cues. The problem with the DEDS skeleton is that it does not allow for smooth state changes under uncertainty in recovering the events. We describe in the next sections techniques that make the transition from a DEDS skeleton into a working hybrid observer for a moving manipulation agent. Stability and stabilizability issues are resolved in the visual observer domain by supplying suitable control sequences to the observer robot at intermittent points in time in order to ``guide'' it into the ``desirable'' set of visual states.

Next: State Modeling and Up: Hybrid and Discrete Previous: Discrete event dynamic
Tue Nov 22 21:30:54 MST 1994