Next: DEDS for Modeling Up: Hybrid and Discrete Previous: Hybrid and Discrete

Discrete event dynamic systems for active visual sensing

An example of a high-level DEDS controller for part inspection can be seen in Figure 5. This finite state machine has some observable events that can be used to control the sequencing of the process. The machine remains in state A until a part is loaded. When the part is loaded, the machine transitions to state B where it remains until the part is inspected. If another part is available for inspection, the machine transitions to state A to load it. Otherwise, state C, the ending state, is reached. If an interruption occurs, such as a misloaded part or inspection error, the machine goes to state D, the error state.

Our approach uses DEDS to drive a semi-autonomous visual sensing module that is capable of making decisions about the visual state of the manipulation process taking place. This module provides both symbolic and parametric descriptions which can be used to observe the process intelligently and actively.

A DEDS framework is used to model the tasks that the autonomous observer system executes. This model is used as a high level structuring technique to preserve and make use of the information we know about the way in which a manipulation process should be performed. The state and event description is associated with different visual cues; for example, appearance of objects, specific 3-D movements and structures, interaction between the robot and objects, and occlusions. A DEDS observer serves as an intelligent sensing module that utilizes existing information about the tasks and the environment to make informed tracking and correction movements and autonomous decisions regarding the state of the system.

To be able to determine the current state of the system we need to observe the sequence of events occurring in the system and make decisions regarding the state of the automaton. State ambiguities are allowed to occur, however, they are required to be resolvable after a bounded interval of events. In a strongly output stabilizable system, the state of the system is known at bounded intervals and allowable events can be controlled (enabled or disabled) in a way that ensures return in a bounded interval to one of a desired and known set of states (visual states in our case).

One of the objectives is to make the system strongly output stabilizable and/or construct an observer to satisfy specific task-oriented visual requirements. Many 2-D visual cues for estimating 3-D world behavior can be used. Examples include: image motion, shadows, color and boundary information. The uncertainty in the sensor acquisition procedure and in the image processing mechanisms should be taken into consideration to compute the world uncertainty.

The observer framework can be utilized for recognizing error states and sequences. This recognition task will be used to report on visually incorrect sequences. In particular, if there is a pre-determined observer model of a particular manipulation task under observation, then it would be useful to determine if something goes wrong with the exploration actions. The goal of this reporting procedure is to alert the operator or autonomously supply feedback to the manipulating robot so that it can correct its actions.

Next: DEDS for Modeling Up: Hybrid and Discrete Previous: Hybrid and Discrete
Tue Nov 22 21:30:54 MST 1994