A framework for modeling and verifying visually guided agents: Design, analysis and experiments

Jana Kosecka, University of Pennsylvania

Abstract

The successful functioning of robotic agents in a dynamically changing environment requires rich sensory input and advanced sensory-motor control. The robotic agents are typically equipped with multiple sensors and multiple actuators. The interactions between agents and the environment can be characterized by both discrete and continuous models. The accomplishment of various tasks is mediated by complex coordination and interaction between individual sensing and control strategies. It is crucial for the reliable and predictable operation of robotic systems that the design be within a structured methodology which supports analysis and modularity. In this thesis we introduce a framework for modeling and verifying autonomous mobile agents which provides a unified approach to the design and analysis of systems comprising of both continuous and discrete event components. Our framework proposes to model elementary sensing and control strategies by making appropriate finite state machine abstractions which capture the discrete event aspects of continuous models. In order to achieve the desired modularity and flexibility we define a Task Specification Language for composing elementary strategies to form more complex missions. The robustness and reliability of individual control strategies is guaranteed by their design and analysis at the continuous level. The verification of the discrete event interactions between strategies is formulated as a Supervisory Control Theory Problem, where we synthesize a supervisor which serves as a run-time scheduler and monitor of the task. The overall framework is verified in a series of experiments where teams of visually guided mobile agents are engaged in various navigation tasks. Within our framework, we formulate the visually guided navigation problem and propose a relational model of the environment embedded in a finite state machine structure which can then be used for automatic generation of the task specification.

Subject Area

Computer science|Electrical engineering

Recommended Citation

Kosecka, Jana, "A framework for modeling and verifying visually guided agents: Design, analysis and experiments" (1996). Dissertations available from ProQuest. AAI9627947.
https://repository.upenn.edu/dissertations/AAI9627947

Share

COinS