Planning, coordination and execution of perceptual and motor actions in sensor-based intelligent robots

Given the complexity of an intelligent robotic system, important design considerations are: high level of autonomy, automatic task planning and sensor-driven capabilities, integration of sensing and actions, and the software system's generality, modularity, and portability. In this dissertation we introduce a System Architecture using Frames for Intelligent Robots and their Environments (SAFIRE). The system architecture consists of Perception, Motor, Task Planner, Knowledge-Base, User Interface and Supervisor modules. Frames, a data structure used in knowledge representation, provide a suitable and flexible scheme for representation and manipulation of the world model, the sensor derived information, as well as for describing the actions required for the execution of a specific task. Using the frame-based structure, we present an automatic task planner, PRISM, which integrates the sensing and motor actions in task planning and control of the SAFIRE system. The closely coupled Perception and Motor functionalities are crucial for the robot to operate successfully in a dynamically changing environment such as the inspection and manipulation task domain. We also describe SIRSIM, a Sensor-based Intelligent Robot Simulator. This unique simulator, combined with the Task Planner, Perception and Motor modules, significantly increases the efficiency and reliability of the system development. Extensive evaluation of the SAFIRE system is conducted exclusively in a real-world environment: Validity of the SAFIRE system architecture is illustrated by designing and implementing a robotic system which utilizes a multisensor robot testbed. The system can accomplish complex tasks such as inspection and manipulation of a control panel and autonomous spiU cleaning. The experimental results show the basic validity of the general architecture as well as the robust and successful performance of the system. The constraints of real-time performance, high-level autonomy, sensor-driven capability, and software system flexibility of the working system motivate resolutions of several important issues in developing intelligent robots: (1) a truly robust system can only be obtained through a carefully designed system architecture and systematically coordinated (integrated) modules performing in concert within the architecture, (2) automatic task planning with integrated sensing and action is critical for a robotic system to be capable of performing complex tasks in highly unstructured and dynamic work environments, (3) integration of plan generation and execution is a crucial feature for a task planner operating in real-world situations in which replanning, error recovery, and interrupt servicing must be considered, and (4) the simulator in an integrated sensor-based robotic system must incorporate simulation of sensory information feedback.