Distributed Autonomous Control Architecture for Intelligent Mobile Robot Systems

Introduction

In the past decades, robots are becoming commonplace in unstructured and dynamic environments, ranging from homes to public and disastrous sites. Intelligent robots, such as humanoid robots, are typically designed to have capability of manipulating objects enough to achieve complicated and variable task goals as well as to have the locomotion capability, while adapting its behavior to particular environment conditions (Elkady et al., 2011; Hammer et al., 2010). Current behavior-based robot architectures so far mainly concerns navigation tasks performed by mobile robots (Arkin, 1987; Brooks, 1987; Connell, 1992), and are not fit for intelligent robots because the task executing environment is not well structured (Gat, 1992; Noreils, et al. 1995). Advanced intelligent mobile robots have to deal with a number of subtasks in real time in order to perform complex tasks with a broad variety of objects autonomously in dynamic and much less structured environments. (Konolige et al., 1997). It is necessary to attain the autonomous synchronization of the various subsystems involved in the navigation and task operation system, because intelligent subsystems have the inherent capability to act asynchronously and concurrently or collaborate with each other.

The use of a single processor system in a multi-tasking kernel with real-time features establishes a common system resource that is shared among all software modules or behaviors, but raises the possibility for mutual interference with negative effects in terms of timeliness with slower and longer modules blocking faster reactive ones, reducing the robot’s reactive capabilities. Controllers with the multiprocessor in parallel can realize real-time performance effectively, dividing the complex task into the several simple subtasks which every processor does respectively and coordinating the processors. Generally, distributed control systems consist of several decision making and executive subsystem controllers, and available information is processed by the subsystem controllers, which communicate and negotiate to come to a common decision using a communication channel, and then the decision is executed by them to fulfill a global goal. From the view of communication networks, two approaches are distinguished for multiprocessor based control system design, namely, centralized and decentralized approaches. In the centralized approach, control of each executive subsystem controller (processor or agent) is based on global information achieved by a central controller. A decision is made in the central mechanism for the system goal and transmitted to executive subsystem controllers. Whereas, in the decentralized approach, each subsystem controller makes decision based on its local information achieved through its own sensors and from the neighbor subsystem controllers. Decentralized control is useful when the global information is not available, because there is no central mechanism to coordinate the overall robot behavior. In contrast to the centralized approach, decentralized control architectures reveal their main advantages when it becomes necessary to maintain the system, to integrate components, and to enhance the system, through system-inherent redundancy without any error model, cooperation of subsystems, addition of new system components, and so on (Camargo et al., 1991). Practically, the main problem of decentralized architectures is coordination between the subsystem controllers to make sure that the system will fulfill an overall or global goal, due to the independent task execution by the subsystem controllers.