An Intelligent Mobile Robotic System

An Intelligent Mobile Robotic System

INTRODUCTION:

Fuzzy logic has been applied in various areas such as knowledge engineering, computer science, manufacturing system, mechatronics and robotics. Furthermore various fuzzy interface methods have been proposed to deal with imprecise information, human language and non-linear mappings. With the progress of computation capabilities, techniques within the field of artificial intelligence (AI) have been developed in order to describe and build intelligent agents that perceive and sense environment, make appropriate decisions and take actions. Recently, human intelligence and life itself have been studied in cognitive science, soft computing, artificial life and computational intelligence in addition to artificial intelligence. The intelligence can be divided into three levels:

Artificial
Biological
Computational

In the strictest sense CI depends on numerical data and doesn’t rely on explicit knowledge. Furthermore CI can be defined as a methodology that involves computing. It aims to construct intelligence from the viewpoint of biology, evolution and self organization. CI tries to construct intelligence by internal description; where as classical Artificial Intelligence (AI) tries to construct intelligence from external description. These intelligent methods have been often used for robotic system. In such system human operators often dictate both the sensing system and control rules, but ideally the robot should automatically perform the given task without assistance. Consequently, the robot must be able to perceive the environment, make decision, to represent sensed data Acquire knowledge and infer rules concerning the environment. Robots that can acquire and usefully apply knowledge or skills are often called as intelligent robot. Intelligence emerges from the close linkage of many capabilities.

An intelligent robot receives a task from a human operator and must accomplish the task in the available space, which may include many obstacles such as people, machining centers and other robots. The intelligent robot should therefore take into account the problem of collision avoidance. Furthermore, the intelligent robot should automatically generate its motion for performing the task. Motion planning fundamentally includes path, trajectory and task planning. Here we define these planning problems as follow. The path-planning problem requires generating the shortest path for robot from a given starting point to a target point while satisfying the spatial constraints. The trajectory-planning problem requires generating a robot trajectory that satisfies the time constraints. The task-planning problem requires finding a sequence of primitive motion commands for solving a given task.
These commands are described by robot programming language. A task-level programming system allows user to describe the task in a high-level language. To accomplish the task mobile robot performs geometric map building and path planning.

Path planning and collision avoidance is closely related. The search space (environmental map) can build by polygonal approximation, cell decomposition and artificial potential fields. In polygonal approximation, obstacles in the workspace are approximated as polygonal on a map. This simplified the search space. In cell decomposition a two-dimensional workspace is divided into N*N cells to reduce the search space. In the artificial potential field method, a robot manipulator moves based on an attractive force from the goal and a repulsive force from the obstacles in the workspace. In the hybrid method of the cell decomposition and artificial potential fields, each cell on the divided cell space is assigned a psudopotential value based on the distance from the obstacles in the workspace.

Evolutionary algorithms have been applied to the path-planning problems. Xaio et al proposed an adaptive evolutionary planner/navigator using various operators to evolve and improve candidate paths. The motion of a robot is constrained fundamentally by its dynamics, kinematics and workspace. The trajectory planning must satisfy the constraint, such as bounded velocity and acceleration, maximal torque and also task dependency. Therefore, path planning should take into account these constraints in trajectory planning. However, these methodologies based on the map constructed a priori and are suitable for offline path planning but it is difficult to build a map a priori because of constraints that may include moving obstacles. Furthermore, in such a case, the control rules of mobile robot must be designed to handle any environmental conditions beforehand. Now a day various adaptive robotic systems have been proposed to adopt under dynamic or unknown environments.

There are two methods that are used for mobile robot so far are subsumption architecture and behavior based artificial intelligence (BBAI). The subsumption architecture uses a layered finite state to present the agent. The design is decomposed into behavior such obstacles avoiding, photo tracing and wall following. In addition, reinforcement learning methods such as the bucket bridge algorithm, the temporal difference method and Q-learning have been discussed as method for inducing rules of behavior through the interaction with environment. This kind of BBAI stresses the importance of the direct interaction between robot and environment. In contrast, classical artificial intelligence is based on representation and manipulation of explicit knowledge. However, the behavior-based approach needs to design a new controller for each task.

Recently the other method describes as behavior analysis and training Monalysa architecture, model based learning, hierarchical intelligent control and others have been proposed. Monalysa architecture relies upon a hierarchical classifier system including reactive and planning modules. A simulated robot has many possible rules and decides what action to perform next. A given task is then logically decomposed into a series of subtasks in the planning modules. On the other hand, hierarchical intelligent control for robotic system comprises.

Learning level of knowledge.
A skill level based on fuzzy rules.
An adaptation level to adjust rules according to environmental condition.

Each level has a feedback loop for information processing between levels. The intelligence of a robotic system depends upon the structure of hardware and software for processing information i.e. structure influences the potential intelligence. Furthermore, the intelligence of a robotic system can evolve through learning and adaptation in dynamic environment. In next section we discuss a robotic system with structured intelligence that emphasized the importance of the entire integrated system architecture. Based on the perceptual information, a robotic system with structured intelligence makes decision and actions deriving from four different levels in parallel.

for more info visit.
http://www.enjineer.com/forum