Institute of Informatics SAS, firstname.lastname@example.org
Over the last few years, a number of studies were reported concerning a machine learning, and how it has been applied to help mobile robots to improve their operational capabilities. One of the most important issues in the design and development of intelligent mobile system is the navigation problem. This consists of the ability of a mobile robot to plan and execute collision free motions within its environment. However, this environment may be imprecise, vast, dynamical and either partially or non-structured. Robots must be able to understand the structure of this environment. To reach their targets without collisions, the robots must be endowed with perception, data processing, recognition, learning, reasoning, interpreting, decision-making and action capacities. The ability to acquire these faculties to treat and transmit knowledge constitutes the key of a certain kind of artificial intelligence. Reproduce this kind of intelligence is, up to now, a human ambition in the construction and development of intelligent machines, and particularly autonomous mobile robots. To reach a reasonable degree of autonomy two basic requirements are sensing and reasoning. The former is provided by on board sensory system that gathers information about robot with respect to the surrounding scene. The later is accomplished by devising algorithms that exploit this information in order to generate appropriate commands for robot. And with this algorithm we will deal in this paper.
We report on the objective of the motion planning problem well known in robotics. Given an object with an initial location and orientation, a goal location and orientation, and a set of obstacles located in workspace, the problem is to find a continuous path from the initial position to the goal position, which avoids collisions with obstacles along the way. In other words, the motion planning problem is divided into two sub-problems, called ‘Findspace’ and ‘Findpath’ problem. For related approaches to the motion planning problem see reference (Latombe, J.C. 1991). The findspace problem is construction the configuration space of a given object and some obstacles. The findpath problem is in determining a collision-free path from a given start location to a goal location for a robot. Various methods for representing the configuration space have been proposed to solve the findpath problem (Brady, M. & all 1982), (Latombe, J.C. 1991), (Vörös, J. 2002). The major difficulties in the configuration space approach are: expensive computation is required to create the configuration space from the robot shape and the obstacles and the number of searching steps increases exponentially with the number of nodes. Thus, there is a motivation to investigate the use of parallel algorithms for solving these problems, which has the potential for much increased speed of calculations. A neural network is a massive system of parallel distributed processing elements connected in a graph topology. Several researchers have tried to use neural networks techniques for solving the find path problem (Bekey, G.A. & Goldberg, K.Y., 1993).
In this paper we introduce a neural networks-based approach for planning collision-free paths among known stationary obstacles in structured environment for a robot Janglová, D. / Neural Networks in Mobile Robot Motion, pp. 15-22, Inernational Journal of Advanced Robotic Systems, Volume 1 Number 1 (2004), ISSN 1729-8806 16 with translational and rotational motion. Our approach basically consists of two neural networks to solve the findspace and findpath problems respectively. The first neural network is a modified principal component analysis network, which is used to determine the “free space” from ultrasound range finder data. Moving robot is modeled as a two-dimensional object in this workspace. The second one is a multilayer perceptron, which is used to find a safe direction for the next robot step on the collision-free path in the workspace from start configuration to a goal configuration while avoiding the obstacles.
The organization of the paper is as follows: section 2 brings out the briefly description of neural network applications in robotics. Our approach to solving the robot motion problem is given in section 3. Our method of motion planning strategy, which depends in using two neural networks for solving the findspace problem and the findpath problem respectively will be described in section 4. Simulation results will be included in section 5. Section 6 will summarize our conclusions and gives the notes for our further research in this area.
To read this full paper please click the "Download Document" button below: