Expert system architecture designed for flexible robotic arms.

Calangiu, Gabriela Andreea ; Stoica, Mihai ; Sisak, Francisc 等

1. INTRODUCTION

To make the robots understand enough of the real world to permit them to act independently proved to be more difficult than it had been initially thought.

Artificial intelligence (AI) has become an essential part of the technology industry, providing the heavy lifting for many of the most difficult problems in computer science (Kurzweil, 2005). The central problems of AI include such traits as reasoning, knowledge, planning, learning, communication, perception and the ability to move and manipulate objects.

Early AI researchers developed algorithms that imitated the step-by-step reasoning that human beings use when they solve puzzles, play board games or make logical deductions (Luger & Stubblefield, 2004). By the late 80s and 90s, AI research had also developed highly successful methods for dealing with uncertain or incomplete information, employing concepts from probability and economics (Russel & Norvig, 2003).

Human beings solve most of their problems using fast, intuitive judgments rather than the conscious, step-by-step deduction that early AI research was able to model (Lakoff & Nunez, 2000). AI has made some progress at imitating this kind of "sub-symbolic" problem solving: embodied approaches emphasize the importance of sensory-motor skills to higher reasoning; neural network research attempts to simulate the structures inside human and animal brains that give rise to this skill.

Knowledge representation (Luger & Stubblefield, 2004) and knowledge engineering (Russel & Norvig, 2003) are central to AI research. Many of the problems machines are expected to solve will require extensive knowledge about the world. Among the things that AI needs to represent are: objects, properties, categories and relations between objects (Nilsson, 1998); situations, events, states and time; causes and effects; knowledge about knowledge (what we know about what other people know); and many other, less well researched domains.

The research interests in flexible robotic arms have increased significantly over the last few years because they possess some interesting advantages, when compared to rigid robotic arms. One of the distinct advantages is significantly higher payload-to-robot weight ratio, which stems from the lightweight structure of the robotic arm itself (Sakawa et al., 1985), (Balas, 1978) and (Morita et al., 2001).

Fuzzy and neuro-fuzzy based systems have established themselves as strong candidates for developing dynamic system models. Their usefulness in the domain of system identification and control is the source of our inspiration to investigate into a neuro-fuzzy based modeling procedure for flexible robotic arms (Chatterjee et al., 2007). The objective is to develop the neuro-fuzzy model for the flexible arm directly on the basis of experimental data collected from the arm.

The rest of the paper is organized as follows. Section 2 presents a detailed description of the architecture of the expert system developed for the flexible robotic arm. Section 3 gives a brief description of the procedures used for controlling the robotic arm and it is focused on describing how the robotic arm percepts and actions, and Section 4 presents the conclusion.

2. THE ARCHITECTURE OF THE EXPERT SYSTEM

The knowledge base is the sum of the particular knowledge for a certain domain. Therefore, there are more than a few methods used for knowledge representation, but the most important ones are:

* semantic networks;

* production rules;

* frames.

The process of creating the knowledge base is the following:

* the knowledge is taken-over from the human expert and from the flexible robot arm.

* the knowledge is modeled according with the requests of the representation method;

* the knowledge is introduced/stored in the data base and validated.

Facts base contains the data of a particular problem which is to be solved (the statement of the problem), but also the facts resulting after the reasoning made by the inference engine on the knowledge base.

The inference engine is the element that processes data in the expert system; it starts by the facts (input data) it activates the corresponding knowledge from the knowledge base, building the reasoning which leads to new facts. The inference engine builds a new development plan for solving the problem depending on the particularities of the problem, using the available knowledge from the specific domain.

[FIGURE 1 OMITTED]

Based on the action of the inference engine in a specific context, the knowledge base improves either with new elements or with the modification of the existing ones. Thus, the inference engine is a program which implements reasoning algorithms (deductive, inductive or mixed) but it is independent from the knowledge base.

The role of the explications module is to present into a wide accessible mode the justification of reasoning made by inference engine, but also it presents the questions to the user.

The acquisition module is used to transform the knowledge, from the manner the user and the robotic arm expresses them, into his internal manner. In the same time this module handles the communication interface with the database.

The user interface realizes the dialogue between the user, the robotic arm and the expert system by pointing out the input data and providing the results for the problem to solve.

3. ROBOT'S ARM PERCEPTS AND ACTIONS

Focused on the reactive control of the robot arm, a multithreaded architecture combining element of belief, desires, and intentions is to be developed. The design and the implementation of the program for control are based on the Teleo Reactive (TR) programs first introduced by Nilsson. TR programs are a rule based programming notation influenced by process control concepts of continuous monitoring and action, but they have parameterized procedures and actions can be TR procedure calls, even recursive calls. A TR procedure is also typically goal directed; its rules are oriented on the exploration of the environment, using tests of sensors reading.

Thus, they are well-suited for implementation of the robotic arm control due to their capacity to select a routine depending on the computational analysis of the sensors acquisition. The TR procedure can be programmed to monitor beliefs inferred from sensor readings and a model of the environment, rather than percepts.

A control thread is used for deciding which TR procedure to invoke based on significant events, such as belief updates or new goals. This allows the robot to switch tasks based on message events or sensor reading events that lead to significant changes in the robots beliefs.

The robot arm has a camera pointing forwards and a gripper with pressure sensors. It has percept routines that are constantly analyzing the camera image to provide updates of its perceptual beliefs:

* see(D,C)--C colored spot at position D in the field of vision, D is left, center or right;

* near(C)--close to a C colored thing;

* touch--touching something;

* at(blue)--at the blue corner;

* holding--something is between grippers. It has action routines that enable it to:

* forward(F)--move forward at speed F;

* turn(D,R)--turn to D (left or right) at rotation speed R;

* close--close grippers;

* open--open grippers.

3.1 Motivating example

First the robot arm must investigate the environment and find an enclosed flat space with empty or partial colored elements scattered about and a corner painted blue.

[FIGURE 2 OMITTED]

Assuming that there are no other objects in the space, the robot arm using only the vision system, the gripper and expert system, must locate, grab and deliver to the blue corner the elements partially colored. It then waits until the element is removed by a human and then repeats the task.

The expert system and TR program are used for actions like: perception of the environment, planning, learning, development and improvement of the knowledge base.

4. CONCLUSION

The sequence of steps taken to reach a conclusion is dynamically synthesized with each new case. It is not explicitly programmed when the system is built.

Expert systems can process multiple values for any problem parameter. This permits more than one line of reasoning to be pursued and the results of incomplete (not fully determined) reasoning to be presented.

Problem solving is accomplished by applying specific knowledge rather than specific technique. This is a key idea in expert systems technology. It reflects the belief that human experts do not process their knowledge differently from others, but they do possess different knowledge. With this philosophy, when one finds that their expert system does not produce the desired results, work begins to expand the knowledge base, not to re-program the procedures.

The further work will consist in development of the presented expert system in order to teach the robotic arm to understand the environment. Also an artificial neural network will be implemented and trained in such a way that the robotic arm can be able to act independently.

5. REFERENCES

Balas, M.J. (1978). Feedback control of flexible systems, IEEE Transactions on Automatic Control, Vol. 23, No. 4, (Aug 1978) pp. 673-679, ISSN 0018-9286

Chatterjee, A.; Chatterjee, R.; Matsuno, F. & Endo, T. (2007). Neuro-fuzzy state modeling of flexible robotic arm employing dynamically varying cognitive and social component based PSO. Measurement, Vol. 40, No. 6, (July 2007), pp. 628-643, ISSN 0263-2241

Kurzweil, R. (2005). The singularity is near, Penguin Books, ISBN 0-670-03384-7, New York

Lakoff, G.& Nunez, R.E. (2000). Where Mathematics Comes From: How the Embodied Mind Brings Mathematics into Being, Basic Books, ISBN 0-465-03770-4, New York

Luger, G.F. & Stubblefield, W.A. (2008). Artificial Intelligence: Structures and Strategies for Complex Problem Solving, Addison-Wesley, ISBN 0-321-54589-3, Reading, Massachusetts

Morita, Y.; Kobayashi, Y.; Kando, H., Matsuno. F.; Kanzawa, T. & Ukai, H. (2001). Robust force control of a flexible arm with a nonsymetric rigid tip body, Journal of Robotic System, Vol. 18, No. 5, (Apr 2001) pp. 221-235, ISSN 0741-2223

Nilsson, N. (1997). Artificial Intelligence: A New Synthesis, Morgan Kaufmann, ISBN 1-55860-535-5, San Francisco

Russel, S. J. & Norvig P. (2003). Artificial Intelligence: A Model Approach, Prentice Hall, ISBN 0-13-790395-2, New Jersey

Sakawa, Y.; Matsuno, F. & Fukushima, S. (1985). Modeling a feedback control of a flexible arm. Journal of Robotic System, Vol. 2, No. 4, (Apr 1985) pp. 453-472, ISSN 0741-2223