Intelligent control of robot engineering systems in uncertainty conditions

Fig. 2. Structural diagram of the intelligent control system of the robotic system.
Traditional control systems for industrial manipulators are divided into several classes. The first class of systems is program control systems.
The system of continuous control of the working body of the manipulator implies trimming the manipulator to the reference model [1]. This control algorithm does not take into account the losses in the MC of the manipulator and it is assumed that all the efforts developed by the drives are transferred to the working body.
The system of software control of the force in the working body is used to control not only the force vector, but also the position vector of the working body. The system of independent control of movement and force in the working body of the manipulator for various degrees of freedom has two control loops with feedback: by position and by force [2].
In the system of coupled control of displacement and force in the working body of the manipulator, the task according to the position vector of the working body is corrected according to the current value of the force vector. This means that when the working body moves, the magnitude of its stroke is adjusted according to the force of impact on the external environment.
The problem of the functioning of a robotic system under conditions of uncertainty is multifaceted. Consider the problem of planning the behavior of a robotic system under uncertainty. To solve it, it is most expedient to use the technology of dynamic expert systems. The knowledge base of such an expert system is adjusted over time. If a production rule base is used, then the composition of the production rules is continuously examined for completeness and consistency [4]. In addition, due to adaptive algorithms, outdated and outdated rules are updated and replaced. At the same time, the issues of training an expert system without a teacher (self-learning) are given special attention due to the fact that monitoring the system of a highly qualified specialist is not economically feasible.
The block of self-learning or self-tuning of the knowledge base of the expert system requires careful study at the stage of designing an intelligent control system for a robotic system. It is the quality of this stage of design work that often determines the effectiveness of solving the problem. It should include subsystems for assessing the completeness and inconsistency of knowledge, assessing the quality of management and correcting knowledge.
Chronologically, the next stage after behavior planning can be the problem of issuing control commands to a robotic system in natural language. To create a natural language interface, in our opinion, the most appropriate implementation tool is the theory of fuzzy sets.
With the help of linguistic variables containing a certain, previously described term-set, a description of the subject area, a limited system of commands and objects that affect the robotic system and change under its action is made. The fuzzification and defuzzification methods used in this case, as well as fuzzy inference algorithms, have a significant impact on the accuracy of processing control actions and the speed of the robotic system.
And finally, the use of neural network control systems for robotic systems. The main advantage of a neural network is that there is no need to know or create a mathematical model of an object, since a neural network is a universal fuzzy approximator.
The object (robotic system) acts as a "black box". The neural network can act as a reference model of a controlled robotic system. It should be noted that this should be a learning multilayer neural network (object identifier). The neural network model is adjusted to the control object by mismatching the output signals of the object and the model. It also forms a training sample for adjusting and correcting the control device in accordance with the selected quality criterion.
The analysis carried out made it possible to synthesize the architecture of an intelligent control system for robotic systems, which is invariant with respect to the specifics of operation. The developed situational identification algorithm makes it possible to build highly informative sensory maps of the external environment. The main approaches to the formation of intelligent control systems for robotic systems are described. The directions of the prospective development of the most effective methods of artificial intelligence used to implement control devices are shown.