Annotation of the results obtained in 2020
A neural network based on the Hodgkin-Huxley biological neuron model was developed, capable of processing and classifying sensory information, while deciding on the type of input stimulus. The mechanism of operation of the proposed neural network is based on the formation of a chimera-like state when one part of the neurons is in an active state and generates spikes, while the other part is silent. The developed network consists of 100 Hodgkin-Huxley neurons, interconnected in accordance with a scale-free topology, and two output neurons. Initially, the main network is not connected to the output layer. In the process of training, these connections change over time, depending on the input signal, so that each of the output neurons is activated at a certain amplitude of the input signal, and the second remains at rest. Output neurons are interconnected by inhibitory connections. The developed network is able to determine the type of input pulse with close to 100% accuracy in a wide range of values of the amplitude of the external current pulse. Near the threshold value of the impulse, which corresponds to the average value of the amplitudes for which the training was carried out, the activation of both output neurons occurs equally, which corresponds to the ambiguity in determining the type of the presented stimulus and is in accordance with the known results of the experimental study of the perception of bistable images. An experimental radiophysical model of the central pattern generator (CPG) proposed at the previous stage of the project has been created. The constructed experimental prototype consists of an active van der Pol oscillator, which is connected to a chain of three linear dissipative oscillators. Maps of the oscillatory modes of the created CPG on the experimental planes of the control parameters have been constructed. The possibility of controlling the switching of the dynamic modes of the CPG, which are responsible for various types of locomotor activity, has been demonstrated. We have shown experimentally that when the frequency detuning of the self-sustained oscillator changes from the frequency of the three passive elements of the heterogeneous ensemble, hard switching occurs between the oscillatory modes. By shifting the oscillation phases of the ensemble elements, it is possible to control the gait of the robot by changing the values of the shift between the angles of rotation of the robot's limbs in the "hip", "knee" and "ankle" joints. The proposed CPG scheme has demonstrated its effectiveness in an experiment with a biped robot model for solving the problem of movement pattern generation that provides walking/running modes. We have shown experimentally that the developed CPG model demonstrates resistance to high noise level. It was found that switching of oscillatory modes in the investigated heterogeneous ensemble is observed even with a signal-to-noise ratio equal to -6 dB, which indicates the stability of the proposed CPG to noise and the prospects of its use in the tasks of controlling anthropomorphic robots. The characteristic features of the neural activity of the cerebral cortex and the trajectory of the gaze were revealed in the process of collecting unambiguous and ambiguous sensory visual information necessary for decision-making. We have shown that with ambiguity in sensory data, an increase in the energy of neural activity in theta and beta ranges is observed in the first 0.2 s after the presentation of a visual stimulus. The first point indicates that the subjects relied on internal representations (memory and contextual thinking), while an increase in energy in the beta range characterizes the activation of the mechanism for processing ambiguity and identifying relevant sensory data required for decision making. Analysis of the trajectory of eye movement using an Eye-Tracker showed that each orientation of the cube has its own characteristic pattern. The degree of ambiguity also affects the gaze trajectory. Thus, in the case of a highly ambiguous stimulus, the subject needs more sensory information to make a decision, which entails frequent switching of observation points. We have proposed an approach for the synthesis of neuro-fuzzy controllers using a modified genetic algorithm and an adaptive neuro-fuzzy inference system. The indicated approach is applied to the construction of Pareto-optimal controls for manipulator movements in an environment with static constraints. It is assumed that the manipulator also has limited mobility in the joints, which, in turn, imposes restrictions on possible movements. Suboptimal controllers were synthesized that minimize both the maximum possible deviations from the desired trajectory of movement and the maximum possible control moments created in the joints of the manipulator. The correctness of the developed controllers was tested on the model, and a comparison was made with the controllers built in the classical way using the semi-definite programming apparatus for the system obtained by linearizing the original along the desired trajectory of motion. The spatial-temporal structure of changes in the level of complexity of the signal of electrical activity of the human brain in the mu-rhythm range (8-14 Hz) was studied using recurrent analysis during the executions of the motor function. It was shown that the execution of hand movements is associated with a statistically significant increase in the regularity of the signal, localized in the region of the motor cortex with the dominant role of the contralateral hemisphere. A hypothesis was formulated that an increase in the regularity of the EEG signal can be interpreted from the standpoint of adaptation of the dynamics of local neuronal populations of the motor cortex to the performance of a motor task under the action of the mechanism of neural plasticity. As a consequence, we have shown that the regularity of fluctuations between assignments, i.e. at pre-stimulus temporal intervals, demonstrates statistically significant growth during the experiment. This may indicate healthy mechanisms for training the motor functions of the brain and adaptation of neural ensembles for more optimal activation during sequentially repeated movements. A method for assessing the complexity and long-term correlations in EEG signals associated with the execution of movements, based on the detrended fluctuation analysis of time series, is proposed. We have shown that movement of the non-dominant (left) hand determines relatively weak changes in the long-range correlations of EEG signals, as compared to the dominant (right) hand. A real-time control system for a two-legged walking robot has been developed without using a copying exoskeleton by analyzing the neural activity of the operator using the methods of recurrence quantitative analysis developed within the framework of the project. The classifier of EEG signals associated with the execution of arm and leg movements is based on a quantitative analysis of recurrent diagrams. The time dependence of the measure of determinism (DET), which has the property of contralaterality, was used as a criterion for classification. The main idea of the proposed classifier is that each type of movement is characterized by its own unique DET pattern. The fast speed of classification is ensured through the use of parallel computing technologies and the corresponding optimization of the algorithm. Signals of the neural activity of the operator's brain in the process of performing movements, first undergo preliminary processing, and then are fed to the input of the classifier, which determines the type of perfect movement (movement with the left/right arm/leg). Based on the information received, a control command is generated for the walking robot. We have shown the presence of a muscle pattern that occurs during training a person while maintaining balance. The muscle pattern includes four muscle pairs: the tibialis anterior muscle on the right leg and the triceps muscle of the calf on the right leg, the tibialis muscle on the right leg and the triceps muscle of the calf on the left leg, the tibialis muscle on the right leg and the quadriceps muscle of the thigh on the right leg, the triceps muscle of the calf on the right leg and the triceps muscle of the calf on the left leg. The results of statistical analysis of behavioral characteristics show a significant increase between sessions in the duration of the longest equilibrium segment and the total duration of all equilibrium segments. This conclusion is confirmed by statistical analysis of muscle activity, which demonstrates that a significant change in the correlation coefficients between 4 pairs of muscles of the detected pattern occurs only between experimental sessions 1 and 2. We have found that the number of muscles involved in the activation pattern is limited. It was shown that the pattern of muscle activation is asymmetric, which confirms the ability of the posture control system to use asymmetric patterns to maintain balance. A mathematical model was constructed for the first time, based on mechanical principles, taking into account the mechanisms of training a person while maintaining balance, muscle control in which is carried out on the basis of neural controllers with feedback, including the deviation of the platform angle, speed, acceleration, displacement of the center of mass of the body, afferent nerve delays and proprioceptive factor. The use of the developed model made it possible to identify a limited set of correlation patterns between the muscles of the legs of a person during training, in which the interaction between the muscles increases during the long-term performance of complex actions associated with maintaining balance. Neural activity in the sensorimotor area of the cerebral cortex shows significant changes in the process of maintaining balance, accompanied by a general decrease in activity in the O1, O2, Oz, Pz, Cpz EEG channels. In the course of multivariate analysis of variance, it was revealed that significant differences are achieved in the high frequency range of 23.75-40 Hz in the right and left parietal areas and in the central frontal area. It should be noted that this difference is achieved at the beginning of the session (200–240 seconds), while the energy in this cluster is maximum in the first session and decreases from session to session. On September 7-9, 2020, the IV International Scientific School "Dynamics of Complex Networks and their Application in Intellectual Robotics" (DCNAIR'2020) (https://events.innopolis.university/dcnair2020) was held at Innopolis University, in which 6 Russian and 9 foreign leading scientists (lecturers), as well as 75 participants - Russian and foreign young scientists under the age of 35 took part Selected DCNAIR 2020 papers was submitted in the IEEE Explore Library and indexed in Scopus.

 

Publications

---

Annotation of the results obtained in 2017
A model network of phase oscillators was developed, which topology is characterized by two scales of connectivity - the existence of a structure of coupled sub-networks. We have studied how the relationship between topologies of structural scales effects the dynamic regimes, in particular, on the implementation of such specific regimes as the chimera state. It is shown that a change in the connectivity of subnets allows to control the relation of the size of synchronous and asynchronous cluster on a global scale. The results obtained within the framework of this task will be used to develop a multiscale network of neural elements in the context of creating a central pattern generator. We numerically model one-layer and multilayer network consisting of Rulkov neurons coupled to each other by synaptical coupling. Couplings inside each layer are excitatory, but between layers a part of couplings are inhibitory and another one is excitatory. We discover the phenomenon of formation of space-time structures and investigate the influence of such parameters as network size, number of stimulated neurons, part of inhibitory couplings between layers and amplitudes of external stimulus and internal noise on it. We discover the phenomenon of coherent resonance depending from the parameters. A new mechanism of structural adaptation in the network of the Rulkov maps is developed on the basis of the spike-timing-dependent plasticity principle. It is shown that the developed model demonstrates a much more complex dynamics in comparison with the classical STPD mechanism. It was revealed that this difference is directly related to the type of structures formed in the process of adaptation: the implementation of an additional homeostatic term in the coupling equation led to the emergence of highly heterogeneous clusters in the structure and the appearance of the scale-free type of structure inherent in real neural ensembles of the brain. These results open the possibility of creating a central pattern generator, the characteristics of which adapt to the external conditions. Preliminary experimental studies are carried out with the registration of human neuronal activity during active movement to control an anthropomorphic hand prosthesis. 56 experimental measurements were performed with conditionally healthy volunteers. Various types of data analysis were carried out: spectral analysis with the help of wavelet spectra, cross-spectral analysis, estimation of phase coherence coefficients, which allowed to pre-select promising pairs of EEG signals for more detailed analysis, as well as frequency ranges of analysis. After choosing the parameters of the methods of analysis, a study of the directional coupling was carried out which made it possible to reveal the significant difference between symmetric EEG leads in opposite directions from the background values for both real and imaginary motions. This indicates the prospect of using the methods of the directional coupling analysis based on the simulation of phase dynamics for solving the problems of constructing the BCI. Using multifractal analysis, the structure of EEG signals recorded in untrained operators during motor activity was studied. On the basis of the obtained results, brain regions are identified for which reliable detection of differences between EEG signals acquired during hand movements and background brain electrical activity can be provided. It is shown that the highest efficiency of classification is achieved by using EEG recordings acquired in the frontal regions of the brain. The technique based on artificial neural networks for recognition and classification of EEG patterns corresponding to different types of movements has been developed, it has demonstrated high efficiency for untrained subjects: recognition accuracy up to 90-95%. Radial basis function network shows the best recognition results. Pre-filtering of the input EEG data using the low-pass filter significantly increases the recognition accuracy (on average by 10–20%), and the low-pass filter with a cutoff frequency 4 Hz shows the best results. We have demonstrated that when using signals from certain groups of electrodes consisting of 6-12 channels, the classification accuracy reaches a value close to the maximum. The last result is important from a practical point of view, because it shows the possibility of using more compact systems for EEG signals registration (with fewer electrodes) while maintaining the required recognition accuracy. First School for Young Scientists “Dynamics of Complex Networks and their Application in Intellectual Robotics” (DCNAIR 2017) was held on 20-22 November 2017 in Yuri Gagarin State Technical University of Saratov (http://www.sstu.ru/nauka/konferentsii/dinamika-slozhnykh-setey-i-ikh-primenenie-v-intellektualnoy-robototekhnike-dcnair-2017.html). DCNAIR 2017 has become an international platform for teaching and exchanging educational, scientific and technical ideas among specialists, especially young scientists and students, working in the field of studying complex networks and intelligent robotics. DCNAIR 2017 promoted active scientific cooperation, both at the Russian and international levels (http://www.sstu.ru/news/uchenye-predstavili-rezultaty-issledovaniy-na-mezhdunarodnoy-ploshchadke-po-intellektualnoy-robotote.html).

 

Publications

---

Annotation of the results obtained in 2018
For the first time, a system consisting of an active oscillator that demonstrates quasi-harmonic self-oscillations (a model of the Van der Pol oscillator) and two passive linear dissipative oscillators coupled with it is proposed as a model of a central rhythm generator (CRG). The signals taken from the generator and the contour are used as signals controlling the movements of the legs in the hip and knee joints of the robot. The CRG regime determines the speed of movement of the robot, gait and the regime of "running" or "walking." A mathematical model of a CRG was developed, and with its use a study of the dynamics of the CRG was performed. It was found that the movement along a part of the plane of the parameters of an external action, corresponding to synchronization through suppression, allows one to realize a soft transition from one oscillation regime to another, which corresponds to the running/walking switch. A numerical and experimental study of the dynamics of a network consisting of delay-coupled neural-like FitzHugh-Nagumo oscillators was carried out. A radiophysical setup was created for experimental research of possibility of controlling oscillatory regimes in an ensemble of neural-like oscillators coupled by tuned connections. It is shown that the use of an adaptively controlled delayed coupling allows one to solve the problem of in-phase synchronization of nonidentical neural-like oscillators in a wide range of control parameters. For an experimental study of a soft-excited self-sustained oscillator modeled by the Van der Pol equation, a radiophysical setup was developed and constructed, which includes a generator with cubic nonlinearity and two additional oscillatory circuits connected to the generator circuit via capacitive coupling. The dependence of the amplitude of oscillations in the main circuit on the capacitance of the capacitor in the main generator, which is responsible for the detuning of the frequencies of oscillator and additional circuits, is experimentally investigated. It is shown that by varying the frequency of self-sustained oscillations of the main generator, one can control the multistable oscillatory regimes in the system and observe hard transitions between them. In the course of this work, the effects of synchronization and the formation of spatio-temporal structures in multilayer networks of phase and neuron-like oscillators were numerically studied. A new phenomenon - the macroscopic chimera-like state of the network - was discovered and studied in detail. This effect consists in dividing the initially identical layers of the network into subgroups that demonstrate different dynamic modes under their interaction on the macro-level. Such states may be associated with the state of doubt of the brain neural network during the making of ambiguous decisions. Analysis of the neural activity of the human brain was conducted during the observation of bistable images and the decision on their interpretation. The magnetoencephalographic experiment data was processed using an artificial neural network trained to classify the states of perception of low ambiguous visual stimuli. Objective criteria were introduced to describe the process of information processing and decision making in a situation with ambiguity of choice. Using these criteria, an algorithm is proposed for an intelligent decision-making system based on the processing of external sensory information. According to the results of the experimental work, according to the developed design, an array of homogeneous experimental data was formed, containing sets of multichannel electroencephalography and myographies recorded from 20 conditionally healthy subjects. A unified algorithm for processing complex arrays of electroencephalographic (EEG) and electromyographic (EMG) signals has been developed. The processing complex is based on the allocation of a person’s motor activity and its objective classification according to myographic signals. It is shown that the use of methods of energy and skeleton estimations of continuous wavelet transform allows to establish an exact connection between the occurrence of muscle activity (increase in the amplitude of EMG) and EEG signals in the sensorimotor recording area. The use of the method of multiscale analysis based on discrete wavelet transform allows to reliably distinguish the signals of the electrical activity of the brain when performing movements. To highlight the general patterns inherent in the vertical posture of the subjects, energy assessments were made of the main neurophysiological modes of the oscillatory activity of the human brain in the moments of the implementation of various types of movements. To minimize the disturbance components caused by involuntary tension of the cervical and shoulder muscle groups, a special algorithm has been proposed, which includes separate recording of signals from several electrodes located on the back of the muscles, and the subsequent activation of filtering based on the calculation of empirical Hilbert-Huang transform modes at times of enhancement the amplitude level of high-frequency activity in the recorded myographic signals, leading to a specific distortion of brain activity. It was revealed that the transition to the vertical position of the subject, close to the natural, causes a significant complication of the frequency structure of the dynamics of the electrical activity of the brain, making the generation of high-frequency oscillations characteristic of the sensorimotor cortex in the classical neurophysiological samples less pronounced. We have proposed methods for control an anthropomorphic prosthetic arm using neural activity signals and muscle activity via direct decoding of biological signals without a copying exoskeleton. It has been shown that movement of the arm causes specific pattern formation on the signal of the corresponding muscle electrical activity, the shape of which keeps similar from one move to other. This allows us to automatically detect the exact moment when the movement starts using preliminary specified threshold. We have shown, that the optimal threshold value allows to correctly detect 100% of movements with 10% false detection rate. It has been shown that the characteristic motor-related patterns of brain electrical activity can be detected with an accuracy of 92% and with 5% false detection rate. When movements are performed under conditions of additional external stimulation, the frontal brain areas exhibit activity increase. It is associated with an attention on the external stimulus. Less than a second before the beginning of the movement and during its execution, both in the case of voluntary motor activity and in the case of additional external stimuli, there is a local decrease in the activity of 8-12 Hz in the left part of the central-frontal cortex (for movement with the right hand). At the same time, in the case of the presence of an additional external stimulus, there is a persistence of activity in the frontal areas and suppression of the activity of 8-12 Hz in the occipital cortex, which also indicates the concentration of attention during sensory information processing. We have proposed experimental design for analysis of electrical signals of brain (EEG) and muscle activity (EMG) during anthropomorphic robot control using copying exoskeleton. Spatial and frequency properties of neuronal activity patterns have been revealed using artificial neural network (ANN) and time-frequency methods during motor commands generation for controlling movements of robot legs. It has been shown that the most accurate classification of motor-related patterns can be reached by using EEG signals in frontal and parietal lobes. Moreover, the most informative frequency range is shown to be 1-12 Hz. Unique and common features of EEG signals, which are associated with control of arms and legs movements, have been analyzed using ANN. It has been shown, that ANN effectively (average accuracy 95%) detects the differences between these patterns when high-frequency EEG component is considered (>30 Hz). In the case when the low-frequency components (<4 Hz) are taken into account, the average classification accuracy decreases to 70%. The intellectual system for anthropomorphic robot control has been developed on a basis of correlated processes of brain and muscle activity. In order to detect EEG motor patterns and find correlation between them and corresponding patterns of EMG, the recurrent ANN has been used. The average recognition accuracy of motor commands is shown to be 90%. Second School for Young Scientists “Dynamics of Complex Networks and their Application in Intellectual Robotics” (DCNAIR 2018) was held on 8-10 October 2018 in Yuri Gagarin State Technical University of Saratov (http://www.sstu.ru/nauka/konferentsii/dcnair-2018-ru.html). DCNAIR 2018 has become an international platform for teaching and exchanging educational, scientific and technical ideas among specialists, especially young scientists and students, working in the field of studying complex networks and intelligent robotics. DCNAIR 2018 promoted active scientific cooperation, both at the Russian and international levels (http://www.sstu.ru/news/v-sgtu-proshla-shkola-konferentsiya-po-intellektualnoy-robototekhnike.html).

 

Publications

---

Annotation of the results obtained in 2019
A neural network of 100 Hodgkin-Huxley neurons has been proposed and implemented. The network demonstrates behavior similar to the behavior of the human brain during decision making in conditions characterized by a lack or ambiguity of information. The operative mechanism of the proposed neural network is based on a chimeric-like state formation, when one part of the neurons is in the active state and generates spikes, while the second part is inactive. The possibility of controlling this state using an external current pulse is shown. This feature allows classifying various states. In the process of analyzing the operation of the network, it is shown that in conditions with a low degree of ambiguity, the neural network unambiguously makes decisions. With increasing ambiguity, it begins to “doubt”, but is more likely to make a decision in the direction of the orientation of the cube with a greater intensity of faces. With maximum ambiguity, competition between output neurons with equal probability can lead to the activation of each of them at the end of an ambiguous stimulus presentation. The role of the internal noise influence in the coding network of sensory information on the decision-making process was discovered: on the one hand, the presence of noise extends the boundaries of the ambiguity domain in decision-making, but on the other hand, it speeds up the decision-making process in the case of uniquely interpreted sensory information. Thus, the effects of stochastic and coherent resonance when adding noise to the original sensor signal can have a positive effect on the decision-making process in an artificial intelligent system. Based on the results of experimental studies, it has been suggested that the search for a solution to the problem of classifying ambiguous information can lead to an increase in the dimension of the feature space. From the point of view of an intelligent system, this requires a multilayer convolution of the input information signal to extract several levels of features. A mathematical model based on the interaction between two networks of neuron-like Hodgkin-Huxley elements was constructed. The model describes the process of perception and processing of sensory information and decision making via analysis of experimental data on brain electrical activity using network theory methods. One network preprocesses and encodes incoming sensory information into a sequence of neural spikes, the other network is involved in processing this signal. It is shown that a change in image contrast corresponds with a change in the degree of correlation of the recorded signals of brain activity is observed. The highest correlation is achieved in the image contrast range of 0.4-0.7. An analysis of functional connections of the experimental data on the brain electrical activity was performed that for this range of contrasts. This analysis demonstrates the characteristic formation of a distributed network structure of functional connections involving neural populations of the prefrontal and parietal regions of the cerebral cortex. A similar local increase in the correlation of neuron activity is achieved due to the influence of internal noise and is possible only at certain values of its intensity (coherent resonance effect). An original approach is proposed for the experimental study of an ensemble of neural-like oscillators, in which the radio-engineering installation provides the possibility of setting both constant and adaptive interaction between generators. This connection can be of any kind, and the change of connections inside the ensemble occurs in real time. Using an example of a system of two mutually connected non-identical oscillators and a circle consisting of 10 identical Fitzhugh-Nagumo oscillators, it is shown that the introduction of adaptive delayed coupling can provide a phase synchronization of the oscillators. This phase synchronization occurs even in the case of a large detuning of their parameters and in cases where a constant coupling cannot provide phase synchronization of all non-identical elements of the ensemble even with large values of the coupling coefficients. The experimental results are in good agreement with the results of numerical studies. It is shown that the adaptive connection of oscillators is promising in the development of CPG responsible for motion control. To implement various types of robot gait and switch between them, it is necessary to provide different types of synchronization between the ensemble oscillators, differing in the magnitude of the phase shift of their oscillations. A model of the central pattern generator (CPG) is proposed in the form of a self-excited oscillator with soft excitation, which is modeled by the Van der Pol equation and has two additional oscillatory circuits connected with the generator circuit via capacitive coupling. It is shown that this model is promising for the generation of gait of an anthropomorphic robot compatible of switching the gait mode. As a result of a detailed bifurcation analysis, it was found that applying an external harmonic effect to the developed system of the CPG in the coexistence mode of two oscillatory modes with different phase ratios allows switching between limiting cycles existing in the phase space of the system. The discovered structure of the space of external influence parameters allows switching between oscillatory modes in a multistable system by applying an external harmonic force to it and controlling its parameters. The created model was reproduced in the radiophysical experiment with the structure of the plane of the parameters corresponding with the mathematical model of the CPG. During the experiment, it was also established that the vibrational mode switching is observed in the system even with a signal-to-noise ratio of 0 dB. A CPG model is proposed that implements switching between the four types of locomotion. The model is based on an ensemble of heterogeneous oscillators. Van der Pol oscillator was chosen as an active oscillator, and the other elements of the ensemble were three linear dissipative oscillators consisting of inductor and capacitor. The elements of the ensemble were connected in a chain through capacitors of the same denomination. The phase ratio between the oscillations of the active and three passive oscillators corresponds to the ratio between the angles of rotation of the robot limbs in the "hip", "knee" and "ankle" joints. Changing the magnitude of the oscillators frequency detuning between two elements of the heterogeneous ensemble leads to a hard switch between the four oscillatory modes. The observed modes are stable and resistant to variations of other parameters, and can be used to generate various types of locomotion. A new model of walking primitives generator is also proposed. The model is based on the use of virtual holonomic connections using a motion generator that does not include an explicit function of the generalized coordinates. A method of stabilizing motion is shown. The method is based on reducing the system to a zero submanifold due to feedback on the residual of virtual connections and their speeds. A database was created consisting of neural and muscle activity signals of a person, corresponding to the persons execution of motor commands (both real and imaginary). The database contains the original EMG, EEG signals, as well as the time-frequency characteristics calculated for them. EEG and EMG signals were filtered using a 5th order Butterworth bandpass filter in the ranges 1-100 Hz and 10-100 Hz. The EEG trials were filtered in the ranges of mu- and beta-bands, which are the most informative in the analysis of the electrical activity of the brain, corresponding to real and imaginary movements. In addition, the database contains the time-frequency characteristics of each trial, such as the wavelet surfaces of EEG trials and the averaged wavelet energies of mu and beta rhythms. A control system for a two-legged walking robot without the use of a copying exoskeleton was developed using an analysis of the human neural activity. The methods were developed for the quantitative analysis of recurrence diagrams. Signals were recorded with an electroencephalograph from the operator in the process of performing movements. The signals were pre-processed to reduce noise and remove artifacts. Then signals were given to the classifier, which determines the type of perfect movement (movement with the left / right hand / foot). Based on the information received, a control command was generated for the walking robot. As a criterion for classification, we used the time dependence of the determinism measure (determinism, DET), which has a pronounced contralaterality property. The interaction between the blocks of the server part of the robot control interface is implemented using the execution queue, in which the execution commands are placed. An intelligent system was created, trained using EMG and EEG signals. The system automatically maintains the balance of a walking robot when the surface tilt varies over time. At the first stage, experiments were conducted to record the neural and muscle activity of a person while maintaining equilibrium on a balance platform. It is shown that during maintaining equilibrium on the balance platform, there is a change in the distribution of the values of the angle of the balance platform from the equilibrium position in each of the three experimental sessions. In the last session, the distribution maximum shifts to the region of small deviations, which indicates an increase in the time the platform is in equilibrium. This indicates a person’s adaptation to changing conditions. Studies have been conducted with an anthropomorphic walking robot. It should be noted that for particular parameters of the experiment, the behavior of the robot in a mathematical modeling environment is close enough to human behavior in full-scale experiments with a balance platform. It was shown that the implementation of feedback on the position of the center of mass is not enough to obtain the peculiarities of the robot’s movement close to that observed in experiments with subjects, while the introduction of an additional feedback channel on the balance-platform orientation made it possible to obtain results that are qualitatively similar to experimental data. On September 9-11, 2019, the III International School of Young Scientists “Dynamics of Complex Networks and Their Application in Intellectual Robotics” (DCNAIR 2019) was held at Innopolis University, in which 5 Russian students took part and 8 foreign academic lecturers, as well as 82 students - Russian young scientists under the age of 35 inclusive (https://lomonosov-msu.ru/rus/event/5553/). DCNAIR 2019 selected content is hosted on the IEEE Explore and indexed on Scopus and Web of Science.

 

Publications

---