JP2020121376A - Control device, control system and control program - Google Patents

Abstract

To provide a control device which autonomously reacts even if an unintentional change of an external environment occurs, and makes a control object stably perform a desired operation, a control system and a control program.SOLUTION: This control device can control a control object by supplying a drive signal to the control object, and comprises a spike signal row creation part and a drive signal creation part. The spike signal row creation part can create an internal state including a fundamental control signal for controlling the control object and a disturbance, and a spike signal row at timing which is defined by dynamics related to the internal state. The drive signal creation part can create the drive signal which is continuously changed in a time series on the basis of the spike signal row.SELECTED DRAWING: Figure 1

Description

The present invention relates to a control device, a control system, and a control program in a system involving complicated and dynamic state changes such as motion/control of robots and moving systems for industrial use, medical use, and home use, a manufacturing plant, and the like.

Industrial and commercial industries such as agriculture, medical fields such as surgery and nursing/nursing care, and households such as cleaning robots and industrial machines are becoming more complex and highly functional. The components of these devices such as robots and industrial machines are not uniform, and the work target and operating environment are not always constant.

Patent Document 1 is proposed as a bipedal walking robot that generates a motion pattern for each individual by repeatedly performing learning in a device to which a neuron network (neural network) is applied. In Patent Document 1, a neuron network is provided as described in claim 1, and as described in FIG. 3 and [0004], it is attempted to improve design accuracy and shorten design time by iterative learning.

JP, 2006-88331, A

The neuron network used in Patent Document 1 generates an output signal by applying a unique weighting coefficient W_k_* to a plurality of inputs, as described in FIG. 11 and paragraphs [0007] to [0008]. There is. This weighting coefficient W_k_* is optimized by iterative learning. Therefore, although the configuration of each device is fixed and the environment can be optimized under a constant environment, it cannot follow an unexpected change in the environment due to disturbance or the like.

The present invention has been made in view of such circumstances, and makes it possible for a controlled object to stably perform a desired operation by reacting autonomously even when an unexpected external environment change occurs. An object is to provide a control device, a control system, and a control program.

According to the present invention, the control device is configured to control the control target by supplying a drive signal to the control target, and includes a spike signal train generation unit and a drive signal generation unit, The spike signal train generation unit is configured to be able to generate a spike signal train at an internal state including a basic control signal and a disturbance for controlling the controlled object, and at a timing defined by dynamics related to the internal state, and the drive signal A control device is provided in which the generation unit is configured to be capable of generating the drive signal that continuously changes in time series based on the spike signal train.

In the control device according to the present invention, the basic control signal is once converted into the spike signal train by the spike signal train generation unit, and then the control target is controlled using the drive signal generated by the drive signal generation unit. .. At this time, even if an unexpected change in the external environment occurs, the control device side reacts autonomously, and the entire control system can perform a desired operation, which is an advantageous effect.

1 is a functional block diagram of a control system including a control device and a control target according to an embodiment of the present invention. The optimization control flow chart in a control apparatus. The block diagram which controls the particle position on a horizontal axis as a control example using a spike signal train. FIG. 3 is a state diagram showing a triple/double/single well potential for particles moving on a horizontal axis. Of the order emergence function, the simulation result diagram regarding the entropy reduction/pattern formation function. Of the order emergence function, the simulation result diagram regarding the function of expanding the pull-in area to the target state. Of the order emergence function, the simulation result diagram regarding the binding function to the natural frequency. The block diagram of the musculoskeletal robot control system which is an example of a control system. FIG. 6 is a diagram showing a moving speed simulation result of a musculoskeletal robot control system in a low friction environment. FIG. 6 is a diagram showing a result of a simulation of a cooperative movement ability of a musculoskeletal robot control system in a low friction environment. FIG. 3 is a functional block diagram of a control system in which the spike signal train generation unit and the drive signal generation unit are external control devices.

Embodiments of the present invention will be described below with reference to the drawings. The various features shown in the embodiments described below can be combined with each other. In particular, in the present specification, the “unit” may include, for example, a combination of hardware resources implemented by a circuit in a broad sense and information processing of software that can be specifically realized by these hardware resources. .. In addition, although various kinds of information are handled in the present embodiment, these pieces of information are digital signal information represented by high and low of a signal value as a binary bit aggregate composed of 0 or 1, and voltage and current are continuous. Communication and calculation can be performed on a circuit in a broad sense by analog signal information that changes dynamically and spike signal information that instantaneously generates voltage and current on the time axis.

Further, the circuit in a broad sense at least appropriately includes a digital circuit (Digital Circuit), an analog circuit (Analog Circuit), an optical circuit (Optical Circuit), circuits (Circuitry), a processor (Processor), a memory (Memory) and the like. It is a circuit realized by combining them. That is, as a digital circuit, an application specific integrated circuit (ASIC), a programmable logic device (for example, a simple programmable logic device (Simple Programmable Logic Device: SPLD), a complex programmable logic device (Complex Logic Program)). CPLD), a field programmable gate array (Field Programmable Gate Array: FPGA), and the like. Examples of the analog circuit include resistors, capacitors (capacitors), passive elements (passive components) such as inductors, diodes (diodes), transistors (discrete semiconductors) such as thyristors (discrete semiconductors), and comparators. (Comparator) and other analog integrated circuits (Analog Integrated Circuit) and the like. A circuit configuration using a D/A converter (Digital-to-Analog Converter) or an A/D converter (Analog-to-Digital Converter) is also possible at the boundary between the digital circuit and the analog circuit. Further, as an optical circuit, a light emitting diode (Light Emitting Diode), a light emitting element (Light Emitter) such as a semiconductor laser (Semiconductor Laser), a light receiving element (Photodetector) such as a photodiode (Photodiode), and an optical fiber (Optical Optical) such as an optical fiber. It includes a waveguide (Optical Waveguide), an optical integrated circuit (Optical Integrated Circuit), and the like.

1. Overall Configuration In Section 1, the overall configuration of a control system 1 including a control device according to the present invention will be described with reference to the drawings. FIG. 1 is a diagram showing a schematic configuration of a control system 1 according to the present embodiment. The control system 1 is a system that includes a control device 2 and a controlled object 3 and that are electrically connected to each other. The controlled object 3 is a robot such as a bipedal locomotive (described later), a moving body, a pacemaker, an electric circuit system, a chemical reaction system, a communication network, a socioeconomic management system, a financial system, a biological network, an animal and plant environment, and the surrounding environment with respect to movement and condition It has a characteristic that the electrical, mechanical or chemical internal state changes due to the change, and requires control in order to perform a desired operation. Items are not limited as long as the internal state is related to the function or operation of the control system 1 and can be detected.

1.1 Control device 2
As shown in FIG. 1, the control device 2 has a communication unit 21, a storage unit 22, and a control unit 23, and these components are electrically connected to the communication bus 20 inside the control device 2. Hereinafter, each component will be further described.

<Communication unit 21>
The communication unit 21 exchanges information with the controlled object 3. Wired communication means such as USB, IEEE 1394, Thunderbolt, and wired LAN network communication are preferable, but wireless LAN network communication, mobile communication such as 5G/LTE/3G, and Bluetooth (registered trademark) communication are included as necessary. Good. These are examples, and a dedicated communication standard may be adopted. That is, it is more preferable to implement it as a set of a plurality of these communication means.

Although FIG. 1 shows a state in which the communication unit 21 is separately connected to the state detection unit 31 and the drive unit 30 in the controlled object 3, the physical connection is one and the inside of the controlled object 3 is shown. May be logically distributed in.

<Memory unit 22>
The storage unit 22 is a volatile or non-volatile storage medium that stores various information. This is, for example, as a storage device such as a solid state drive (SSD), or a random access memory (Random Access Memory) for storing temporarily necessary information (arguments, arrays, etc.) related to the calculation of a program. It may be implemented as a memory such as RAM). Also, a combination of these may be used.

In particular, the storage unit 22 stores various parameters related to control execution contents, individual characteristic information such as shape, size, material, and weight related to the controlled object 3, and past setting information at the time of continuous control including in the middle of optimization.

Further, the storage unit 22 stores various programs and the like related to the control device 2 executed by the control unit 23. Specifically, for example, an operation procedure regarding the controlled object 3 having a plurality of driving elements such as a plurality of muscles/tendons and joints like a bipedal walking robot, a basic control signal generation unit 231, a spike signal that constitutes the control unit 23, and the like. These are initial values and update procedures of parameter groups used in the column generation unit 232 and the drive signal generation unit 233.

<Control unit 23>
The control unit 23 processes and controls the overall operation related to the control device 2. The control unit 23 is, for example, a central processing unit (CPU) (not shown). The control unit 23 realizes various functions of the control device 2 by reading out a predetermined program stored in the storage unit 22. Specifically, based on the information given in advance for each controlled object 3 and the state information received from the controlled object 3 internal state detection unit 31 via the communication unit 21, the basic control signal generation unit 231 and the spike signal sequence generation unit. The function of generating the drive signal AS to the controlled object 3 through the drive signal generating unit 232 and the drive signal generating unit 233 and executing the control is applicable.

That is, the information processing by the software (stored in the storage unit 22) is specifically realized by the hardware (control unit 23), so that the basic control signal generation unit 231, the spike signal sequence generation unit 232, and the drive. It may be executed as the signal generator 233. In addition, in FIG. 1, it is described as a single control unit 23, but the actual configuration is not limited to this, and a plurality of control units 23 may be provided for each function. Also, a combination thereof may be used. Hereinafter, the basic control signal generator 231, the spike signal string generator 232, and the drive signal generator 233 will be described in more detail.

[Basic control signal generation unit 231]
The basic control signal generation unit 231 is one in which information processing by software (stored in the storage unit 22) is specifically realized by hardware (control unit 23). The basic control signal generation unit 231 is a non-spike signal based on the state information obtained from the state detection unit 31 of the controlled object 3 via the communication unit 21 and the parameters given in advance for each controlled object 3. The control signal CS is generated. The control algorithm is not limited, and various algorithms such as feedback control, feedforward control, model predictive control, and control using deep learning can be used.

In order to maximize the use of the ordered emergence function, which will be described later, it is desirable that the basic control signal generation unit 231 has a frequency characteristic of a control algorithm and parameter setting that does not have a strong natural frequency/natural frequency peak.

[Spike signal sequence generation unit 232]
The spike signal sequence generation unit 232 is one in which information processing by software (stored in the storage unit 22) is specifically realized by hardware (control unit 23), and the hardware is the digital circuit and analog described above. Composed of a combination of circuits.

The spike signal train generation unit 232 receives the basic control signal CS generated by the basic control signal generation unit 231, and includes a neuron (not shown in FIG. 1) that is an element that generates the spike signal train ST. .. The spike signal sequence generation unit 232 operates similarly to a neuron network that stochastically generates impulse-like action potentials in a living body, that is, a stochastic spiking neuron network (sSNN). As the neurons in the spike signal sequence generation unit 232, there are LIF (leaky integrate-and-fire) neurons, Poisson spike models, Hodgkin-Huxley models, burst ignitable models, and the like. It is possible to apply a model that generates a spike signal train at a timing defined by a basic control signal and an internal state including a disturbance, and dynamics related to the internal state.

When a plurality of neurons exist in the spike signal sequence generation unit 232 (sSNN), it is possible to have arbitrary synaptic connections between the neurons. At that time, all neurons are designed not to fire synchronously. Specifically, for example, each neuron receives an independent noise, or a different firing threshold (described later) or reset potential (described later) is set for each neuron.

Here, the case of the LIF neuron will be described in more detail using mathematical expressions. The j-th LIF neuron in the i-th spike signal sequence generation unit 232 (sSNN) can be expressed as [Equation 1] [Equation 2] [Equation 3] described below.



Here, v_ij is the potential, dots on the variables are time-differentiated, γ is the attenuation coefficient, b_ij is the bias input (current), I is the input signal, D_i is the noise intensity in the i-th spike signal train generation unit 232, and ξ_ij is Gaussian (normal distribution) noise of unit intensity, v^θ is a firing threshold, v_ij^R is a reset potential, τ_ref is an impossible period, and k is a spike signal generation sequence number.

When the internal potential v_ij of the target LIF neuron reaches v^θ, it fires and mathematically generates a spike signal described by the Dirac delta function δ, and the internal potential is reset to v_ij^R. Then, the τ_ref is disabled. In the LIF neuron shown in [Equation 1], a large number of LIFs are prevented from firing synchronously by adding a disturbance in the form of Gaussian noise. As is clear from [Equation 1], it is possible to prevent the synchronous firing by individually setting the firing threshold v^θ and the reset potential v_ij^R in addition to the noise component application. The spike signal train σ_ij in each LIF neuron can be formally described by [Equation 4].

At this time, if a plurality of LIF neurons are present in the i-th spike signal train generation unit 232, all spike signal trains output from each LIF neuron are utilized. At that time, a method of using the average of the spike signal train or a method of linearly weighting each LIF neuron can be applied. The new spike signal train thus obtained is used as the spike signal train ST which is the output of the spike signal train generator 232.

That is, the spike signal string generation unit 232 has a plurality of neurons that form a neuron network, and is configured to be able to generate the spike signal string ST based on the signals output by each of the plurality of neurons.

[Drive signal generator 233]
The drive signal generation unit 233 is one in which information processing by software (stored in the storage unit 22) is specifically realized by hardware (control unit 23), and the hardware is the digital circuit and analog circuit described above. It is composed of a combination of.

The drive signal generation unit 233 has a function of converting the spike signal sequence ST generated by the spike signal sequence generation unit 232 into a drive signal to be supplied to the drive unit 30 of the controlled object 3. This drive signal is generated based on the post-synaptic potential (PSP) in the biological neuron network.

A method similar to the synapse can be used to generate the drive signal AS corresponding to the post-synaptic potential from the spike signal train ST. Specifically, it is a low-pass filter, a classical α-functional synapse model, a square wave synapse model, a dynamic synapse model, and the like.

When the signal processing method in the drive signal generation unit 233 is a low-pass filter, the post-synaptic potential (PSP) y_i that is the basis of the output and the activity A_i of the controlled target 3 drive unit 30 are respectively [Equation 5][ Equation 6] can be used.


Here, τ_s is a synapse time constant, N is the number of LIF neurons in the target i-th spike signal sequence generation unit 232, g_i^A is an amplification gain, and A_i^0 is an offset.

Although the drive signal AS is transmitted from the drive signal generation unit 233 to the drive unit 30 by a direct connection in FIG. 1, after the A/D (Analog-to-Digital) conversion is performed, the communication unit 21 outputs a digital signal. It is also possible to adopt a configuration in which the signal is transmitted to the drive unit 30 via.

1.2 Control target 3
The controlled object 3 is specifically, for example, a robot or a moving system that executes a mechanical task. Note that the control device 2 of the present invention does not limit the technical field of the task, but can be an electric circuit system or a chemical reaction system as long as it is a dynamically acting one and can detect a change in operating state due to environmental changes. The system can also be the controlled object 3.

[Drive unit 30]
The drive unit 30 operates the controlled object 3 based on a control signal from the outside when executing a task. Specifically, for example, when the controlled object 3 is a robot, it is a motor, an actuator such as pneumatic/hydraulic, or the like, but is not limited to these.

[State detection unit 31]
The state detection unit 31 detects the internal state of the controlled object 3 including the case where the controlled object 3 changes in environment due to disturbance during operation. The internal state includes the position of a point of interest in the controlled object 3, velocity, acceleration, rotation, angular velocity, angular acceleration, mechanical mechanical information such as force and moment, electrical information such as voltage, current and resistance, and sound and It is, but not limited to, physical information of light, fluid dynamics information such as temperature, pressure and flow velocity, and chemical information such as concentration, pH and molecular weight. The detected internal state can be transmitted to the communication unit 21 as state information. That is, in other words, the state information is information indicating an internal state that changes due to the behavior of the control target and environmental changes at the point of interest in the control target.

2. Optimization Method of Control System 1 Section 2 describes a parameter optimization method of the control device 2 in the control system 1. Here, for example, when the controlled object 3 is a bipedal walking robot, a task of moving to a designated point is executed, but the optimization method of the present invention depends on the controlled object 3 and the type of the task. It is not limited.

FIG. 2 shows an optimization flow for optimizing the parameters of the control device 2 by repeatedly executing the basic tasks a plurality of times in an environment with little disturbance.

[Start optimization]
(Step S1)
Each parameter group in the basic control signal generation unit 231, the spike signal sequence generation unit 232, and the drive signal generation unit 233 is initialized. Information stored in the storage unit 22 can be used as the parameter value used for the initialization. The information stored in the storage unit 22 is not only the information continuously stored in a non-volatile manner, but also information input by the user from the outside in consideration of the individual characteristics of the controlled object 3 and the external environmental condition at the start of work. Including.

The basic control signal generation unit 231 initializes the parameters used in the adopted algorithm. Specifically, for example, in PID (Proportional-Integral-Differential) control, which is one type of feedback control, a proportional gain K_P, an integral gain K_I, a differential gain K_D, and the like are used.

The spike signal sequence generation unit 232 initializes all parameters as a stochastic spiking neuron network (sSNN). For example, when the spike signal sequence generation unit 232 is configured by LIF neurons, the number of neurons N and variables included in the [Formula 1] and [Formula 2], specifically, the attenuation coefficient γ, the bias input b_ij, the noise intensity D_i, and the firing The threshold v^θ, the reset potential v_ij^R, and the disabled period τ_ref.

The drive signal generation unit 233 initializes parameters relating to the method adopted to generate the drive signal AS from the spike signal train ST. For example, when the low-pass filter method is adopted, the time constant τ, the passband gain, etc. are set as filter characteristic values. When all or part of the drive signal generation unit 233 is configured as an electrical analog circuit, it is possible to set the resistance value of the resistor, the capacitance value of the capacitor, or the like fixedly or semi-fixedly in advance. ..

(Step S2)
The parameters in the basic control signal generation unit 231, the spike signal sequence generation unit 232, and the drive signal generation unit 233 are updated. Although all parameters can be updated, the general directionality of the control unit 23 as a whole largely depends on the basic control signal CS, and therefore it is desirable to mainly optimize the basic control signal generation unit 231. Regarding the spike signal sequence generation unit 232 and the drive signal generation unit 233, for example, a control method may be adopted in which the number of neurons N, the synapse time constant τ_s, and the noise intensity D are the update targets, and other parameters related to the synapse connection intensity are not updated. It is possible. Each time the parameter is determined not to converge in step S4 described below, the parameters are updated.

(Step S3)
The controlled object 3 executes a basic task under the control of the control device 2. The state detection unit 31 detects the above-described various internal states of the control target 3 and transmits the state information to the communication unit 21 in the control device 2. The basic control signal generation unit 231 calculates the evaluation value of the controlled object 3 as a whole. Every time it is determined in step S4 described later that the basic task is not converged, the number of times of execution of the basic task is increased to the first time, the second time, and the third time.

(Step S4)
The basic control signal generation unit 231 determines whether or not the entire system has converged. If it is determined that the values have not converged (NO), the process returns to step S2 to continue from the parameter updating work. As a learning algorithm at the time of updating parameters, an evolution strategy such as a genetic algorithm can be applied, but the learning algorithm is not limited thereto. If it is determined that they have converged (YES), the work ends.
[End of optimization]

3. Order emergence function In Section 3, the order emergence function of the control device 2 in the configuration of the present invention will be described in detail. This is to generate a spike signal train ST using the spike signal train generator 232 (sSNN) from the basic control signal CS generated by the basic control signal generator 231, and then drive the spike signal train ST with the drive signal generator 233. This is a function unique to the control device 2 of the present invention having a configuration for generating the signal AS, and is not conventionally known.

In this section, as an example showing the order emergence function, a case having two stochastic spiking neuron networks (sSNN) is shown in FIG. Here, the position of particles existing on the horizontal axis in FIG. 3 is controlled. In FIG. 3, S_0 and S_1 are sSNN, and include the spike signal train generation unit 232 and the drive signal generation unit 233 of the present invention. The two sSNNs receive I_0(t) and I_1(t) as input signals, respectively. Here, the input signal I_*(t) is defined as a difference amount between the current position x(t) of the particle and the target positions x_0^g and x_1^g. A LIF neuron (see Section 1) is used as the spike signal sequence generation unit 232, and a low-pass filter (see Section 1) is used as the drive signal generation unit 233.

3.1 Entropy reduction/pattern formation function Fig. 4A shows a case where the position of a particle having a mass in the triple well potential function is controlled by sSNN. It should be noted here that the center (x=0) is not the minimum value but the minimum value of the potential, and there are places where the potential is minimum on both sides of the center.

5A and 5B show the simulation results of the particle position movement state under the environment of FIG. 4A. The horizontal axis represents time t, and the vertical axis represents particle position x. Further, FIG. 5A shows the case where the number of neurons N=2, and FIG. 5B shows the case where the number of neurons N=150. ApEn is entropy (Approximate Entropy) calculated from the moving state. In FIG. 5A, in which the number of neurons is small, the absolute value of the amount of movement of particles is larger than that in FIG. 5B, but this is because the potential minimum position existing in two places is moved periodically, Due to such periodicity, the entropy ApEn has a small value. As described above, in the stochastic spiking neuron network (sSNN), the higher the spike property, the more the entropy reducing function and the pattern forming function are expressed.

3.2 Function of Enlarging Entrainment Area in Target State FIG. 4B shows a case where the position of a particle having a mass in the double well potential function is controlled by sSNN. Here, the center (x=0) is the maximum value of the potential, and as with the mountain climbing problem (mountain car task), the maximum value of the potential x=0 cannot be reached directly from the bottom of the valley, and there is a reaction. The problem setting requires the help of external force.

The initial position x_0 and the initial velocity v_0 of the particle were changed variously, and the case where the particle could stay within the range of the center [-0.1, 0.1] for a certain time or longer was defined as the pull-in area for simulation. The results are shown in FIGS. 6A and 6B. 6A shows the case where the number of neurons N=1, and FIG. 6B shows the case where the number of neurons N=100. White areas in FIGS. 6A and 6B are pull-in areas. Further, FIG. 6C shows a simulation result of a pull-in area ratio (basin rate) when the number N of neurons is changed with the bias input b as a parameter. As is clear from FIGS. 6A, 6B, and 6C, the stochastic spiking neuron network (sSNN) has a function of enlarging the attraction region, and the smaller the number of neurons included in the sSNN, the greater the attraction region. Often expressed strongly.

3.3 Binding Function to Natural Frequency FIG. 4C shows a case where the position of a particle having a mass in the spring-mass system is controlled by sSNN. Here, the spring-mass system is equal to the single well potential function. A spring-mass system in which no control is performed has a natural frequency f_0 (natural frequency) determined by the spring constant k and the mass m of particles. It is known that when the normal feedback control is performed on the spring mass system, the natural frequency f_0 is modulated due to the influence of the gain of the feedback control.
FIG. 7 shows the result of simulating the SNR (signal-to-noise ratio) with respect to the natural frequency f_0 by performing control by sSNN with the center position (x^g=0) as the target. In FIG. 7, the horizontal axis represents the synaptic time constant τ_s, the vertical axis represents the amplification gain g^A, and the whiter region indicates that the SNR is high. A resonance phenomenon to the natural frequency f_0 can be confirmed in a very wide parameter range where the natural frequency f_0 is 1 to 10 Hz. In addition, since the white stripe pattern extends in the vertical direction in many regions, when the spring-mass system is driven using the stochastic spiking neuron network (sSNN), there is almost no effect on the natural frequency f_0. It's obvious.

4. Robot Control System In Section 4, a simulation result of a bipedal robot control system using a robot as a control target 3, more specifically, a musculoskeletal robot will be described as an embodiment.

FIG. 8 shows a functional schematic diagram of the musculoskeletal robot control system used in the simulation. The left side of FIG. 8 shows a structure of a skeleton (link), a joint (joint), and a muscle (muscle line in the drawing, a part of which is omitted). As the robot driving unit 30, eight muscles and multi-joint muscles are provided for each leg. Are connected. Further, as the robot state detection unit 31 (Sensory Input in FIG. 8), a muscle generating force, a muscle length, a joint angle, a body posture, a center of gravity position, a sole reaction force, a 9-axis inertial measurement device for each bone (link) ( This is a configuration capable of measuring the triaxial posture, triaxial acceleration, and triaxial angular velocity obtained by the Inertia Measurement Unit).

Each of the left and right legs has a plurality of phases such as a stationary motion and a swing motion, and each phase has a different reflex activation rule. The reflex activation rule (Reflex System: basic control signal generation unit 231 in FIG. 8) is a combination of a positive feedback control rule for generated force, a feedback control rule for muscle length, and a PD (proportional derivative) control rule for joint angle or body posture. Built in.

FIG. 9A, FIG. 9B, and FIG. 9C show the results of center-of-gravity movement speed simulation in a slippery low-friction environment. 9A and 9B, the horizontal axis represents time t, and the vertical axis represents the center-of-gravity moving speed v^g. The friction coefficient μ is normally 10 (broken line in the figure), but in a low friction environment (solid line in the figure), the friction coefficient μ is set as low as 0.04 at time t=[10, 40]. As shown in FIGS. 9A and 9B, stable bipedal walking is performed in both the normal environment and the low friction environment. At that time, the low friction environment shifts to the lower speed side as a whole than the normal environment. FIG. 9C shows the relationship between the gait frequency and the amplitude, and transitions to the low frequency side in a low friction environment.

FIG. 10 shows the results of the cooperative motor performance simulation in the situation where a large slip has occurred. In FIGS. 10A and 10C, the horizontal axis x indicates the position, x=[4,16] is the low friction section, and the friction coefficient μ=0.04, and otherwise the friction coefficient μ=10. The sampling interval in FIG. 10A is 0.1 s, and the sampling interval in FIG. 10C is 0.25 s. On the low friction zone (slippery band in the figure), the black band indicates the slip of the right foot, and the gray band indicates the slip of the left foot. 10B and 10D show the relationship between the time t and the center-of-gravity moving speed v^g corresponding to FIGS. 10A and 10C, respectively.

In FIG. 10A and FIG. 10B, a slip of about 0.5 m occurs for a time of 0.5 s or more, but the walking can be adapted to this. At this time, it should be noted that the sliding distance between the left and right feet is asymmetric. Further, in FIGS. 10C and 10D, v^g is extremely decreased near x=16 m, t=13 s, which is the end point of the low friction section, which is a little longer from the right foot slip just before the end of the low friction section to the normal section. It indicates that the vehicle was about to fall when entering. Even in this situation, it is possible to return to normal walking by using the fact that the toes and the like are in contact with the non-slip ground, which is the normal section (μ=10), as a foothold. As described above, the robot control system using the control device 2 according to the present invention has a very high adaptability, which has been difficult to realize by the conventionally known control of only the reflection circuit.

For the coordinated operation in the bipedal robot control system described in this section, it is necessary to reduce the entropy of the motion sequence. Further, in order to avoid falling, it is necessary to control ZMP (zero-moment point) within a certain range. As described in Section 3, the ordered emergence function (see 3.1 Entropy reduction/pattern formation function, 3.2 Target state expansion region expansion function) of the control device 2 of the present invention works effectively. Therefore, it can be said that the instant fall avoidance function is realized.

5. Modification Note that the present embodiment may be further devised in the following manner.

In the fourth section, the embodiment of the bipedal walking robot control system has been described. However, in the mobile system, the external environment generally changes with the movement, and the characteristics of the control device 2 of the present invention having the order emergence function are described. You can take advantage of it. In addition, the order emergence function can be utilized either as an unmanned fully autonomous control system or as an auxiliary control system of a manned system. Specific examples of the moving system include a multi-legged walking robot, a wheel/caterpillar robot, an unmanned aircraft (Unmanned Aerial Vehicle: UAV, drone), an unmanned surface vehicle (USV), and an unmanned submersible (Unmanned). Underwater Vehicle (UUV), automobiles including autonomous driving, aircraft, ships, etc., but are not limited to these.

Furthermore, it is also applicable to industrial, medical, agricultural, and household robots whose work objects are frequently changed in transportation and processing.

The order emergence function of the control device 2 according to the present invention is not limited to the technical field as long as it is the control system 1 that needs to adapt to changes in the state. That is, it can be exerted not only for mechanical movement but also for electrical fluctuation or chemical reaction fluctuation. Furthermore, it can be expected to be applicable to control of financial systems, control of inflow/outflow/propagation of information in communication networks such as the Internet, and air conditioning systems. Therefore, for example, it can be applied to a cardiac pacemaker or an artificial heart-lung machine that utilizes the function of binding to natural frequencies (see 3.3) among the function of emergence of order.

As shown in FIG. 1, the spike signal train generation unit 232 and the drive signal generation unit 233 according to the present invention should be configured as one control device 2 together with the basic control signal generation unit 231, the communication unit 21, the storage unit 22, and the like. However, the spike signal train generation unit 232 and the drive signal generation unit 233 can be used as an external control device.

FIG. 11 shows a functional block diagram. In FIG. 11, 2b is a basic control device that generates a basic control signal CS, and 3 is a control target. Although the conventional control is possible only by combining the basic control device 2b and the controlled object 3, the external control device 2a is connected to supplement the conventional control. The external control device 2a is provided with a spike signal train generation unit 232 and a drive signal generation unit 233. By adding the external control device 2a to the existing control system, the order emergence function described in Section 3 can be utilized and the function/performance of the control system 1 can be improved.

Furthermore, spike signal sequences are not limited to those generated stochastically due to disturbance, and spike signal sequences generated deterministically using chaos, etc. if they have sufficient complexity and unpredictability. However, the same function can be obtained.

6. Conclusion As described above, according to the present embodiment, even if an unexpected external environment change occurs, the control device 2 that reacts autonomously and enables the entire control system 1 to perform a desired operation. Can be implemented.

The control device 2 is configured to control the control target 3 by supplying a drive signal to the control target 3, and includes a spike signal train generation unit 232 and a drive signal generation unit 233. The column generation unit 232 is configured to be able to generate the spike signal sequence ST at a timing defined by an internal state including a basic control signal CS for controlling the controlled object 3 and the disturbance, and the drive signal generation unit is The drive signal AS that continuously changes in time series is generated based on the spike signal train ST.

Moreover, the following control system 1 can be implemented by this.

The control system 1 includes a control target 3 and a control device 2 that controls the control target 3.
The control target 3 is at least one of a robot, a mobile body, a pacemaker, an electric circuit system, and a chemical reaction system, and the control device 2 is the control device 2 described above.

Software for implementing the control device 2 or the control system 1 as hardware can also be implemented as a program. Then, such a program may be provided as a computer-readable non-transitory recording medium, or may be provided so as to be downloadable from an external server, or the program may be activated by an external computer. Then, so-called cloud computing, in which each function can be performed by the client terminal, may be performed.

Such a control program is for controlling a controlled object, and causes a computer to execute a spike signal sequence generation function and a drive signal generation function. According to the spike signal sequence generation function, the control target 3 is controlled. The spike signal train ST is generated at a timing defined by the basic control signal CS for controlling and the internal state including the disturbance, and according to the drive signal generation function, the spike signal train ST is time-series based on the spike signal train ST. The drive signal AS that continuously changes is generated.

Lastly, various embodiments according to the present invention have been described, but these are presented as examples and are not intended to limit the scope of the invention. The novel embodiment can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. The embodiment and its modifications are included in the scope and gist of the invention, and are also included in the invention described in the claims and the scope equivalent thereto.

1: control system 2: control device 2a: external control device 2b: basic control device 20: communication bus 21: communication unit 22: storage unit 23: control unit 231: basic control signal generation unit 232: spike signal sequence generation unit 233 : Drive signal generation unit 3: Control object 30: Drive unit 31: State detection unit CS: Control signal ST: Spike signal train AS: Drive signal I: Input signal ApEn: Entropy N: Number of neurons x_0: Initial position v_0: Initial velocity b: Bias input f_0: Natural frequency τ_s: Synapse time constant g^A: Amplification gain v^g: Center-of-gravity moving speed μ: Friction coefficient

Claims (7)

A control device, which is configured to control the control target by supplying a drive signal to the control target,
A spike signal train generation unit and a drive signal generation unit,
The spike signal string generation unit is configured to generate a spike signal string at a timing defined by a basic control signal for controlling the controlled object and an internal state including a disturbance, and dynamics related to the internal state,
The drive signal generation unit is configured to be capable of generating the drive signal that continuously changes in time series based on the spike signal train.
Control device. The control device according to claim 1,
Further comprising a communication unit and a basic control signal generation unit,
The communication unit is configured to be able to receive the state information of the control target, where:
The state information, at a point of interest in the control target, is information indicating an internal state that changes due to the behavior of the control target and environmental changes,
The basic control signal generation unit is configured to generate the basic control signal based on the state information,
Control device. In the control device according to claim 1 or 2,
The spike signal train generation unit,
Having multiple neurons that make up a neuron network,
It is configured such that the spike signal train can be generated by one neuron network based on the signals output by each of the plurality of neurons.
Control device. The control device according to any one of claims 1 to 3,
The neuron of the spike signal sequence generation unit is configured to be able to generate the spike signal by using generation of a stochastic impulse-like action potential in a living body as a model.
Control device. A control system,
A control target, and a control device for controlling the control target,
The control target is at least one of a robot, a mobile body, a pacemaker, an electric circuit system, a chemical reaction system, a communication network, a socioeconomic management system, a financial system, a biological network, and animals and plants,
The control device is the control device according to any one of claims 1 to 4.
Control system. The control system according to claim 5,
The control target is a musculoskeletal robot, and includes a robot drive unit and a robot state detection unit,
The robot driving unit includes a plurality of bones, a plurality of joints, a muscle that applies a tensile force between adjacent bones, and/or a multi-joint muscle that applies a tensile force across the plurality of bones,
The robot state detection unit is configured to detect at least one state of muscle force, muscle length, joint angle, body posture, center of gravity position, sole reaction force, 3-axis posture, 3-axis acceleration, and 3-axis angular velocity. Will be
Control system. A control program for controlling a controlled object,
It causes the computer to execute the spike signal train generation function and the drive signal generation function.
According to the spike signal train generation function, a spike signal train is generated at a timing defined by a basic control signal for controlling the controlled object and an internal state including a disturbance,
According to the drive signal generation function, the drive signal that continuously changes in time series is generated based on the spike signal train,
Control program.