KR20100022647A - Robot control system based analog neuron - Google Patents

Abstract

PURPOSE: An analog neuron-based robot control system is provided to produce other pattern signals although one neuron circuit is out of order by forming a neuron network and producing pattern signals. CONSTITUTION: An analog neuron-based robot control system comprises a sensor signal receiver(120), a sensing pattern generation part(130) and a motion pattern generation part(140). The sensor signal receiver receives sensor signals from a sensor(110) mounted on the sensor signal receiver. The sensing pattern generation part converts operation potention into pattern signals expressed as the pulse frequency. The motion pattern generation part outputs motion signals for driving an actuator(150). The neuron circuit comprises a negative resistor, transistor impedance, leak resistance, and cell membrane capacity. The motion pattern generation part comprises a signal converter and an analog switch.

Description

Robot control system based analog neuron

The present invention relates to a robot control system, and more particularly, to an analog neuron-based robot control system.

In general, a device that performs an operation similar to a human by using an electric or magnetic action is called a robot. The etymology of the robot is said to come from Slavic ROBOTA (slave machine). Since the late 1960s, robots began to spread in Korea, and industrial robots, such as manipulators and transfer robots, have been developed and utilized for the purpose of automation and unmanned production.

Recently, researches on a legged mobile robot having a movable leg have been actively conducted. In particular, a legged mobile robot that models a human motion is called a humanoid robot. The robot can be crawled, crawled, grouped 4, group 6, or the like. The bipedal upright system can realize flexible movement work such as being able to cope with uneven places or obstacles such as walking surface with irregularities on the work path, and discontinuous walking such as stairs and ladders. However, the biped upright approach is difficult and unstable in posture control or walking control compared to other movement methods.

Research on walking patterns of robots for stable walking of legged mobile robots is being actively conducted, and a representative method is walking research on robots using CPG (central pattern generator). CPG mathematically models neurons to generate signal patterns that are similar to the information transmission and processing methods of biological organs. The walking pattern method of the robot using the conventional CPG is a software method through modeling, which simplifies the characteristics of the neuron and simply generates a certain pattern, so that the mechanism of the complex neuron cannot be accurately expressed. In other words, since only a part of the functions of the actual neurons are used in the software-based CPG, the walking pattern of the biomimetic robot is not sufficiently represented.

There is a need for a robot control system capable of adaptively expressing a walking pattern of a robot.

The technical problem to be solved by the present invention is to provide an analog neuron-based robot control system.

Analog neuron-based robot control system according to an embodiment of the present invention is a sensor signal receiving unit for generating a motion potential by receiving a sensor signal from a sensor mounted on the robot, a neuron network consisting of a plurality of neuronal circuits A sensory pattern generator for converting the sensory pattern signal into a pattern signal represented by a pulse frequency, and an exercise pattern generator for converting the sensory pattern signal into an exercise signal for driving an actuator related to the operation of the robot.

The neuron circuit may include a negative resistance, a transistor impedance, a leakage resistance, and a membrane capacitor. The negative resistance may be provided by a silicon controlled rectifier (SCR) thyristor. The gate variable of the negative resistance may be a membrane potential and an inductor load potential.

The motion pattern generator is a signal converter for converting a command signal indicating a unit operation or a basic operation of the robot into a voltage signal and whether the motion neuron outputs the pattern signal as the motion signal by comparing the voltage signal with a comparison voltage. It may include an analog switch for controlling. The analog switch may include a plurality of comparators as many as the unit operation or the basic operation, and the comparison voltage of each comparator may be determined differently. The exercise neuron may include a voltage follower outputting the same signal as the pattern signal, an integrator for filtering the pattern signal output from the voltage follower, and a pattern signal passing through the integrator to drive the actuator. It may include an amplifier to amplify the signal.

Analog neuron circuits, which perform functions similar to those of real biological neurons, can be used to make robots move more like humans. In addition, by constructing a neuron network using a plurality of neuron circuits to generate a pattern signal, even if one neuron circuit fails, it is possible to generate another pattern signal to control the movement of the robot according to the environment.

1 shows an example of the structure of a quadruped walking robot.

Referring to FIG. 1, the group 4 walking robot 10 may be manufactured by mimicking the appearance and skeletal structure of the group 4 walking animal. The quadruped walking robot 10 is provided between a pair of front legs 400R and 400L supporting the ground, and each front legs 400R and 400L are coupled to each front legs 400R and 400L. At the same time is coupled to the scapula part 200 to buffer the impact load generated in each of the front leg (400R, 400L), the pelvis (300) and the pelvis (300) are spaced apart from the scapula (200) by a predetermined distance A pair of rear leg parts 500R and 500L which support the ground is provided.

The scapular part 200 may be provided by combining the head 100 corresponding to the head when imitating the appearance of the multi-pedal walking animal, and the head 100 has a CCD (Charge Coupled Device) camera 120 for image recognition. And a separate lighting means and various detection sensors (for example, distance detection sensor, etc.) may be mounted.

The quadruped walking robot 10 is provided with a control system for controlling the operation of the robot. The control system receives the detection signals from various sensors to determine the operating range of the robot and to control the motion related devices. The control system may be provided with a microchip or the like, and there is no limitation on the position at which the quadruped walking robot 10 is mounted.

Hereinafter, a robot control system for controlling the operation of the robot will be described. Although the four-legged walking robot 10 has been exemplified above, the robot control system is not limited to the shape of the robot or the number of legs.

Figure 2 compares the structure of the robot control system which is a temporary example according to the present invention and the biological nervous system.

Referring to FIG. 2, a general biological nervous system includes sensory neurons, intermediate neurons, and motor neurons. Each neuron and neuron is connected by a synapse.

The proposed robot control system has a structure corresponding to that of the biological nervous system. The robot control system corresponding to the biological nervous system includes an artificial dendrite that collects sensor signals from various sensors, and an artificial sensory neuron cell body that generates sensory neuron signals by receiving an action potential from the artificial dendrite. sensory neuron cell body) and an artificial motor neuron cell body that receives sensory neuron signals and generates motor neuron signals. In the robot control system, sensory neuron signals and motor neuron signals are expressed as pulse frequencies, as are signals of actual biological nervous system neurons. That is, neuron information varies with the period of the signal and is not affected by amplitude.

Figure 3 is a block diagram showing a robot control system according to an embodiment of the present invention.

Referring to FIG. 3, the robot control system may include a sensor signal receiver 120 that receives sensor signals from various sensors 110, a sensory pattern generator 130 that converts sensor signals into sensory pattern signals, and an actuator that senses sensory pattern signals. 150 includes a movement pattern generator 140 for converting the movement pattern signal for driving.

The sensor 110 may be provided as a sensor having various functions attachable to a robot such as a visual sensor for image recognition, an acoustic sensor for voice recognition, a pressure sensor for contact recognition, and the like. The sensor signal measured by each sensor 110 may be collected by the sensor signal receiver 120 in the form of a binary signal. The sensor signal receiver 120 synthesizes the sensor signals collected from each sensor 110, generates an operation potential, and transmits the movement potential to the sensory pattern generator 130.

The sensory pattern generator 130 may include a plurality of sensory neurons. Sensory neurons are composed of neuron circuits, and a plurality of neuron circuits may be connected to each other to form a neuron network. In neuronal networks, a number of neuronal circuits work together to generate vibration patterns. The sensory pattern generator 130 generates a sensory pattern signal represented by a pulse frequency and transmits the motion potential to the exercise pattern generator 140. The sensory pattern signal expresses the motion potential in frequency. The sensory pattern signal may vary depending on the number of neuron circuits included in the sensory pattern generator 130 and the connection structure of the neuron network.

The exercise pattern generator 140 includes a plurality of exercise neurons. The plurality of motor neurons may form a neuron network to generate a moving signal. Alternatively, one motor neuron may be in charge of one actuator 150 or several motor neurons may be in charge of one actuator 150 to transmit an exercise signal to the actuator 150. The motion signal may be information indicating an operation order, an operation range, and the like of the plurality of actuators 150.

The exercise pattern generator 140 may selectively activate a plurality of exercise neurons according to a command signal. The exercise pattern generator 140 may determine whether the exercise neuron is operated according to the command signal. That is, the movement pattern generator 140 controls the movement of the robot by controlling the movement neurons. The operation of the robot is determined according to the complex operation of the plurality of actuators 150, and the command signal may be a unit motion or a signal representing a basic operation of the robot such as forward, backward, right turn, left turn, etc. constituting the complex operation. have. The unit motion refers to the operation of one actuator 150, and the basic motion refers to the basic operation of the robot operated by the at least one actuator 150. The command signal may be a signal transmitted by the user to the robot through a remote controller or a signal transmitted from a sensor attached to the robot. The command signal may be an analog signal. That is, a signal transmitted from a remote controller or a sensor is converted into an analog signal through a neuron circuit or a neuron network and input to the exercise pattern generator 140. The command signal may be a signal generated through a neuron circuit or a neuron network of the sensory pattern generator 130, or may be a signal generated through a neuron circuit or a neuron network provided separately. Alternatively, the command signal may be an analog signal acquired through past actions of the robot. When the stored analog signal is applied to the movement pattern generator 140, the robot may perform the same movement as the past actions.

Actuator 150 is a driving device related to the operation of the robot is not limited to the driving method or structure.

4 shows a neuron circuit according to one embodiment according to the present invention.

Referring to FIG. 4, a neuron circuit circuits the function of a living neuron. Neuronal circuits include negative resistance and transistor impedance. In neuronal circuits, R stands for leakage resistance and C stands for membrane capacitor.

The negative resistance may be provided as a silicon controlled rectifier (SCR) thyristor to which the PNP transistor and the NPN transistor are bonded. Negative resistance gate parameters are membrane potential and inductor load potential. Negative resistance represents sodium channels in living neurons.

Transistor impedance shows a phase shift result similar to inductance with a 90 Hz phase shift of RC and a 180 dB phase shift of the transistor. Transistor impedance represents potassium channels in living neurons. That is, the transistor impedance generates the same signal as that caused by the movement of potassium ions in the living neurons.

5 shows a neuronal network, an embodiment according to the present invention. 6 shows a neuronal network, which is another embodiment according to the present invention.

5 and 6, a neuron network is composed of a plurality of neuron circuits and synapses connecting the neuron circuits to each other. There is no limit to the number of neuronal circuits and synapses that make up a neuronal network. Synapses transmit signals from one neuronal circuit to another. One synapse may be connected between neuron circuits to transmit a signal in only one direction, or two synapses may be connected to transmit a signal in both directions.

FIG. 5 shows three neuron circuits connected by six synapses, and FIG. 6 shows four neuron circuits connected by eight synapses. A neuron network is a synapse of many neuronal circuits, including two, three, four, ... Up to two synapses may be used when two neuronal circuits are used, up to six synapses may be used when three neuronal circuits are used, and up to 12 synapses may be used when four neuronal circuits are used. Depending on the synapse connection state, the pattern signal generated in the neuronal network may vary. For example, if one neuron circuit breaks in a neuron network, the neuron network is connected and the neuron network generates signals of different patterns using the remaining neuron circuits and synapses.

FIG. 7 illustrates an example of a pattern signal generated in the neuron network of FIG. 5. FIG. 8 illustrates an example of a pattern signal generated in the neuron network of FIG. 6. It can be seen that the pattern signal generated differs depending on the number of neuronal circuits and synaptic connections that form the neuron network.

9 is a block diagram showing an exercise pattern generator according to an embodiment of the present invention.

Referring to FIG. 9, the exercise pattern generator 140 includes a signal converter 210, an analog switch 220, and a exercise neuron 230.

The signal converter 210 may be configured as a frequency to voltage converter. The signal converter 210 converts a command signal into a voltage signal. The command signal may be a signal transmitted from a user's remote controller or a signal transmitted from a sensor attached to the robot. The command signal may be an analog signal converted to mean a unit operation or a basic operation of the robot through a neuron circuit or a neuron network.

The analog switch 220 controls whether the plurality of motor neurons 230 are operated. That is, the analog switch 220 serves as a pre-synaptic of the biological nervous system. The analog switch 220 may be configured with as many comparators as the number of unit motions or basic motions of the robot, and each comparator is connected to at least one motor neuron 230 that is responsible for the unit motion or basic motion of the robot. do. Each comparator has a specific comparison voltage, and compares the input voltage signal with the comparison voltage to turn on the connected motor neuron 230 when the voltage signal is equal to or greater than the comparison voltage. The pattern signal is transmitted to all the exercise neurons 230, but only the exercise neurons 230 that generate the exercise signals according to the pattern signals.

For example, if the basic motion of the robot consists of forward, backward, right and left, four comparators for each basic motion can be provided and each comparator The comparison voltage of may be determined as V1ref, V2ref, V3ref, and V4ref. If the voltage signal converted by the signal converter 210 is equal to or greater than V1ref, the analog switch 220 operates only the motor neurons 230 connected to the comparator having a comparison voltage of V1ref, and other motor neurons 230 are operated. Don't let that happen.

The exercise neuron 230 converts the pattern signal into an exercise signal capable of driving the actuator and outputs the exercise signal. The motor neurons 230 may be provided in at least a number corresponding to the number of unit motions or basic motions of the robot. For example, when the basic motion of the robot is forward, backward, right turn, and left turn, at least four motor neurons 240 may be provided, and each motor neuron 240 is responsible for each basic motion.

10 is a block diagram illustrating an example of a motor neuron according to the present invention.

Referring to FIG. 10, the motor neuron 230 is composed of a voltage follower 310, an integrator 320, and an amplifier 330.

The voltage follower 310 outputs the same signal as the input pattern signal and applies it to the integrator 320. The integrator 320 filters the pattern signal output from the voltage follower 310. The integrator 320 may be provided as a low pass filter that cuts a frequency higher than a predetermined frequency and passes a low frequency. The frequency determined by the integrator 320 may be set to the same value or different value for each motor neuron. If the frequency of the integrator 320 is determined differently for each motor neuron 230, only a specific pattern signal can be generated from the corresponding motor neuron 230, and the motion generated from the motor neuron 230 as the pattern signal is changed. The signal can vary. The amplifier 330 amplifies and outputs the pattern signal passed through the integrator 320 into a motion signal having a size appropriate to drive an actuator.

11 shows the operation of a robot control system according to an embodiment of the present invention.

Referring to FIG. 11, a sensor signal obtained from various sensors mounted on a robot is generated as a pattern signal through a neuron network provided in the sensory pattern generator. The pattern signal generated in the neuron network may be transmitted to a plurality of motor neurons. The pattern signal is expressed in frequency irrespective of the vibration width of the signal. A command signal indicating a unit motion or basic motion of the robot may be applied to the motion pattern generator, and a motion neuron operated according to the command signal is determined. The command signal may mean a basic motion of the robot such as forward, backward, right turn, and left turn, and a plurality of motor neurons may be divided into motor neurons that are responsible for the basic motion of the robot. The pattern signal generated from the exercise pattern generator may be transmitted to all motor neurons, but only the motor neurons operated according to the command signal output the pattern signal as an exercise signal capable of driving the actuator to operate the robot.

In the control of the robot, the walking or driving pattern of the robot may be determined according to how the neuron network is configured. Compared to numerically controlling the control of the robot, as proposed, the robot control system using analog neuron network composed of a structure similar to the biological nervous system is suitable for the control of the biomimetic robot and the behavior of the robot. I can express it better.

In the digital way of expressing neurons in general software, if one neuron fails, the robot will fail or runaway. However, in an analog method using a neuron network composed of neuron circuits, even if one neuron fails, the remaining neurons generate a signal suitable for a condition, thereby realizing proper movement like a real organism. For example, if a person injuries one leg, walking in a different pattern than walking normally or walking in a different pattern using crutches, a quadruped robot generates another pattern signal in the neuron network when one leg fails. This allows the robot to move on the remaining three legs.

As mentioned above, preferred embodiments of the present invention have been described in detail, but those skilled in the art to which the present invention pertains should understand the present invention without departing from the spirit and scope of the present invention as defined in the appended claims. It will be appreciated that various modifications or changes can be made. Accordingly, modifications to future embodiments of the present invention will not depart from the technology of the present invention.

1 shows an example of the structure of a quadruped walking robot.

Figure 2 compares the structure of the robot control system which is a temporary example according to the present invention and the biological nervous system.

Figure 3 is a block diagram showing a robot control system according to an embodiment of the present invention.

4 shows a neuron circuit according to one embodiment according to the present invention.

5 shows a neuronal network, an embodiment according to the present invention.

6 shows a neuronal network, which is another embodiment according to the present invention.

FIG. 7 illustrates an example of a pattern signal generated in the neuron network of FIG. 5.

FIG. 8 illustrates an example of a pattern signal generated in the neuron network of FIG. 6.

9 is a block diagram showing an exercise pattern generator according to an embodiment of the present invention.

10 is a block diagram illustrating an example of a motor neuron according to the present invention.

11 shows the operation of a robot control system according to an embodiment of the present invention.

Claims (7)

In the analog neuron-based robot control system, A sensor signal receiver configured to receive a sensor signal from a sensor mounted on the robot and generate an operation potential; A sensory pattern generation unit for converting the operation potential into a pattern signal represented by a pulse frequency using a neuron network including a plurality of neuron circuits; And An analog neuron-based robot control system including a movement pattern generator for outputting the pattern signal as a movement signal for driving an actuator related to the operation of the robot. The analog neuron-based robot control of claim 1, wherein the neuron circuit includes negative resistance, transistor impedance, leakage resistance, and membrane capacitor. system. The analog neuron-based robot control system of claim 2, wherein the negative resistance is formed of a silicon controlled rectifier (SCR) thyristor. The analog neuron-based robot control system of claim 2, wherein the gate variable of the negative resistance is a membrane potential and an inductor load potential. The method of claim 1, wherein the movement pattern generation unit A signal converter converting a command signal indicating a unit operation or basic operation of the robot into a voltage signal; And Analogue neuron-based robot control system comprising an analog switch for controlling the operation of the motor neuron outputting the pattern signal as the motion signal by comparing the voltage signal with the comparison voltage. The analog neuron-based robot control system of claim 5, wherein the analog switch includes a plurality of comparators as many as the unit operation or the basic operation, and the comparison voltage of each comparator is determined differently. The method of claim 5, wherein the motor neuron A voltage follower outputting the same signal as the pattern signal; An integrator for filtering the pattern signal output from the voltage follower; And And an amplifier for amplifying the pattern signal passing through the integrator into a motion signal for driving the actuator.

Images