Control device

Abstract

A control device 10 controls a machine device on which drive units Aa, Ab are provided. The control device 10 includes detection units Da, Db, processing units Pa, Pb, Pc, and drive control units Ca, Cb. The detection units Da, Db convert detection signals output from sensors Sa, Sb into sensory information and transmit the converted sensory information to communication networks Na, Nb. The processing parts Pa, Pb, Pc analyze sensory information transmitted to the communication networks Na, Nb, generate action command information, and transmit the generated action command information to the communication networks Na, Nb. The drive control units Ca, Cb analyze the action command information transmitted to the communication networks Na, Nb and control the drive units Aa, Ab on the basis of the analyzed action command information.

Description

control device CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application is based on and claims the benefit of priority from Japanese Patent Application No. 2023-219072 filed on December 26, 2023, and Japanese Patent Application No. 2024-75423 filed on May 7, 2024, the entire contents of which are incorporated herein by reference.

[0002] The present disclosure relates to a control device.

[0003] A conventional robot is described in Patent Document 1 below. This robot includes a tactile detection unit, a generation unit, a memory unit, an analysis unit, and a control unit. The tactile detection unit detects external contact. The generation unit generates contact information based on the detection results. The memory unit accumulates and stores the contact information for a predetermined period of time. The analysis unit analyzes the tendencies of each user based on the accumulated contact information. The control unit controls a mechanism for causing the robot to take action, and dynamically changes the action based on the analysis results.

[0004] Japanese Patent Application Laid-Open No. 2017-119336

[0005] The control device described in Patent Document 1 has room for improvement in terms of versatility.

[0006] An object of the present disclosure is to provide a control device that can enhance versatility.

[0007] A control device according to one aspect of the present disclosure is a control device for controlling a mechanical device provided with multiple drive units, and includes a detection unit, a processing unit, and a drive control unit. The detection unit converts a detection signal output from a detection unit sensor provided in the mechanical device when the detection unit detects predetermined information into sensory information and transmits the converted sensory information to a transmission path. The processing unit analyzes the sensory information transmitted from the detection unit to the transmission path, and generates action command information, which is information on an action to be taken by the mechanical device, based on the analysis of the sensory information, and transmits the generated action command information to the transmission path. The drive control unit analyzes the action command information transmitted from the processing unit to the transmission path, and controls the drive units based on the analyzed action command information.

[0008] With this configuration, the detection unit, processing unit, and drive control unit operate based on sensory information and behavioral command information, making it possible to achieve greater versatility compared to conventional control devices that operate based on numerical signals or complex control signals.

[0009] FIG. 1 is a block diagram showing a schematic configuration of a control device according to a first embodiment. FIG. 2 is a diagram showing a schematic configuration of a robot device according to a first embodiment. FIG. 3 is a block diagram showing a portion of the configuration of a control device according to a first embodiment. FIG. 4 is a block diagram showing a portion of the configuration of a control device according to a first embodiment. FIG. 5 is a block diagram showing a configuration of a detection unit according to a first embodiment. FIG. 6 is a block diagram showing a configuration of a neural network according to a first embodiment. FIG. 7 is a block diagram showing a configuration of a memory controller according to a first embodiment. FIG. 8 is a diagram showing a schematic example of information included in a semantic library according to a first embodiment. FIG. 9 is a block diagram showing a configuration of a processing unit according to a first embodiment. FIG. 10 is a diagram showing a schematic example of information included in a semantic library according to a first embodiment. FIG. 11 is a block diagram showing a configuration of a drive control unit according to a first embodiment. FIG. 12 is a block diagram showing a configuration of a processing unit according to a second embodiment. FIG. 13 is a block diagram showing a configuration of a drive control unit according to a second embodiment. FIG. 14 is a block diagram showing a partial configuration of a detection unit and a processing unit according to a third embodiment. FIG. 15 is a block diagram showing a partial configuration of a processing unit and a drive control unit according to a third embodiment.

[0010] An embodiment of a control device will be described below with reference to the drawings. To facilitate understanding of the description, the same components in the drawings will be denoted by the same reference numerals as much as possible, and duplicate descriptions will be omitted.

[0011] First Embodiment First, an overview of a control device according to a first embodiment will be described.

[0012] 1, a control device 10 according to this embodiment is mounted on a machine 20 that is to be controlled. The machine 20 is, for example, a humanoid robot or an industrial robot. The control device 10 includes a plurality of high-speed detection units Da, a plurality of low-speed detection units Db, a plurality of high-speed drive control units Ca, a plurality of low-speed drive control units Cb, a plurality of upper-level processing units Pa, a plurality of middle-level processing units Pb, and a plurality of lower-level processing units Pc.

[0013] The multiple high-speed detection units Da, the multiple high-speed drive control units Ca, the multiple upper processing units Pa, and the multiple middle processing units Pb are communicably connected to one another via a high-speed communication network Na. Information transmitted to the high-speed communication network Na is broadcast information, and is processed arbitrarily by the multiple high-speed detection units Da, the multiple high-speed drive control units Ca, the multiple upper processing units Pa, and the multiple middle processing units Pb.

[0014] The plurality of low-speed detection units Db, the plurality of low-speed drive control units Cb, the plurality of upper processing units Pa, the plurality of middle processing units Pb, and the plurality of lower processing units Pc are communicably connected to one another via a low-speed communication network Nb. Information transmitted to the low-speed communication network Nb is also information transmitted in a broadcast manner, and is processed arbitrarily by the plurality of low-speed detection units Db, the plurality of low-speed drive control units Cb, the plurality of upper processing units Pa, the plurality of middle processing units Pb, and the plurality of lower processing units Pc.

[0015] The high-speed detection unit Da and the low-speed detection unit Db are components of the mechanical device 20 that detect predetermined physical quantities. To this end, the high-speed detection unit Da and the low-speed detection unit Db each include a sensor Sa for detecting the predetermined physical quantities, and a sensor Sb for detecting the predetermined physical quantities. The sensors Sa and Sb are mounted on the mechanical device 20 and include, for example, a camera sensor, an image sensor, an acoustic sensor, a posture sensor, a taste sensor, a radio wave sensor, a gas sensor, a temperature sensor, a contact sensor, and a pressure sensor. The sensor Sa mounted on the high-speed detection unit Da is required to operate faster than the sensor Sb mounted on the low-speed detection unit Db. Whether various sensors are classified as the high-speed detection unit Da or the low-speed detection unit Db depends on the configuration of the mechanical device 20 that is the control target of the control device 10. For example, if the mechanical device 20 is a humanoid robot, the mechanical device 20 may be equipped with a camera sensor for acquiring information about its surroundings and a contact sensor for detecting whether an object has come into contact with the robot. In this case, image information captured by the camera sensor is used to control the entire robot, so it is desirable for the image information to be processed as quickly as possible. Therefore, the camera sensor is classified as one of the sensors Sa of the multiple high-speed detection units Da. On the other hand, contact information detected by the contact sensor is rarely used as information of high urgency when controlling the robot. Therefore, the contact sensor is classified as one of the sensors Sb of the multiple low-speed detection units Db.

[0016] The detection units Da and Db convert the detection signals output from the sensors Sa and Sb by detecting a predetermined physical quantity into sensory information, abstract the converted sensory information, and transmit it to the high-speed communication network Na. For example, if the sensor Sb is a contact sensor and is provided on the robot's right hand, the low-speed detection unit Db generates verbalized sensory information such as "The robot's right hand is touching a hard object" or "The robot's right hand is touching a cold object" based on the detection signal from the sensor Sb. In this embodiment, the process of verbalizing the sensory information corresponds to the process of abstracting the sensory information. The low-speed detection unit Db then transmits the verbalized sensory information to the low-speed communication network Nb. Note that the sensory information includes information representing the five human senses and information about the location where the sensation was obtained. When the sensor Sa detects a predetermined physical quantity and outputs a detection signal, the high-speed detection unit Da also transmits sensory information corresponding to the detection signal to the high-speed communication network Na.

[0017] The sensory information transmitted from the high-speed detection unit Da to the high-speed communication network Na is acquired by a plurality of upper processing units Pa and a plurality of middle-level processing units Pb. The sensory information transmitted from the low-speed detection unit Db to the low-speed communication network Nb is acquired by a plurality of upper processing units Pa, a plurality of middle-level processing units Pb, and a plurality of lower processing units Pc. Each processing unit Pa, Pb, Pc analyzes the sensory information acquired via each communication network Na, Nb, and generates action command information, which is information on the action to be taken by the mechanical device 20, based on the analysis results. The action command information includes information that determines the action of the mechanical device 20 as a whole, information that determines the operation of each part of the mechanical device 20, etc.

[0018] The upper processing unit Pa, the middle processing unit Pb, and the lower processing unit Pc have different processing contents. For example, if the mechanical device 20 controlled by the control device 10 is a humanoid robot, the upper processing unit Pa generates action command information that determines the movement of the entire robot, such as the direction in which the mechanical device 20 should move. The middle processing unit Pb and the lower processing unit Pc generate action command information that determines the movement of each of the robot's multiple parts. For example, the multiple middle processing units Pb generate action command information that determines the movement of each of the robot's multiple hands. The multiple lower processing units Pc generate action command information that determines the movement of each of the robot's multiple arms. The processing of the middle processing unit Pb takes priority over the processing of the lower processing unit Pc. The processing of the upper processing unit Pa takes priority over the processing of the lower processing unit Pc and the middle processing unit Pb. One of the multiple upper processing units Pa, the multiple middle processing units Pb, and the multiple lower processing units Pc can exchange various information with an external device De, such as a random number generator.

[0019] For example, if the low-speed detection unit Db transmits sensory information such as "The robot's right hand is touching a cold object" to the low-speed communication network Nb and the sensory information is received by the middle-level processing unit Pb, the middle-level processing unit Pb analyzes the sensory information and generates action command information such as "Immediately move the robot's right hand away from the object." The middle-level processing unit Pb then transmits the generated action command information to the high-speed communication network Na and the low-speed communication network Nb. The other upper processing units Pa and lower processing units Pc also operate in the same way based on the sensory information transmitted from the detection units Da and Db to the high-speed communication network Na and the low-speed communication network Nb.

[0020] The behavior command information transmitted from the upper processing unit Pa and the middle processing unit Pb to the high-speed communication network Na is received by a plurality of high-speed drive control units Ca. Furthermore, the behavior command information transmitted from the upper processing unit Pa, the middle processing unit Pb, and the lower processing unit Pc to the low-speed communication network Nb is received by a plurality of low-speed drive control units Cb. The high-speed drive control unit Ca controls the high-speed drive unit Aa provided in the mechanical device 20. The low-speed drive control unit Cb controls the low-speed drive unit Ab provided in the mechanical device 20. The drive units Aa and Ab are motor devices, speaker devices, lighting devices, etc. that constitute the mechanical device 20. The high-speed drive unit Aa is a device that is required to operate faster than the low-speed drive unit Ab. Whether the various devices provided in the mechanical device 20 are classified as high-speed drive units Aa or low-speed drive units Ab depends on the configuration of the mechanical device 20 that is the control target of the mechanical device 20. For example, if the mechanical device 20 is a humanoid robot, the actuator device that moves the robot's hands is classified as a high-speed drive unit Aa because it requires high-speed operation, while the actuator device that moves the robot's arms is classified as a low-speed drive unit Ab because it does not require such high-speed operation.

[0021] The drive control units Ca and Cb control the drive units Aa and Ab, which are the objects of control, based on action command information transmitted from the processing units Pa, Pb, and Pc to the communication networks Na and Nb. For example, if the middle processing unit Pb transmits action command information such as "immediately move the robot's right hand away from the object" to the high-speed communication network Na and the action command information is received by the high-speed drive control unit Ca, the high-speed drive control unit Ca analyzes the action command information and operates the high-speed drive unit Aa, which is an actuator device that moves the robot's hand.

[0022] (Specific Configuration of Mechanical Device and Control Device) Next, a specific configuration of the control device 10 when the mechanical device 20 shown in FIG. 1 is a humanoid robot device will be described.

[0023] 2, the robot device 30 of this embodiment includes a body 31, a head 32, shoulders 33a and 33b, arms 34a and 34b, hands 35a and 35b, a waist 36, and legs 37a and 37b. The robot device 30 is powered by a battery 38 installed in the waist 36.

[0024] The head 32 is provided with a camera sensor 40, a distance measurement sensor 41, an audio sensor 42, and a speaker device 43. The camera sensor 40 captures an image of the area in front of the robot and acquires image data of the area in front of the robot. The distance measurement sensor 41 is a sensor that measures the distance to an object in front of the robot. The audio sensor 42 is a sensor that detects sounds around the robot. The speaker device 43 functions as the robot's mouth and emits voice, music, etc.

[0025] The shoulders 33a and 33b respectively include built-in actuators 330a and 330b, which actuate the arms 34a and 34b. In this embodiment, the actuators 330a and 330b function as drive units.

[0026] Contact sensors 50 to 53 are provided on the arms 34a and 34b, respectively. Contact sensors 54 and 55 are also provided on the hands 35a and 35b, respectively. Furthermore, contact sensors 56 to 59 are also provided on the legs 37a and 37b, respectively. The contact sensors 50 to 59 are sensors that can detect whether or not an object has come into contact with them. The contact sensors 50 to 59 can also detect the physical properties of the object they come into contact with, such as elasticity or viscosity, and output a detection signal corresponding to the detected physical properties.

[0027] The hand portions 35a and 35b respectively incorporate actuator devices 350a and 350b, which actuate the hand portions 35a and 35b.

[0028] Next, a specific configuration of the control device 10 mounted on the robot device 30 will be described with reference to FIGS.

[0029] The control device 10 includes a high-speed communication network Na as shown in FIG. 3 and a low-speed communication network Nb as shown in FIG.

[0030] As shown in FIG. 3, high-speed detection units Da1 to Da5, high-speed drive control units Ca1 to Ca3, upper processing units Pa1 to Pa3, and middle processing units Pb1 and Pb2 are connected to the high-speed communication network Na.

[0031] The high-speed detection units Da1 to Da5 process the detection signals of the camera sensor 40, the distance measurement sensor 41, the audio sensor 42, and the contact sensors 54 and 55. The high-speed drive control units Ca1 to Ca3 control the speaker device 43 and the actuator devices 350a and 350b, respectively.

[0032] The high-speed detection unit Da1 generates sensory information related to the vision of the robot device 30 based on the image information acquired by the camera sensor 40, and transmits the generated sensory information to the high-speed communication network Na. The sensory information related to the vision is, for example, information such as "The traffic light in front of the robot is red."

[0033] The high-speed detection unit Da2 generates sensory information related to the vision of the robot device 30 based on the information on the distance to a predetermined object acquired by the distance measurement sensor 41, and transmits the generated sensory information to the high-speed communication network Na. The sensory information related to the vision is, for example, information such as "There is an obstacle 5 meters ahead of the robot."

[0034] The high-speed detection unit Da3 generates sensory information related to the robot's hearing based on the audio information acquired by the audio sensor 42, and transmits the generated sensory information to the high-speed communication network Na. Sensory information related to hearing is, for example, information such as "I can hear the sound of a train."

[0035] The high-speed detection units Da4 and Da5 generate sensory information related to the sense of touch of the robot device 30 based on the contact information acquired by the contact sensors 54 and 55, and transmit the generated sensory information to the high-speed communication network Na. Sensory information related to the sense of touch is, for example, information such as "The right hand of the robot is in contact with a hard object."

[0036] The upper processing units Pa1 to Pa3 generate action command information based on the sensory information transmitted from the high-speed detection units Da1 to Da5 to the high-speed communication network Na, and transmit the generated action command information to the high-speed communication network Na. For example, the upper processing unit Pa1 generates action command information that determines the overall behavior of the robot device 30 by using the sensory information transmitted from the high-speed detection units Da1 to Da5. The upper processing unit Pa2 also generates action command information that determines the overall behavior of the robot device 30 by using the sensory information transmitted from the high-speed detection units Da1 to Da5. The upper processing unit Pa1 generates action command information related to the behavior of the larger robot device 30 than the upper processing unit Pa2. Note that, because the upper processing unit Pa1 is positioned higher than the upper processing unit Pa2, the action command information of the upper processing unit Pa1 takes priority over the action command information of the upper processing unit Pa2. The upper processing unit Pa3 generates action command information related to, for example, sound by using the sensory information transmitted from the high-speed detection units Da1 to Da5.

[0037] The right-hand middle processing unit Pb1 uses the sensory information transmitted from the high-speed detection units Da1 to Da5 to generate action command information that determines the action of the right hand 35a of the robot device 30, and transmits the generated action command information to the high-speed communication network Na. The left-hand middle processing unit Pb2 uses the sensory information transmitted from the high-speed detection units Da1 to Da5 to generate action command information that determines the action of the left hand 35b of the robot device 30, and transmits the generated action command information to the high-speed communication network Na.

[0038] The high-speed drive control unit Ca1 controls the speaker device 43 based on action command information related to voice among the action command information transmitted from the processing units Pa1 to Pa3, Pb1, and Pb2 to the high-speed communication network Na.

[0039] The high-speed drive control unit Ca2 controls the actuator device 350a of the right hand portion 35a based on the action command information related to the right hand among the action command information transmitted from each processing unit Pa1 to Pa3, Pb1, Pb2 to the high-speed communication network Na.

[0040] The high-speed drive control unit Ca3 controls the actuator device 350b of the left hand portion 35b based on the action command information related to the left hand among the action command information transmitted from each processing unit Pa1 to Pa3, Pb1, Pb2 to the high-speed communication network Na.

[0041] As shown in FIG. 4, low speed detection units Db1 to Db8, low speed drive control units Cb1 and Cb2, and lower processing units Pc1 to Pc3 are connected to the low speed side communication network Nb.

[0042] The low-speed detection units Db1 to Db8 generate sensory information related to the sense of touch of the robot device 30 based on the contact information acquired by the contact sensors 50 to 53 and 56 to 59, and transmit the generated sensory information to the high-speed communication network Na. Sensory information related to the sense of touch is, for example, information such as "The right foot of the robot is in contact with a soft object."

[0043] The lower processing units Pc1 to Pc3 generate behavioral command information that determines the behavior of the arms 34a and 34b of the robot device 30 based on the sensory information transmitted from the contact sensors 50 to 53 and 56 to 59, and transmit the generated behavioral command information to the low-speed communication network Nb.

[0044] The low-speed drive control units Cb1 and Cb2 control the actuator devices 330a and 330b provided on the shoulders 33a and 33b, respectively, based on the behavioral command information related to the arms 34a and 34b among the behavioral command information transmitted from each of the lower processing units Pc1 to Pc3 to the low-speed communication network Nb.

[0045] (Configuration of the detection units) Next, the configuration of the detection units Da1 to Da5 and Db1 to Db8 will be specifically described. Note that since the detection units Da1 to Da5 and Db1 to Db8 have the same or similar configurations, the following description will be focused on the configuration of the high-speed detection unit Da4, which corresponds to the contact sensor 54 of the right hand portion 35a.

[0046] 5, the high-speed detection unit Da4 includes a pre-processing unit 70, a recognition unit 71, a learning unit 72, an estimation unit 73, a determination unit 74, an assignment unit 75, a verbalization processing unit 76, a data output unit 77, a communication processing unit 78, a data receiving unit 79, a write processing unit 80, a data compression unit 81, and a memory controller 82. The recognition unit 71, the learning unit 72, the estimation unit 73, the memory controller 82, and the verbalization processing unit 76 are connected to each other via a data / memory bus 83 so as to be able to communicate with each other.

[0047] The pre-processing unit 70 performs pre-processing on the detection signal output from the contact sensor 54. The pre-processing includes filtering and the like.

[0048] The recognition unit 71 recognizes contact information detected by the contact sensor 54 by analyzing the detection signal of the contact sensor 54 that has been preprocessed by the preprocessing unit 70. The recognition unit 71 is configured with a neural network NN as shown in FIG. 6. As shown in FIG. 6, the neural network NN includes multiple input layers 1000, multiple intermediate layers 1001, and multiple output layers 1002. The preprocessed detection signal of the contact sensor 54 is input to the input layer 1000. The intermediate layer 1001 acquires information from the input layer 1000 and performs various calculations based on the acquired information. The output layer 1002 multiplies the input layer 1000 and the intermediate layer 1001 by predetermined weights and outputs multiple data processed with an activation function. The output layer 1002 of the recognition unit 71 outputs physical property information detected by the contact sensor 54. This detected physical property information includes, for example, numerical data such as hardness, elasticity, and viscosity.

[0049] The learning unit 72 sequentially learns the correspondence between the detection signal of the contact sensor 54 that has been pre-processed by the pre-processing unit 70 and the physical property information detected by the contact sensor 54. The learning unit 72 acquires the physical property information detected by the contact sensor 54 from the determination unit 74. The learning unit 72 is configured with a neural network NN as shown in Fig. 6. The learning unit 72 uses the detection signal of the contact sensor 54 that has been pre-processed by the pre-processing unit 70 as input information and outputs estimated physical property information by using the learned information.

[0050] The estimation unit 73 outputs estimated physical property information by using machine learning, a computer, etc., with the detection signal of the contact sensor 54 that has been pre-processed by the pre-processing unit 70 as input information.

[0051] The determination unit 74 calculates final detected physical property information by combining the detected physical property information output from the recognition unit 71 and the estimated physical property information output from the learning unit 72 and the estimation unit 73 using weighting coefficients, etc. The determination unit 74 transmits the calculated final detected physical property information to the learning unit 72. The learning unit 72 performs a learning process based on the final detected physical property information transmitted from the determination unit 74 and the detection signal of the contact sensor 54 when the physical property information is detected.

[0052] The assigning unit 75 generates sensory information corresponding to the final detected physical property information output from the determining unit 74, based on the detected physical property information. Specifically, as shown in FIG. 7 , a semantic library 821 is stored in a storage device 820 controlled by a memory controller 82. The semantic library 821 stores information indicating correspondences between a plurality of pieces of physical property information and a plurality of pieces of sensory information, for example, as shown in FIG. 8 . Based on the final detected physical property information output from the determining unit 74, the assigning unit 75 identifies, from the plurality of pieces of physical property information shown in FIG. 8 , the physical property information that is closest to the current detected physical property information, and outputs sensory information corresponding to the identified physical property information. In this way, the assigning unit 75 corresponds to a part that assigns human sensory information to the final detected physical property information output from the determining unit 74.

[0053] The language processing unit 76 converts the sensory information transmitted from the assigning unit 75 into a data format that can be transmitted to the high-speed communication network Na. The language processing unit 76 is configured with a neural network NN as shown in FIG. 6. When sensory information such as "sticky" is transmitted from the assigning unit 75, the language processing unit 76 generates sensory information such as "The robot's right hand is in contact with a sticky object" so that the position where the physical property is detected can be identified. The sensory information generated by the language processing unit 76 is transmitted to the high-speed communication network Na via the data output unit 77 and the communication processing unit 78. In this embodiment, the language processing unit 76 is an example of an abstraction unit for the detection unit that abstracts the sensory information.

[0054] The data compression unit 81 compresses the data of the detection signal of the contact sensor 54 that has been pre-processed by the pre-processing unit 70. The compressed data of the detection signal of the contact sensor 54 compressed by the data compression unit 81 is transmitted to the high-speed communication network Na via the data output unit 77 and the communication processing unit 78.

[0055] The communication processing unit 78 performs processing to transmit data sent from the data output unit 77 to the high-speed communication network Na, and also receives various information transmitted to the high-speed communication network Na. The various information transmitted to the high-speed communication network Na is information transmitted from other high-speed detection units Da1 to Da3, Da5, high-speed drive control units Ca1 to Ca3, upper processing units Pa1 to Pa3, and middle processing units Pb1 and Pb2 connected to the high-speed communication network Na. The information received by the communication processing unit 78 is transmitted to the memory controller 82 via the data receiving unit 79 and the write processing unit 80.

[0056] The memory controller 82 controls the storage device 820 shown in Fig. 7. In addition to a semantic library 821, the storage device 820 stores feature data 822, related data 823, parameters 824, label data 825, and original data 826. The memory controller 82 updates the various data stored in the storage device 820 based on data transmitted from the write processing unit 80, for example. The various data stored in the storage device 820 is also used by the determination unit 74 and the like.

[0057] (Configuration of Processing Units) Next, the configuration of each of the processing units Pa1 to Pa3, Pb1, Pb2, Pc1 to Pc3 will be described. Note that since each of the processing units Pa1 to Pa3, Pb1, Pb2, Pc1 to Pc3 has the same or similar configuration, the following description will be focused on the configuration of the right-hand middle processing unit Pb1 that controls the right-hand unit 35a.

[0058] As shown in FIG. 9 , the right-hand middle processing unit Pb1 includes an external I / F (interface) 90, a pre-processing unit 91, a recognition unit 92, a learning unit 93, an other processing unit 94, an estimation unit 95, a determination unit 96, an assignment unit 97, a verbalization processing unit 98, a data output unit 99, a communication processing unit 100, a data receiving unit 101, a write processing unit 102, a data compression unit 103, a memory controller 104, an analysis unit 105, a behavior planning unit 106, and a self-recognition unit 107. The recognition unit 92, the learning unit 93, the other processing unit 94, the estimation unit 95, the verbalization processing unit 98, the memory controller 104, the behavior planning unit 106, and the self-recognition unit 107 are communicatively connected to one another via a data / memory bus 140. Among the elements shown in FIG. 9 , elements having the same names as those shown in FIG. 5 perform similar operations, and therefore the following description will focus on the differences.

[0059] The external I / F 90 is a part to which an external device De other than the robot device 30 can be connected, and acquires information transmitted from the external device De. The external device De is, for example, a unique random number generator. The information acquired by the external I / F 90 is transmitted to the recognition unit 92, learning unit 93, other processing unit 94, and estimation unit 95 via the pre-processing unit 91. The other processing unit 94 is a part that performs processing on the information transmitted from the external device De that has been pre-processed by the pre-processing unit 91, which is different from the processing performed by the recognition unit 92, learning unit 93, and estimation unit 95.

[0060] In the right-hand middle processing unit Pb1, for example, when sensory information transmitted from the high-speed detection unit Da4 to the high-speed communication network Na is received by the communication processing unit 100, the sensory information is input to the verbalization processing unit 98 via the communication processing unit 100 and the data receiving unit 101, converted into a data format that can be analyzed by the analysis unit 105, and then input to the analysis unit 105. At this time, if the sensory information is, for example, information that "the robot's right hand is in contact with a sticky object," the analysis unit 105 extracts physical property information from the information. The extracted physical property information includes information on the position where the sensation was detected and physical property information indicating the sensory information. The physical property information indicating the sensory information includes, for example, numerical data for hardness, elasticity, viscosity, etc. The analysis unit 105 transmits the extracted physical property information to the memory controller 104 as an analysis result.

[0061] The action planning unit 106 acquires the analysis results of the analysis unit 105 via the memory controller 104 and the determination unit 96. The action planning unit 106 is configured with a neural network NN as shown in FIG. 6. The action planning unit 106 uses the analysis results of the analysis unit 105 as input information to generate recommended action information to be performed by the right hand unit 35a. The recommended action information includes the movement distance, movement direction, movement speed, etc. of the right hand unit 35a. The recommended action information for the right hand unit 35a generated by the action planning unit 106 is input to the determination unit 96.

[0062] The self-recognition unit 107 acquires the analysis results of the analysis unit 105 via the memory controller 104 and the determination unit 96. The self-recognition unit 107 is configured with a neural network NN as shown in FIG. 6. The self-recognition unit 107 uses the analysis results of the analysis unit 105 as input information to generate recognition information indicating the current situation in which the right hand 35a is placed. The recognition information includes information on whether or not any object is in contact with the right hand 35a, information on the distance from the right hand 35a to any object, and the like. The recognition information of the right hand 35a generated by the self-recognition unit 107 is input to the determination unit 96.

[0063] The determination unit 96 generates final movement information for the right hand 35 a based on the recommended movement information for the right hand 35 a generated by the action planning unit 106 and the recognition information for the right hand 35 a generated by the self-recognition unit 107. For example, when the determination unit 96 determines that the right hand 35 a is in contact with an object based on the recognition information of the right hand 35 a, the determination unit 96 increases the value of the movement speed included in the recommended movement information for the right hand 35 a and transmits the recommended movement information reflecting the correction to the assignment unit 97 as final recommended movement information.

[0064] The assigning unit 97 generates action command information to be performed by the right hand 35a from the final recommended action information for the right hand 35a output from the determining unit 96. Specifically, a meaning library 1040 indicating correspondence between a plurality of pieces of action information and a plurality of pieces of action command information, for example, as shown in Fig. 10, is stored in a storage device controlled by the memory controller 104. The assigning unit 97 identifies action information that is closest to the current recommended action information from the plurality of pieces of action information shown in Fig. 10 based on the final action information for the right hand 35a output from the determining unit 96, and outputs action command information corresponding to the identified action information.

[0065] The language processing unit 98 converts the behavioral command information transmitted from the assigning unit 97 into a format suitable for transmission to the high-speed communication network Na. The language processing unit 98 is configured with a neural network NN as shown in FIG. 6. When behavioral command information such as "Move the right hand 35a upward at a speed of ●●" is transmitted from the assigning unit 97, the language processing unit 98 generates behavioral command information such as "Immediately move the robot's right hand upward." The behavioral command information generated by the language processing unit 98 is transmitted to the high-speed communication network Na via the data output unit 99 and the communication processing unit 100. In this embodiment, the language processing unit 98 is an example of a processing unit abstraction unit that abstracts behavioral command information.

[0066] (Configuration of the drive control unit) Next, the configuration of each of the drive control units Ca1 to Ca3, Cb1, and Cb2 will be described in detail. Note that since each of the drive control units Ca1 to Ca3, Cb1, and Cb2 has the same or similar configuration, the following description will be focused on the configuration of the high-speed drive control unit Ca2 that controls the actuator device 350a of the right-hand unit 35a.

[0067] 11 , the high-speed drive control unit Ca2 includes a pre-processing unit 110, a recognition unit 111, a learning unit 112, an estimation unit 113, a determination unit 114, an assignment unit 115, a verbalization processing unit 116, a data output unit 117, a communication processing unit 118, a data receiving unit 119, a write processing unit 120, a data compression unit 121, a memory controller 122, an analysis unit 123, a control unit 124, a driver 125, and a sensor 126. The recognition unit 111, the learning unit 112, the estimation unit 113, the memory controller 122, and the verbalization processing unit 116 are communicatively connected to one another via a data / memory bus 128. Among the elements shown in FIG. 11 , elements having the same names as those shown in FIG. 5 perform similar operations, and therefore the following description will focus on the differences.

[0068] In the high-speed drive control unit Ca2, for example, when the communication processing unit 118 receives action command information transmitted from the right-hand middle processing unit Pb1 to the high-speed communication network Na, the action command information is input to the language processing unit 116 via the communication processing unit 118 and the data receiving unit 119, converted into a data format processable by the analysis unit 123, and then input to the analysis unit 123. At this time, if the action command information is information such as "Immediately move the robot's right hand upward," the analysis unit 123 generates a control signal for the actuator device 350a from the information. Specifically, the analysis unit 123 sets the drive amount, drive direction, etc. of the actuator device 350a required to immediately move the right hand 35a of the robot device 30 upward, and then transmits a control signal corresponding to the set drive amount and drive direction to the control unit 124.

[0069] The control unit 124 controls the actuator device 350a via the driver 125 based on the control signal transmitted from the analysis unit 123. This causes the actuator device 350a to perform an action according to the action command information.

[0070] The sensor 126 detects the motion state of the right hand 35a of the robot device 30 and outputs a detection signal corresponding to the detected motion state. In this embodiment, the right hand 35a corresponds to the target object. The sensor 126 may also detect the motion state of the actuator device 350a. The detection signal of the sensor 126 is input to the recognition unit 111, the learning unit 112, and the estimation unit 113 via the pre-processing unit 110, and operation result information corresponding to the detection signal of the sensor 126 is generated by the assignment unit 115. This operation result information is transmitted to the high-speed communication network Na via the language processing unit 116, the data output unit 117, and the communication processing unit 118. In this embodiment, the language processing unit 116 is an example of an abstraction unit for a driving unit that abstracts the operation result information.

[0071] The action result information transmitted to the high-speed communication network Na by the high-speed drive control unit Ca2 of the right hand 35a is received by, for example, the communication processing unit 100 shown in FIG. 9 and input to the analysis unit 105 via the data receiving unit 101 and the language processing unit 98, and the analysis unit 105 extracts physical property information. The physical property information extracted by the analysis unit 105 is input to the action planning unit 106 via the memory controller 104 and the determination unit 114. This allows the action planning unit 106 to analyze whether the right hand 35a of the robot device 30 is moving appropriately based on the action command information it transmitted, based on the physical property information. Furthermore, if the right hand 35a of the robot device 30 is not moving appropriately, the action planning unit 106 generates more appropriate recommended action information for the right hand 35a and outputs the generated recommended action information to the determination unit 96. This allows the right hand 35a of the robot device 30 to move more appropriately. In other words, feedback control of the right hand 35a of the robot device 30 is realized.

[0072] (Actions and Effects of the Control Device 10 of the Present Embodiment) As described above, the control device 10 of the present embodiment controls the robot device 30 (mechanical device) provided with the speaker device 43 and the actuator devices 330a, 330b, 350a, and 350b. In the present embodiment, the speaker device 43 and the actuator devices 330a, 330b, 350a, and 350b correspond to a plurality of drive units. The control device 10 includes detection units Da1 to Da5, Db1 to Db8, processing units Pa1 to Pa3, Pb1, Pb2, and Pc1 to Pc3, and drive control units Ca1 to Ca3, Cb1, and Cb2. The detection units Da1 to Da5 and Db1 to Db8 convert the detection signals output from the sensors 40 to 42 and 50 to 59 (detection unit sensors) provided on the robot device 30 into sensory information when the sensors 40 to 42 and 50 to 59 detect predetermined information, verbalize (abstract) the converted sensory information, and transmit it to the communication networks Na and Nb (transmission path). The processing units Pa1 to Pa3, Pb1, Pb2, and Pc1 to Pc3 analyze the sensory information transmitted from the detection units Da1 to Da5 and Db1 to Db8 to the communication networks Na and Nb, and based on the analysis results, generate action command information that is information on an action to be taken by the robot device 30, and verbalize (abstract) the generated action command information and transmit it to the communication networks Na and Nb. The drive control units Ca1 to Ca3, Cb1, and Cb2 analyze the behavioral command information transmitted from the processing units Pa1 to Pa3, Pb1, Pb2, and Pc1 to Pc3 to the communication networks Na and Nb, and control the speaker device 43 and the actuator devices 330a, 330b, 350a, and 350b based on the analyzed behavioral command information.

[0073] With this configuration, the detectors Da1-Da5, Db1-Db8, the processors Pa1-Pa3, Pb1, Pb2, Pc1-Pc3, and the drive controllers Ca1-Ca3, Cb1, Cb2 operate based on verbalized sensory information and behavioral command information, enabling greater versatility compared to conventional control devices that operate using numerical signals or complex control signals. Furthermore, the ability to distribute computing resources also reduces costs and power consumption. Furthermore, compared to devices that operate using numerical signals or complex control signals, the amount of information exchanged between each element can be reduced, significantly reducing communication traffic.

[0074] The detection units Da1 to Da5 and Db1 to Db8 each include sensors 40 to 42 and 50 to 59 (sensors for the detection unit), a recognition unit 71 (recognition unit for the detection unit), an assignment unit 75 (generation unit for the detection unit), a language processing unit 76 (abstraction unit for the detection unit), and a communication processing unit 78 (communication unit for the detection unit). The recognition unit 71 recognizes predetermined information detected by the sensors 40 to 42 and 50 to 59 based on the detection signals of the sensors 40 to 42 and 50 to 59. The assignment unit 75 generates sensory information using the predetermined information recognized by the recognition unit 71. The language processing unit 76 languageizes (abstracts) the sensory information generated by the assignment unit 75. The communication processing unit 78 transmits the sensory information languageized (abstracted) by the assignment unit 75 to the communication networks Na and Nb.

[0075] According to this configuration, it is possible to verbalize predetermined information detected by the sensors 40 to 42 and 50 to 59 and transmit it to the communication networks Na and Nb.

[0076] The detection units Da1 to Da5 and Db1 to Db8 further include a learning unit 72 (detection unit learning unit). The learning unit 72 learns the relationship between the detection signals of the sensors 40 to 42 and 50 to 59 and predetermined information detected by the sensors 40 to 42 and 50 to 59. The assignment unit 75 further uses the learned information of the learning unit 72 to generate sensory information.

[0077] This configuration makes it possible to generate sensory information with higher accuracy.

[0078] The processing units Pa1 to Pa3, Pb1, Pb2, and Pc1 to Pc3 each include a communication processing unit 100 (communication unit for processing unit), an analysis unit 105 (analysis unit for processing unit), a behavior planning unit 106, a self-recognition unit 107, an assignment unit 97 (generation unit for processing unit), and a verbalization processing unit 98 (abstraction unit for processing unit). The communication processing unit 100 receives sensory information from the communication networks Na and Nb. The analysis unit 105 analyzes the sensory information received by the communication processing unit 100. The behavior planning unit 106 plans behavior for the robot device 30 based on the analysis result of the sensory information by the analysis unit 105. The self-recognition unit 107 recognizes the current situation of the robot device 30 based on the analysis result of the sensory information by the analysis unit 105. The assignment unit 97 generates behavior command information using the behavior plan for the robot device 30 planned by the behavior planning unit 106 and the current situation of the robot device 30 recognized by the self-recognition unit 107. The verbalization processing unit 98 verbalizes (abstracts) the behavioral command information generated by the assigning unit 97. The communication processing unit 100 transmits the behavioral command information verbalized (abstracted) by the verbalization processing unit 98 to the communication networks Na and Nb.

[0079] According to this configuration, it is possible to verbalize the behavior command information generated based on the behavior plan of the robot device 30 planned by the behavior planning unit 106 and the current situation of the robot device 30 recognized by the self-recognition unit 107 and transmit it to the communication networks Na and Nb.

[0080] The processing units Pa1 to Pa3, Pb1, Pb2, and Pc1 to Pc3 further include an external I / F 90 (external communication unit) and a recognition unit 92 (processing unit recognition unit). The external I / F 90 receives predetermined external input information transmitted from an external device De. The recognition unit 92 recognizes the predetermined external input information received by the external I / F 90. The assignment unit 97 further uses the predetermined external input information recognized by the recognition unit 92 to generate action command information.

[0081] According to this configuration, it is possible to generate more accurate action command information.

[0082] The processing units Pa1 to Pa3, Pb1, Pb2, and Pc1 to Pc3 further include a learning unit 93 (processing unit learning unit). The learning unit 93 learns the relationship between external input information from the external device De and the action command information. The assigning unit 97 further uses the learning information from the learning unit 93 to generate the action command information.

[0083] According to this configuration, it is possible to generate more accurate action command information.

[0084] The drive control units Ca1 to Ca3, Cb1, and Cb2 each include a communication processing unit 118 (drive unit communication unit), an analysis unit 123 (drive unit analysis unit), and a control unit 124. The communication processing unit 118 receives action command information from the communication networks Na and Nb. The analysis unit 123 analyzes the action command information. The control unit 124 controls the speaker device 43 and the actuator devices 330a, 330b, 350a, and 350b based on the analysis results of the action command information by the analysis unit 123.

[0085] According to this configuration, it is possible to control the speaker device 43 and the actuator devices 330a, 330b, 350a, and 350b based on the action command information.

[0086] The drive control units Ca1 to Ca3, Cb1, and Cb2 each include a sensor 126 (drive unit sensor), a recognition unit 111 (drive unit recognition unit), an assignment unit 115 (drive unit generation unit), and a linguistic processing unit 116 (drive unit abstraction unit). The sensor 126 detects, for example, at least one of the operating states of the actuator devices 330a, 330b, 350a, and 350b and the operating states of the actuator devices 330a, 330b, 350a, and 350b. The recognition unit 111 recognizes at least one of the operating states of the actuator devices 330a, 330b, 350a, and 350b detected by the sensor 126 and the operating states of the actuator devices. The assignment unit 115 generates operation result information using at least one of the operating states of the actuator devices 330a, 330b, 350a, and 350b and the operating states of the actuator devices recognized by the recognition unit 111. The verbalization processing unit 116 verbalizes (abstracts) the operation result information generated by the adding unit 115. The communication processing unit 118 transmits the operation result information verbalized (abstracted) by the verbalization processing unit 116 to the communication networks Na and Nb.

[0087] According to this configuration, it is possible to verbalize the operation result information generated based on at least one of the operating states of the actuator devices 330a, 330b, 350a, 350b and the operating states of the driven objects and transmit it to the communication networks Na and Nb.

[0088] The drive control units Ca1 to Ca3, Cb1, and Cb2 each include a learning unit 112 (drive unit learning unit). The learning unit 112 learns the relationship between the detection signal of the sensor 126 and at least one of the operating states of the actuator devices 330a, 330b, 350a, and 350b and the operating state of the drive target. The assigning unit 115 generates operation result information based on the learning information of the learning unit 112.

[0089] This configuration makes it possible to generate more accurate action result information.

[0090] Second Embodiment Next, a control device 10 according to a second embodiment will be described, focusing on differences from the control device 10 according to the first embodiment.

[0091] 1, the control device 10 of this embodiment further includes a plurality of abnormality detection units Dc and a plurality of abnormality-state drive control units Cc. The plurality of abnormality detection units Dc and the plurality of abnormality-state drive control units Cc are communicatively connected via a control signal line Nc. A plurality of high-speed drive control units Ca, a plurality of low-speed drive control units Cb, a plurality of upper-level processing units Pa, a plurality of middle-level processing units Pb, and a plurality of lower-level processing units Pc are connected to the control signal line Nc.

[0092] The abnormality detection unit Dc is a part that detects abnormalities in the mechanical device 20. The abnormality detection unit Dc has a sensor Sc for detecting abnormalities. The sensor Sc is, for example, a battery status sensor that monitors the status of a battery. The abnormality detection unit Dc transmits abnormality information detected by the sensor Sc to a control signal line Nc.

[0093] The abnormality drive control unit Cc is a part that executes fail-safe control when it receives abnormality information transmitted from the abnormality detection unit Dc to the control signal line Nc, i.e., when some abnormality occurs in the mechanical device 20, to ensure the safety of the mechanical device 20 and maintain the functionality of the mechanical device 20. The abnormality drive control unit Cc controls a fail-safe drive unit Ac provided in the mechanical device 20. The fail-safe drive unit Ac is, for example, a brake device.

[0094] (Specific Configuration of Mechanical Device and Control Device) As shown by the dashed lines in FIG. 2 , the robot device 30 further includes a battery state sensor 130 , an emergency stop switch 131 , and a brake device 132 .

[0095] The battery status sensor 130 is provided in the battery 38 of the robot device 30 and detects the status of the battery 38 .

[0096] The emergency stop switch 131 is a switch that can be operated to forcibly stop the robot device 30 when some abnormality occurs in the robot device 30.

[0097] The brake device 132 is a device that generates a braking force on the soles of the feet of the robot device 30, thereby stopping the movement of the robot device 30. The brake device 132 is used, for example, to prevent the robot device 30 from falling over.

[0098] 3, the upper processing unit Pa1 receives an output signal from the emergency stop switch 131. The upper processing unit Pa1 can directly control the brake device 132 by transmitting a control signal to the brake device 132 via the control signal line Nc.

[0099] The host processing unit Pa3 receives a detection signal from the battery state sensor 130. The host processing unit Pa3 monitors the state of the battery 38 based on the detection signal from the battery state sensor 130, and if any abnormality occurs in the battery 38, transmits information about the abnormality to the control signal line Nc.

[0100] As indicated by the dashed line in FIG. 4, the first sub-processing unit Pc1 can directly control the braking device 132 by transmitting a control signal to the braking device 132 via the control signal line Nc.

[0101] (Configuration of Processing Units) Next, the configuration of each of the processing units Pa1 to Pa3, Pb1, Pb2, Pc1 to Pc3 will be described. Note that since each of the processing units Pa1 to Pa3, Pb1, Pb2, Pc1 to Pc3 has the same or similar configuration, the configuration of the upper processing unit Pa1 will be described below as a representative.

[0102] As shown in FIG. 12, the upper processing unit Pa1 further includes an abnormality processing unit 108 and an abnormality estimation unit 109.

[0103] For example, when the emergency stop switch 131 is turned on, the abnormality processing unit 108 receives a notification indicating that the emergency stop switch 131 has been turned on from the emergency stop switch 131 via the control signal line Nc. Furthermore, when an abnormality occurs in the battery 38, the abnormality processing unit 108 receives a notification indicating this from the upper processing unit Pa3 via the control signal line Nc. The abnormality processing unit 108 transmits these notifications to the abnormality estimation unit 109.

[0104] When the abnormality estimation unit 109 receives a notification that the emergency stop switch 131 has been turned on, or when it receives a notification that an abnormality has occurred in the battery 38, it generates an emergency stop signal to forcibly stop each part of the robot device 30, and transmits the generated emergency stop signal to the control signal line Nc via the abnormality processing unit 108. When this emergency stop signal is transmitted to the brake device 132 via the control signal line Nc, the brake device 132 is activated and the robot device 30 is brought to an emergency stop.

[0105] (Configuration of the drive control unit) Next, the configuration of each of the drive control units Ca1 to Ca3, Cb1, and Cb2 will be described in detail. Note that since each of the drive control units Ca1 to Ca3, Cb1, and Cb2 has the same or similar configuration, the following description will be focused on the configuration of the high-speed drive control unit Ca2 that controls the actuator device 350a of the right-hand unit 35a.

[0106] 13, the high-speed drive control unit Ca2 is further provided with an abnormality processing unit 127. When the abnormality processing unit 127 receives an emergency stop signal via the control signal line Nc, it instructs the control unit 124 to forcibly interrupt and execute processing to bring the actuator device 350a to an emergency stop. The control unit 124 forcibly stops the actuator device 350a based on the instruction from the abnormality processing unit 127.

[0107] The above process is similarly performed in the other drive control units Ca1, Ca3, Cb1, and Cb2, so that the speaker device 43 and the actuator devices 330a, 330b, and 350b are also forcibly stopped.

[0108] (Actions and Effects of the Control Device 10 of the Present Embodiment) As described above, the control device 10 of the present embodiment further includes a battery state sensor 130 and an emergency stop switch 131 (abnormality detection unit) for detecting an abnormality in the robot device 30. When an abnormality in the robot device 30 is detected by the battery state sensor 130 or the emergency stop switch 131, each of the drive control units Ca1 to Ca3, Cb1, and Cb2 forcibly stops the speaker device 43 and the actuator devices 330a, 330b, 350a, and 350b.

[0109] With this configuration, the robot device 30 can be operated more safely.

[0110] Third Embodiment Next, a control device 10 according to a third embodiment will be described, focusing on differences from the control device 10 according to the first embodiment.

[0111] 14 , in the control device 10 of this embodiment, the verbalization processing unit 76 of the high-speed detection unit Da4 has a first neural network NN10. The first neural network NN10 is a neural network that can convert the sensory information output from the assignment unit 75 of the high-speed detection unit Da4 into abstract sensory information. The verbalization processing unit 76 transmits the calculation results of each hidden layer 2001 when the sensory information output from the assignment unit 75 of the high-speed detection unit Da4 is input to each input layer 2000, as well as configuration information of the first neural network NN10 when the calculation results are obtained, to the high-speed communication network Na via the data output unit 77 and the communication processing unit 78.

[0112] The configuration information of the first neural network NN10 includes information on the connection weights between the nodes of each layer, for example, the connection weights wa between the nodes of the plurality of intermediate layers 2001 and the plurality of output layers 2002. １１ ~wa １ｎ , ..., wa ｍ１ ~wa ｍｎ The configuration information of the first neural network NN10 includes information on the connections between nodes in each layer, for example, information indicating which of the multiple output layers 2002 each of the multiple intermediate layers 2001 is connected to.

[0113] 14, the language processing unit 98 of the middle-level processing unit Pb1 also has a first neural network NN10. When the language processing unit 98 acquires configuration information of the first neural network NN10 from the high-speed communication network Na via the communication processing unit 100 and the data receiving unit 101, it reflects the configuration information in its own first neural network NN10. As a result, the first neural network NN10 of the language processing unit 76 of the high-speed detection unit Da4 and the first neural network NN10 of the language processing unit 98 of the middle-level processing unit Pb1 have the same configuration.

[0114] In addition, after the language processing unit 98 shares the configuration of the first neural network NN10 with the language processing unit 76 of the high-speed detection unit Da4 in this manner, it acquires the calculation results of each intermediate layer 2001 from the high-speed communication network Na via the communication processing unit 100 and the data receiving unit 101, and then reflects the calculation results of each intermediate layer 2001 in its own first neural network NN10, thereby acquiring abstracted sensory information from each output layer 2002.

[0115] 15 , in the control device 10 of this embodiment, the verbalization processing unit 98 of the middle-level processing unit Pb1 has a second neural network NN20. The second neural network NN20 is a neural network that can convert the behavioral command information output from the assignment unit 97 of the middle-level processing unit Pb1 into abstract behavioral command information. The verbalization processing unit 98 transmits the calculation results of each intermediate layer 3001 when the behavioral command information output from the assignment unit 97 of the middle-level processing unit Pb1 is input to each input layer 3000, as well as configuration information of the second neural network NN20 when the calculation results are obtained, to the high-speed communication network Na via the data output unit 99 and the communication processing unit 100.

[0116] The configuration information of the second neural network NN20 includes information on the connection weights between the nodes of each layer, for example, the connection weights wb １１ ~wb １ｑ , ..., wb ｐ１ ~wb ｐｑThe configuration information of the second neural network NN20 includes information on the connections between nodes in each layer, for example, information indicating which of the multiple intermediate layers 3001 is connected to which of the multiple output layers 3002.

[0117] 15, the language processing unit 116 of the high-speed drive control unit Ca2 similarly includes a second neural network NN20. When the language processing unit 116 acquires configuration information of the second neural network NN20 transmitted from the middle-level processing unit Pb1 from the high-speed communication network Na via the communication processing unit 118 and the data receiving unit 119, the language processing unit 116 reflects the configuration information in its own second neural network NN20. As a result, the second neural network NN20 included in the language processing unit 98 of the middle-level processing unit Pb1 and the second neural network NN20 included in the language processing unit 116 of the high-speed drive control unit Ca2 have the same configuration.

[0118] In addition, after the language processing unit 116 shares the configuration of the second neural network NN20 with the language processing unit 98 of the middle-level processing unit Pb1 in this manner, it acquires the calculation results of each intermediate layer 3001 sent from the middle-level processing unit Pb1 from the high-speed communication network Na via the communication processing unit 118 and the data receiving unit 119, and then reflects the calculation results of each intermediate layer 3001 in its own second neural network NN20, thereby acquiring abstracted behavioral command information from each output layer 3002.

[0119] The other detectors Da1-Da3, Da5, and Db1-Db8 may have the same or similar configuration as the high-speed detector Da4. The other processors Pa1-Pa3, Pb2, and Pc1-Pc3 may have the same or similar configuration as the middle-level processor Pb1. The other drive controllers Ca1, Ca3, Cb1, and Cb2 may have the same or similar configuration as the high-speed drive controller Ca2.

[0120] (Actions and Effects of the Control Device 10 of the Present Embodiment) As described above, the high-speed detection unit Da4 of the present embodiment has the input layer 2000 and intermediate layer 2001 of the first neural network NN10 for abstracting sensory information, and transmits to the high-speed communication network Na (transmission path) the calculation results of the intermediate layer 2001 when the sensory information is input to the input layer 2000. The middle-level processing unit Pb1 has the intermediate layer 2001 and output layer 2002 of the first neural network NN10, and acquires abstracted sensory information from the output layer 2002 using the calculation results of the intermediate layer 2001 transmitted to the high-speed communication network Na by the high-speed detection unit Da4.

[0121] Even with this configuration, abstracted sensory information can be easily acquired.

[0122] The middle-level processing unit Pb1 of this embodiment has an input layer 3000 and an intermediate layer 3001 of a second neural network NN20 for abstracting behavioral command information, and transmits to the high-speed communication network Na (transmission path) the calculation results of the intermediate layer 3001 when behavioral command information is input to the input layer 3000. The high-speed drive control unit Ca2 has an intermediate layer 3001 and an output layer 3002 of the second neural network NN20, and acquires abstracted behavioral command information from the output layer 3002 using the calculation results of the intermediate layer 3001 transmitted to the high-speed communication network Na by the middle-level processing unit Pb1.

[0123] Even with this configuration, it is possible to acquire abstracted action command information.

[0124] Other Embodiments The present disclosure is not limited to the above specific examples.

[0125] For example, the configuration of the control device 10 described above is not limited to the humanoid robot device 30, but can also be applied to industrial robots, etc. Furthermore, the configuration of the control device 10 described above is not limited to robots, but can also be applied to any mechanical device.

[0126] The language processing units 76, 98, 116 may perform, for example, a process of compressing various pieces of information or a process of symbolizing various pieces of information, instead of a process of verbalizing various pieces of information.

[0127] Design modifications made by a person skilled in the art to the above specific examples as appropriate are also included within the scope of the present disclosure as long as they comprise the features of the present disclosure. The elements of each of the above specific examples, as well as their arrangement, conditions, shape, etc., are not limited to those exemplified and can be modified as appropriate. The elements of each of the above specific examples can be combined as appropriate as long as no technical contradictions arise.

[0128] The neural network NN in each embodiment is not limited to one having a single intermediate layer 1001, but may have multiple intermediate layers. For example, if the neural network NN is a convolutional neural network (CNN), it may have a convolutional layer, a pooling layer, a fully connected layer, etc. as intermediate layers. The same applies to the first neural network NN10 and the second neural network NN20 in the third embodiment.

Claims

1. A control device for controlling a mechanical device provided with a plurality of drive units, the control device comprising: a detection unit that converts a detection signal output from a sensor for a detection unit into sensory information by detecting predetermined information by the sensor for the detection unit provided in the mechanical device, and transmits the converted sensory information to a transmission line; a processing unit that analyzes the sensory information transmitted from the detection unit to the transmission line, generates action command information, which is information on an action to be performed by the mechanical device, based on the analysis result of the sensory information, and transmits the generated action command information to the transmission line; and a drive control unit that analyzes the action command information transmitted from the processing unit to the transmission line and controls the drive unit based on the analyzed action command information.

2. The control device according to claim 1, wherein the detection unit comprises: the sensor for the detection unit; a recognition unit for the detection unit that recognizes the predetermined information detected by the sensor for the detection unit based on a detection signal of the sensor for the detection unit; a generation unit for the detection unit that generates the sensory information using the predetermined information recognized by the recognition unit for the detection unit; an abstraction unit for the detection unit that abstracts the sensory information generated by the generation unit for the detection unit; and a communication unit for the detection unit that transmits the sensory information abstracted by the abstraction unit for the detection unit to the transmission line.

3. The control device according to claim 2, wherein the detection unit further comprises a learning unit for the detection unit that learns a relationship between a detection signal of the sensor for the detection unit and the predetermined information detected by the sensor for the detection unit, and the generation unit for the detection unit generates the sensory information using the learning information of the learning unit for the detection unit.

4. The control device according to claim 3, wherein the recognition unit for the detection unit, the abstraction unit for the detection unit, and the learning unit for the detection unit are configured by a neural network.

5. The processing unit includes a communication unit for the processing unit that receives the sensory information from the transmission path, an analysis unit for the processing unit that analyzes the sensory information received by the communication unit for the processing unit, an action planning unit that plans the actions of the mechanical device based on the analysis result of the sensory information by the analysis unit for the processing unit, a self-recognition unit that recognizes the current situation of the mechanical device based on the analysis result of the sensory information by the analysis unit for the processing unit, a generation unit for the processing unit that generates the action command information using the action plan of the mechanical device planned by the action planning unit and the current situation of the mechanical device recognized by the self-recognition unit, and an abstraction unit for the processing unit that abstracts the action command information generated by the generation unit for the processing unit. The communication unit for the processing unit transmits the action command information abstracted by the abstraction unit for the processing unit to the transmission path. The control device according to claim 1.

6. The processing unit further includes an external communication unit that receives external input information transmitted from an external device, and a recognition unit for the processing unit that recognizes the external input information received by the external communication unit. The generation unit for the processing unit further uses the external input information recognized by the recognition unit for the processing unit to generate the action command information. The control device according to claim 5.

7. The processing unit further includes a learning unit for the processing unit that learns the relationship between the external input information and the action command information. The generation unit for the processing unit further uses the learning information of the learning unit for the processing unit to generate the action command information. The control device according to claim 6.

8. The recognition unit for the processing unit, the abstraction unit for the processing unit, and the learning unit for the processing unit are configured by a neural network. The control device according to claim 7.

9. The drive control unit includes a communication unit for the drive unit that receives the action command information from the transmission path, an analysis unit for the drive unit that analyzes the action command information, and a control unit that controls the drive unit based on the analysis result of the action command information by the analysis unit for the drive unit. The control device according to claim 1.

10. The drive control unit further includes a sensor for the drive unit that detects at least one of the operating state of the drive unit and the operating state of the object to be moved by the drive unit, a recognition unit for the drive unit that recognizes at least one of the operation of the drive unit and the operation of the object detected by the sensor for the drive unit based on the detection signal of the sensor for the drive unit, a generation unit for the drive unit that generates operation result information using at least one of the operation of the drive unit and the operation of the object recognized by the recognition unit for the drive unit, and an abstraction unit for the drive unit that abstracts the operation result information generated by the generation unit for the drive unit. The communication unit for the drive unit transmits the operation result information abstracted by the abstraction unit for the drive unit to the transmission line. The control device according to claim 9.

11. The control unit further includes a learning unit for the drive unit that learns the relationship between the detection signal of the sensor for the drive unit and at least one of the operation of the drive unit and the operation of the object. The generation unit for the drive unit generates the operation result information using the learning information of the learning unit for the drive unit. The control device according to claim 10.

12. The recognition unit for the drive unit, the abstraction unit for the drive unit, and the learning unit for the drive unit are configured by a neural network. The control device according to claim 11.

13. As the processing unit, it includes a plurality of lower-level processing units, a plurality of middle-level processing units whose processing is prioritized over the lower-level processing units, and a plurality of upper-level processing units whose processing is prioritized over the lower-level processing units and the middle-level processing units. The control device according to claim 1.

14. As the detection unit, it includes a plurality of low-speed detection units and a plurality of high-speed detection units that operate faster than the low-speed detection units. The control device according to claim 1.

15. As the drive control unit, it includes a plurality of low-speed drive control units and a plurality of high-speed drive control units that operate faster than the low-speed drive control units. The control device according to claim 1.

16. The control device further includes an abnormality detection unit that detects an abnormality in the mechanical device. When the drive control unit detects an abnormality in the mechanical device by the abnormality detection unit, it forcibly stops the drive unit. The control device according to claim 1.

17. The detection unit has an input layer and an intermediate layer of a first neural network for abstracting the sensory information, transmits the operation result of the intermediate layer when the sensory information is input to the input layer to the transmission path, and the processing unit has the intermediate layer and an output layer of the first neural network, and acquires the abstracted sensory information from the output layer using the operation result of the intermediate layer transmitted to the transmission path by the detection unit. The control device according to claim 1.

18. The processing unit has an input layer and an intermediate layer of a second neural network for abstracting the action command information, transmits the operation result of the intermediate layer when the action command information is input to the input layer to the transmission path, and the drive control unit has the intermediate layer and an output layer of the second neural network, and acquires the abstracted action command information from the output layer using the operation result of the intermediate layer transmitted to the transmission path by the processing unit. The control device according to claim 1.

Images