TW200307592A - Two-leg walking robot and design method for two-leg walking robot - Google Patents

Abstract

The present invention discloses a method for simultaneously designing the motion pattern and the mechanical construction of a two-leg walking robot, which is useful to reduce the time required for designing and to enhance the design precision, comprising plural steps (S1, S2) of representing the two-leg walking robot (1) by various models and designing the robot in accordance with the models. In the plural steps (S1, S2), the design method comprising a step wherein the design data derived in the previous step is referred to in the next step is used. The plurality of steps (S1, S2) comprises a first step and a second step. In the first step the two-leg walking robot is represented by means of a combination model (S11) of cubic parts, and the physical parameter and the motion pattern of the combination model of cubic parts are calculated by using a first evolutionary algorithm. In the second step, the arranged location of constituent parts in each cubic part is calculated by using a second evolutionary algorithm. The evolutionary algorithm may employ a genetic algorithm, while the motion pattern may be generated by a neural network (S12). Whereby, an optimized design can be automatically performed, the design precision may be enhanced and the designing efficiency may be promoted.

Description

200307592 发明 Description of the invention: [Technical field to which the invention belongs] The present invention relates to a design method of a two-legged walking robot that simultaneously designs a motion pattern and a mechanism design, and a two-legged walking robot designed by the design method. [Prior art] The development of robots was performed in the order of specification determination, mechanism design, assembly, operation confirmation, and redesign. First of all, in determining the specifications, 'the size of the robot needs to be set to μ, what action to perform, what kind of motor or sensor to perform when performing the action, and what kind of power system (electric system) device, etc. ~ Y⑺π σ and 5 丨 u Generally speaking, the "plan is to ignore the quality of each element, and only consider the size and dry choice.: = T ° mputer Aided Des •, computer-aided design) The two forces of the structure are surfaced. Then according to the drawing, process, assemble each component, install

Power system equipment, and made into a robot targeted by A Yicha. Then, the production machine crying, L74 L test "work" production, tongue two 'D action confirmation. In most cases " production of various parts inside the active fence will need to be redesigned. Insufficient Temple asks 逑. This day / Xiuhua calculation method also produces crying lamb λ secret "Research on building wheel type is also underway [plus seven]) to express the machine crying man armor in this system with a link model (link evolutionary calculation The method has been published by ^^ 4 tart (magic mu: | ation) i motion mode. ° key, 纟 ° length and each link_ 314618 5 200307592 However, the above-mentioned conventional method uses the evolutionary metering on computer simulation,

In the method of designing a robot, the robot can only be represented by a simple link L ^ main tree, and the sub-error is calculated by the evolutionary calculation method for the long production of each link and the motion mode of each link, so it is _ species It is composed of a majority of households: a practical design method of a robot, which is less practical. v

Therefore, the specific method of simultaneously designing the actual robot's motion mode and mechanism design has not been revealed, and a problem arises that the manufacturing technology of the robot using this design method cannot be realized concretely. In addition, the regulatory decisions and institutional design in the past development process mostly depend on the years of experience and intuition of the development technicians, and in most cases ’need to be redesigned repeatedly, resulting in rising costs. In particular, a two-footed walking robot is composed of many elements and requires a high degree of control. However, because its design technology has not yet been established, it has a problem that it must take a great deal of cost and time to develop. [Summary of the Invention] Therefore, in view of the above-mentioned problems, the present invention provides a design method of a two-footed walking robot that can simultaneously design a motion mode and a mechanism design that help shorten design time and improve design accuracy, and according to the design method, Designed a two-legged walking robot for the purpose. In order to achieve the above object, the design method of the two-legged walking robot of the present invention is formed by: in the procedures for designing the two-legged walking robot, the two-legged walking robot is represented by different models, and most of the programs are designed according to the model. It is characterized in that the majority of procedures include a procedure that refers to the design information obtained in the previous procedure in the next procedure.

6 31461S 200307592 According to the invention and designed because of low development costs. -Different models are used in the overnight program. This can shorten the design time, improve the design accuracy, and reduce the use of the three-dimensional parts. The first evolutionary calculation of the legal parameters and the movement mode is performed! The trick is to arrange it in each three-dimensional zero program. In addition, most programs of the present invention express a two-legged walking robot in combination with a model, and calculate an object program in a combined model of a three-dimensional part; and, by the second term of the component placement position in the second evolutionary piece counting, Because most programs include programs with physical parameters and movement modes of the nose; and the best design in the evolutionary sequence, in order to improve the accuracy of J 2 can be automated, and make the design more efficient. Most programs of the present invention include a two-legged walking robot, and by a second multi-joint model, the program refers to a motion mode program, which is characterized in that it can be lower than most program motion modes. According to the above structure, by Xiguan is a two-legged walking robot that expresses a two-legged walking robot, and the length and motion mode of the bond are performed in the following steps: ΐ Reduce design time, improve design accuracy, and make design efficiency: The physical parameters of the present invention include: the position of each three-dimensional part, the position and the moment of inertia, which are characterized by the obtained physical parameters as the root of the second evolutionary calculation method. According to the above structure, since the physical parameter is used as the target value group in the algorithm, the following two can be shortened: the formula calculates π r 5 and time, and

31461S 7 200307592 In addition, the evolutionary formula of the present invention has used the genetic algorithm to calculate the π method, which can optimize various parameters-the efficiency of the method. Ten different methods are automated and designed. In addition, the kinematics of the present invention can be improved by a neural network. Therefore, the freedom of design can be improved. Measures 1 and 6 are more efficient.

In addition, the weighting coefficients of the neural ^, ^ _ paths of the present invention are obtained by the genetic method and the π method. The neural-like network / finding and adding coefficients are obtained by Gong Zeng, so various parameters and methods can be used to make the design more efficient. The method of calculating the difference is automated, and in addition, the neural network of the present invention is incorporated, and the signaling circuit with positive feedback is operated for 4 s. Grouped and contained as a factorial neural network can be used as a signaling according to the above composition 'It is easy to form a two-legged walking robot: ^ Ding Zuo's neurons, so U Tian's motion pattern. In addition, a signal generating device for a neural network signal such as the present invention. Use this to generate a periodic signal mode. In addition, the two-legged walking robot of the present invention is composed of two human legs. The human model is made up of: three-dimensional parts are used to make the father: the robot, and the first evolutionary calculation method is used to ^ three-dimensional parts. In the combined model, 1 | into 1 sequence, the first order of the physical parameters of the B σ standing parts, and the second evolutionary calculation of the internal structure & + # Τ 开 去 异 异 arranged in each three-dimensional part Made. The design method formed by Ren Xu is a two-legged walking machine with the above-mentioned structure. Since the machine is a master in most programs, it is 314618 8 200307592 with different model performances. It can shorten the design time. , Design accuracy, reduce the cost of development into a talented person ^ '^, and can produce low-cost products. In addition, the ^ f + Η ^ ^ step robot of the present invention is provided with: a god slow element & neural network that operates by including a pair and serving as a circuit having positive daggers and praises; and Ke Yan ,, Gong Wu. ^ The cyclical signal generation device is not in the mode. The device can produce the correct movement footsteps = crying ^ The two-footed walking robot invented by: computer host; two-dimensional parts within ^ Li_1 14; CPU in two-footed walking robot; and three-dimensional The parts are formed by the CPT, and the in and out interfaces are formed. Among them, the Rattles machine uses computer simulation to generate the neural-like, weekly superposition coefficients of the predetermined motion pattern of the > x network. And by using the formula of the pain neural network determined by the weighting coefficient of τ Gongbei, the two-legged walking robot can perform walking motions according to the appropriate motion fork. The two-legged walking robot of the present invention is connected through the aforementioned computer host, and is transmitted through U. Futian 1 ^ ° Every data is received through the interface. In addition, it is best to use the three-dimensional parts and input and output of the two-footed walking robot, and the input and output are separated. Know the access interface 'for function dispersion and load sharing. In addition, the three-dimensional parts of the present invention are preferably formed by TTX 1, W ^, J-field motors and sensors, and the 赭 is generated by the CPU in the two-foot walking robot according to the instructions of the program. Of torque. Or RAM, a two-legged walking machine with the above-mentioned structure. It is known that the neural network 314618 9 200307592 is used to perform the walking action according to the appropriate exercise mode. [Embodiment]

Embodiments of the present invention will be described below with reference to the drawings. The same numbers in the figures represent the same or equivalent. FIG. 1 is a flowchart of a design method of a two-legged walking robot according to an embodiment of the present invention. The design is started. In step S1, the two-footed walking robot 5 is represented by a combination model of three-dimensional parts and is designed based on the combination model of the three-dimensional parts. In step S2, a main part model in which the main part is arranged in the designed three-dimensional part is set, and the configuration design of the main part is performed to complete the design. Fig. 2 is a flowchart of another design method of the two-legged walking robot according to the embodiment of the present invention. At the beginning of the design, in step S3, the two-legged walking robot is represented by a multi-joint link model to design the length and motion mode of each link. In step S4, the two-foot walking robot is represented by a combination model of three-dimensional parts, and is designed based on the combination model of the three-dimensional parts. In step S5, the main part model in which the main part is arranged in the designed three-dimensional part is set, and the configuration design of the main part is performed to complete the design. The difference between the design method in Fig. 2 and the design method described in Fig. 1 lies in the addition of step S3 in which a two-footed robot is represented by a multi-joint link model for design. In step S4, reference is made to the design data obtained in step S3 to perform the design. If this is wrong, the design time in step S4 can be shortened and the design accuracy can be improved. Next, a detailed design procedure of each step will be described. Figure 3 is a detailed flow chart of the design procedure of the three-dimensional part model of κ, knowing shape, and sadness in this issue 10 314618 200307592. It corresponds to step S1 in Figure 1 and step S4 in Figure 2. After designing the opening, In step S11, a model of a two-legged walking robot represented by a combination of three-dimensional parts is set. An example of the model is not shown in Figure 4 and Figure 4. The first figure shows a two-legged walking robot showing the embodiment of the present invention. An example of a combination model of three-dimensional parts. A two-legged walking robot with a three-dimensional shape of a three-dimensional shape with an active unit as shown in FIG. 4 and a 纟 σ-combination-type walking object as the design object. Step § 1 9 Φ, Jie Shi, ^ l & 1 1212, constitute the movement mode of each three-dimensional part, such as the joints of the rumors, the angles, and the neural network (neura) such as ττ 寺network) 〇 Next, using genetic net 瞀, mPr ^ + · Λ Ί /, π method (Genetic Alg0rithm), the neural network-like parameters (such as weighted pass order, cone V, number) and the three-dimensional parts Physical parameters (e.g. position, size, center of gravity罟 Lu position 丨 behavioral moment) is optimized. First, in step S 1 3 Φ, gland bucket, s λ-, the parameters of the pain neural network and the physical parameters of the three-dimensional parts correspond to the chromosomes and 1々, I Therefore, the group of individuals in the period is formed. Then, as described later, the design information of step si S3 in Figure 2 is reflected in the table / number of the initial individuals, so that the design of step S 4 can be achieved. Count the time and improve the precision-Fan Douhan fork is the same as δ and calculate the accuracy. In step S14, in order to complete the design, the loop below 0 is different (1 number of times, it will end when it reaches a predetermined number of times. If it does not reach step S15 When you reach the pre-count number of times, you can then proceed to the fitness evaluation of each individual in step S15. At this time, 314618 11 200307592 fitness is set on the 4th, such as biped walking, energy consumption, and interference between parts. Waiting for design = walking distance of steps' 1 work restriction Jg wo n’t need to reflect the fitness directly.. / ,,,,,, and timing machine functions to expand or reduce the difference in fitness You can also use V in step s16 and choose one At present, there are several methods for selection by the method of selection. At this time, the probability of the fitness obtained by the individual selection is selected. 'According to step S17, the selected intersection position determination method is currently in the current state. There are several methods at this time, for example, it is possible to randomly determine a cross, 0. For example, individuals form new individuals. Sub-tables are replaced in step S18, causing individuals to have mutations 11). That is, a well-known mutation method has several methods. At present, there are several methods to calculate the distance between two individuals created by crossover. The closer the distance is, the more likely the method of θ is to produce a mutation. Calculate the number of loops' When the predetermined number of times is reached, the parameters of the neural network such as the optimization and the best physical parameters of the three-dimensional parts can be obtained in the Γ version of the gene whose fitness is optimized. The parameters of this type of neural network, as well as the three-dimensional parts if the central axis ^ σ majority, can be provided as a reference for the design information in step S2 or step S5 of the next procedure. By performing the above steps with κ, a combination model of three-dimensional parts can be used to design a scoop two-legged walking robot with an optimal motion pattern plate 314618 200307592. The figure shows the details of the main part design procedure of the embodiment of the present invention. "Tian Heng Wang Tu" corresponds to step S2 in Figure 1. At the beginning of the design, it is set to v ^ S5 phase # ^ φ ^ in step θ in step S21, and the layout of the main parts arranged in the three-dimensional part is set. An example of the edge type is shown in FIG. 6A. The selector is arranged in each of the main parts of the present invention. The "happiness" three-dimensional part is shown by the tooth h. The rotation of the three-dimensional part 100 The shaft 101 is conveyed to I ::: 1.5. It conveys the ㈣ of the motor and measures the three-dimensional parts; ::: to perform the rotation. The sensor 10 is used to detect the position of the… part 1. Then, by The genetic algorithm optimizes the configuration of the main components as a whole. The internal factors ΠΓ, 2: 'make the design parameters correspond to the chromosomes and are arranged on the basis of 102, # 二 卿 ... The design parameters refer to, for example, the system For horse 103 > 1〇4ν ιπς a, at the position of Dou 105 and sensor 106. Square; in step S23, in order to pull the bundle _ and reach the predetermined earth potential $; bundle 6 leaves, count nose For the following number of loops, go to step S24. "... stop. When the number of times has not been reached, then, from < ^ S24, the fitness for each individual is the setting and [[degrees of ... at this time, the reason for the design process in step S 4] Step S] or the design data of the two-stroke best motion mode and the two moments of the mechanism of the foot walking robot shown in FIG. 2. The size, gravity center position, inertia, and middle, the obtained fitness is not rate, but can also be imported into the heart directly reflected in the 4 timing, can be function to expand or reduce the difference in fitness. 3 丨 46] 8 13 200307592 In step S25, the individual for mating is selected. There are several ways to choose this method. For example, it can be calculated according to the probability corresponding to the obtained fitness. In step S26, the selected individuals enter each other. How to decide the position ... Intersection: For example, the position of an intersection can be randomly determined, and before and after = examples, a new individual is formed. We think that in step S27, the individual is mutated. That is, one-change gene. There are several ways in which this square, square, square, square, square, square, square, and square are present; there are several methods that have been used in Muli by people, such as the calculation of the distance between two individuals and the distance between them. Example A method with a higher probability of generating a mutation. ~, I am afraid that I will use 丄 to calculate the number of loops. When the predetermined number is reached, the optimized design parameters in the individual genes of the station will be obtained. The most suitable is that by performing the above steps on each three-dimensional part, there is a presence.目 之 ㈣SL „2 目 之 ㈣ ;; ^ The closest to the shovel obtained in S ^ The design program of S4 walking robot, * 4 means a machine ^ order ^ two-legged person. 乂 豕 ... Also plan to make the final form Figure 7 of the two-legged walking robot is a detailed flowchart of the link mold of the embodiment of the present invention. The details of the #, 孓 °, and Ye program are, in which, the multi-joint link is expressed as ... after the 'walking robot,' # — Χ is in the two feet of 6 and 60 objects. The model is palpitated. An example of the model is shown in Figure 8. Figure 8 illustrates a two-legged walking robot that illustrates the embodiment of the present invention. 314618 One of the joint model of the joint ^ Zv〗 picture. Using the joint board 4 with the joint 5 to represent the two-footed walking robot 3. In step 7 of Fig. 7, the key can be formed to rotate each key The movement mode is, for example, the opening angle of Guan Lang ^ — a neural network such as Torque Temple. The receiver 'hunting by the value, the palace's insult, the method of the neural network-like parameters (such as weighted Ding') Number) and the multi-joint model $ each link length is optimized. In step S33, the detection volume „μ in your Xiguanlang model The length of the keycuffs corresponds to the number of gods, kung fu, kuro, and the number of wheat in Kushiro. The multi-segment $ group one version of the subgroup is placed in the gene to generate the initial one in step S 3 4 in order to be out of control ... Beam design. Calculate the following number of loops. After reaching the predetermined number of times, Ya ceases to fire at step S35. ® If it has not reached the predetermined number of times, it will go to the involved and evaluate the fitness of each individual. This is suitable, for example, to set the walking distance of one step of a two-footed walking robot.

Probability when choosing, and the adaptability of differentiating == does not require a direct backshift difference. The function can be expanded or reduced to adapt to J. In step S36, send Lantian ,,,,,,,,,, and, the one selected by the parent. At this time, in the method of selecting which individual to refer to, there are already several methods known to people. For example, you can choose according to the probability that you have obtained the tadpoles, 7 shi and heart rate. In S37, there are several methods for determining the positions of the selected individuals in each other. Yi decides a crossing position randomly and knows it. For example, available individuals.奂 Individuals form a new 314618 15 200307592: Step S3 ", causing the individual to mutate. That is to say, the fixed mutation method has been used to determine the distance between the two individuals that have been created by crossover. The closer the distance is, the more likely it is that the mutation will occur at the same rate. When the number of ten different loops reaches IP 胄 ϋ + Gubei 古 A number 蚪, the optimal design parameters can be obtained in the individual genes with the most fitness. By implementing the above steps, the passenger and the head can be designed as a two-legged walking robot that is represented by a link that is connected with the Xiguan. It has the optimal link length and motion mode in step S4 in Figure 2. 蕤 月 扁 _ ☆ When the individual is formed in the initial stage of the advancement and the fitness is evaluated, the optimized capital is calculated. The length and exercise mode are set. The test of Shibei Cang will improve the design time and design accuracy in S4. Next, we will explain the structure of a neural-like network—the neural dimension], Fan & a α 9 Figure k Select page does not hunt the two pictures of the motion pattern generated by the class. / The model of the human link model indicates that the joint 41 is connected to the link 51, the chain 52, and the link 52, and the joint 42 is to link the link 52 to the link 53. The knot 53 and the link 54 are connected, and the joint 41 is not connected to the end of the golden knot #, t & side of the link 5 1 or one M that is 43 is not connected to one side of the taxi 4 ^ μ # ^ # The ends of ~ 4 are grounded to the ground 50. According to the example in FIG. 9, 处 Λ, / ν ^ ^ μ 51 at the chain is grounded to the ground 50. Here, the opening angles of the joints 41, 42 A, 〇 ^ a are respectively set to Θ], 2, Θ 3, and the joints 4 1, 4, 2, and 4 3, respectively. Set to T], IV D3,, chain '% 5 1, 54 from the ground 5 reaction forces are set to Exj, 3146J8] 6 200307592

Ex2 〇 Brother 10 is based on the link-like neural network hibiscus shown in Figure 9. Xi 4, g, ▲ γ, the k-path composition of the motion pattern. This type of neural network is a neural network with a double-layered structure such as Yushan's double-layered structure, with the angles of inversion ㊀, ㊀, Ω £ χ], Εχ2, and 1w 2 y 3 as the points of penetration. Giant T], Ding 2 and Ding 3 exist in the neurons of the middle layer NA], NA, · · All the input points are combined, and the nerves 7L ΝΒ that are present in the NM and the interlayer, 2, ... Combined with all rounds. In addition, the neuron meridians existing in the output layer Ν ,, Να2, ..., ΝΑ phase "2 series knife and god. ,, 7t _ χ η phase ,, '° δ, and the nerves that exist in the output layer The Wu NTS is combined with the neurons NBi, nr n n] NB2. ·., Then „. ^ Warp element NT], NT2, and T2 respectively output torque D, T2, and T3. Figure 11 (a) is a schematic diagram showing the combination of neurons existing in the middle layer ... Ω). When the neuron wealth is set to A ', the input weights of M, EΥ θ], θ2, and θ3 are respectively set to wk], wk2, wk3, wk4, and%, which are expressed by the following formula ⑴. PAk l + e'QAk (1) Here, QAk is a medium variable and is expressed by the following formula (2). QAk = wk rEXl + wk 2.Ex2 + wk 3 · ^ 1 + wk 4 ^ 2 + wk ye ,,, -3 (2) The 11th (b) picture shows the neurons existing in the middle layer, 2 ..... n) combined schematic. Set the output of neuron NBk to PBk, Εχ], Εχ2, Θ], ㊀2, ㊀3, and the input weightings can be represented by _, wk], wk5, _wk4, and wk3, respectively. \ ι 314618 200307592 So-here, in the bond model shown in Figure 9, the key joint C, the link 52 and the link 54 3 correspond to each other, and form a shake, which is opposite to the two-phase phase I. Therefore, PBk is expressed by Equation (3). l + e'QBk (3) Here, QBk is a media variable, which is represented by the following formula (4) PB ^ No QBk = wkrEx2 + Wk2.Exi + Wk3 ^ _Wk 4., 2 + ^ 5 ^ 乐乐 U (C) The graph shows the neurons NT present in the output layer. 纟 士 八: Intent: Set the output of the neuron% to the torque W God ::

Fortunately for Ak, the weighting of PAk is set as w, and the following formula 数 is used. Xiao 4 Ding

T (5 l + e'QTi is shown by the following formula (6), QT] is the media variable (6) QTi-iwAk.PAk k = l The "(d) picture shows that The nerves of the emergence layer N N are not thought about. Set the neuron NT order 6 6 and the NT 2 wheel output as the torque NAk wheel output plus the PAl plus 8 'will be the nerve. When the knife type is set to wC, the following formula (7) is expressed. 钤 moment D 2 is represented by l + e'QT2 ⑺ Here, the team is a media variable, which is represented by the following formula (8). 314618 18 ZZΌΌόΌ / ^ Z η 叩 ck.PAk (8) Brother 1 1 (e) The picture is very-schematic. The combination of the nerve: the proposed neuron N τ3 is called the output pair PBk: " Execution 22, is the torque Τ3, the neuron is expressed by the following formula… ... When you do n’t set it to ^, the torque is 3 series, T: 1 + ^ QT3 (9): The second series match her, which is expressed by the following formula (1.) ( 10) QT3 ^ JwBk.PB kl shows the opening angle of the neural network according to the above description ㊀ ^ ㊀ "^ and reaction two 'can form joint points, and output joint X] Εχ2 as the torque of wheel-in P卩, T2, & ^ Inside and outside the road, as explained above Inheritance ",,: ,, ## The weighted 内, 你 method in the road can make the neuron-like j 入 large purchase k, WCk (k = 1 to η) in the gray gene of Lu Biyu to determine &Quot; "in Neural Networks

(k = 1 to η). With this weighted "WAk, ~ wC, what the robot needs to do when walking? = Can generate two-foot steps. Then we will explain the neural network-like 1: periodic exercise mode. The two steps of the neural-like network to generate the movement mode : The machine V2 diagram is a schematic diagram of the model. ^ 仃 Other link mold joints 61, 62, 63, 64, 65 of the dispel person are connected to each other and their adjacent joints by bonds 6:68, 69, By the connection, the link with the toeboard 7] 314618 19 200307592 is connected to the joint 61, and pe 7Π ^ 'is connected, and the chain agency with the toeboard 72 is connected to 70. In addition, the above Yi, 0, and Yuedan 73 are combined with Guanlang 65, 66. In addition, the chain between the two lines is 62, 63, 64, 67, 68, 69. The joint 61 and Wei X ζ have a degree. In addition, the joint 62-W surface has a self-buffering 9 series that is bent so that the toe board 7 i and the toe board 72 form a parallel angle with the ground. H In addition, 'joint 61, 6 5, 6 & 0 shape 65 Φ 66, 70 series is set to be able to perform period such as sine wave waveform Sportsman. ^ 'In addition, the off is 61, 62, 63, 〇6 "67, 68, 69, 70" 64 ^ φ4, 、, φ9, φ]. The angle system is set to φ], Φ brother Figure 1j is based on other types of neural crest-type vomiting fields in the U-th graph of the U-th graph of the ten objects, ^ ^ ,, χ, and the structure diagram. The signal generator 80 Periodic waveforms such as sine wave, k > 1 ^ 'to supply the joint μ corresponding to the left and right waist joints of the two-legged walking machine crying-signal at an angle of φ6, and at the same time 4 angle ° 5, joint 66 shape The corresponding joints 61 of the foot joints form the left and right sides of the φ chat-step robot A, ^ form an angle Φ], the joints 70 form the signal neurons 8 1 of the angular production φ 10, and the Shenkaikai 84 series correspond to the two foot steps, respectively. Crying machine worker, neuron 83, and neuron force di ... ⑼Read Jin, Λ neurons of the left and right femoral joints. Use alien 89 to reduce nasal neurons 8 丨 fish ~ According to the result of the reduction, provide the angle of 04 to Low: the output signal of ^ element 82, and Genzen, AQ,, 4 give joints 64. Use the addition to cry 90 and reduce the neuron 83 84 + neurons wheel "^ for angle section 67 φ7ιι. Number, and based on the subtraction results, in addition, ‘Neural 85 and God ’s 8 R ^ ,,, f, r ^ Gongwu 86, Neuron ^, and Neural 8 8 series correspond to two-legged walking machines crying. . The nerves of the left and right knee joints 314618 20 200307592 yuan. The adder 91 subtracts the output signals of the neuron 85 and the neuron 86, and provides an angle Φ3 to the joint 63 according to the result of the subtraction. The output signals of the neuron 87 and the neuron 88 are subtracted using the addition cry 92, and the angle Φ 8 is provided to the joint 6 8 according to the subtraction result. In addition, neuron 8 i and neuron 8 2, neuron 83 and neuron 84, neuron 85 and neuron 86, sacral muscle 87, and neuron 88, neuron 82 «and neuron 84 are respectively, It also operates as a signaling circuit with positive feedback. "Receiver", the inner structure of the neuron of brother i. The internal structural block diagram of the neuron is shown in Fig. 14-Gang, X, and X. Neuron 93 is an input signal L / y Γ Ul 5; it is a combination of binding, 0 (white: (black circle on the input signal represents a circle)) is a representative excitation binding. In the m neuron 83 in the figure, U is omitted from the neurons. , Heart] only exists in the output of the same value: "Table in all the neurons in the majority of the output points of the calculation of the formula 03) ° ^ operator 94 performs the formula ⑴), formula U2),

du dt dv dTf (v) + U0 + u2 (11) y (12)

Among them, f (X) (1 3), where u 俜 phase fire is caused by fatigue 谇 μ, the internal variable of the membrane potential of the 4th meridian, and the internal pressure of Nadu

The text τ 1 and τ 2 are time constants, and t J 314618 21 200307592 is the weighting coefficient, and the various systems are misunderstood. E According to the above description, "Tian, 'the coefficient of Pi Lao degree, k is a constant. 62, 63, 64, 65 The structure of the neural network can output joints 61, Φ 3, ① 4, 0) 5, angles of φ > φ,, 〇 Φ 1, φ 2, 6 Φ7, Φ *, Φ9, Φ] ... In the first, by using the above-mentioned genetics, the gods in the first path can open up a wide range, and open the method, so that the song can be applied to the neural network. Configured at

In addition, the weighting of the neural network is determined, and the soil road can generate a two-legged walking robot. Second, the periodic motion pattern. … The week of the joints needed for the movement. In the above description, the genetic algorithm is used as the evolutionary counting method, but it is not limited to this kind of algorithm. For the evolutionary calculation: method, for example, Using evolutionary strategies ("I: uonary S⑽egies (ES)") that focus on mutations, focusing on the evolution of species rather than the individual's Evolutionary Programming (Ep), and using tree-like structures to process genetic information Genetic Programming (Gp) ^, and can perform the same calculations as genetic algorithms. Next, the two-legged walking robot designed according to the above design method will be described. Fig. 15 is a block diagram of a two-legged walking robot according to an embodiment of the present invention. The two-legged walking robot 120 is composed of a large number of three-dimensional parts 100, but only one of them is shown here. In addition, the figure shows only one set of CPU (Central Processing Unit), (Read Only Memory) 112, RAM (Raildom Access Memory, Random Access &丨 Thinking) 11 3. Fortunately, don't enter or exit the interface 11 4, but you can also configure multiple groups inside the robot 1 20 as needed to form a structure that can achieve the functions of 314618 200307592 dispersion and load knife. The motor 102 and the sensor 106 are built into the three-dimensional component 100 and connected to the input / output interface 114. The CPU 111, ROMU2, RAMU3, input / output interface 114, and communication interface 115 are connected by using a busbar ΐ6, and are built in the two-footed walking robot 120. In addition, the two-legged walking robot 120 is connected to the computer host 130 via a communication line (common line) U7, and its data is sent and received using the communication interface i $. The host computer 130 executes the aforementioned design method by computer simulation, and determines the weighting coefficients of the neural network such as the predicted motion pattern. Then, a program for implementing a neural network determined and determined by the weighting coefficient is transmitted to the RAM 1 1 3 through the communication line 117. Here, the communication line 117 is wired, but may be wireless. In addition, the program designed by the host computer 130 can be written into the non-volatile memory, and it can also be ROM112. In addition, the program can also be written into the external storage medium (not shown) in the host computer 130, and the two-legged walking robot 120 can read the formula using an external storage medium reading device (not shown) and Transfer to RAM113 〇Λ 王 丨 y cut soil TF inches, if necessary, remove the connection line in, so that the CPU ⑴ executes the program built in RoMn2 or ^ 'and through the input / output interface 114 to the sensor 1〇 6 Measure the body = the movement of the piece, and drive the motor 1G2 to generate the torque of the three-dimensional part. By this, the two-foot walking robot 120 can perform the walking action according to the appropriate motion mode indicated by the program. As can be understood from the above description, according to the present invention, in most processes 314618 23 200307 ^ 2, different models are used to express and the design time is shortened to increase the k-steps of the robot, so it can be reduced. In addition, due to the present invention. Second, reduce development costs. The program of the physical parameters and motion modes; and =: = the calculation method to calculate the configuration position of the object into parts. ^ The calculation method from the calculation method to obtain a good design to improve the design accuracy, the invention of each order t Multi-joint model performance two feet = make: plan efficiency. In addition, the length and movement mode are combined with a stepping light robot, and the design time of each link is designed to improve the design precision for reference. Therefore, the physical power obtained during the design can be reduced. In addition, because of the target value group: "At first, it provides the accuracy of the second evolutionary calculation method. In addition, the design time of = :: 1 sequence is improved, and the algorithm used to design the sigh leaves 0 is derived. It is possible to make the 夂 ... / law system based on genetic calculations, and to make 嗖 'Γ 玄 ° * the calculation method for optimizing the automatic neural network outside, because the operation system is classified by generating. Therefore, the design can be improved. Degree of freedom and design efficiency. In addition, due to the weighting system of the neural network-like: 2 ::: The calculation method for optimizing the force and number is improved from the second circuit. In addition, the neural network includes It can be used as a signaling neuron for Ding action, so it is easier for a biped walking machine to cry and generate an appropriate exercise mode. In addition, because of the god-like "road-attached person's nails, the road is equipped with a device that can generate periodic movements" Therefore, it is easier to produce a two-legged walking robot. In addition, 'two-legged walking robots have different wheel shapes and designs in most programs, so design time can be shortened. 3) 46) 8 24 200307592' two-legged walking robots can generate correct P-low development Cost-effective products. In the meantime, people are equipped with similar neural networks and signal generating devices, and exercise modes. Industrial applicability ^ As mentioned above, the present invention can be helpful at the same time when designing a two-legged walking robot; the couple can shorten the design time and improve the design accuracy ^ M, η. ^ Also < Machine, and by shortening the design time, improving the design accuracy, and lowering the development cost, it can contribute to the rain foot walking robot. '' Simple illustration of seat block L diagram】 The present invention can be further understood based on the additional drawings used to explain the following detailed invention and embodiments of the present invention. Second, the embodiments shown in the attached drawings do not specifically or limit the present invention, and are only intended to facilitate the description and understanding of the present invention. Method = is the design of a two-legged walking robot according to the embodiment of the present invention. Figure 2 is a flowchart of the design method of the present invention. Figure 3 is a detailed flowchart of the present invention. Others of the two-legged walking robot of the Ming embodiment The program for designing a three-dimensional part model of the Ming embodiment is shown in Fig. 4. An example of the model of the embodiment of the present invention is shown in combination with the three-dimensional parts. Figure 5 of the two-foot walking robot is detailed flow chart of the present invention. 314618 25 200307592 _'shows an example of a built-in model of a three-dimensional part arranged in the embodiment of the present invention. The details of the link model design procedure of the embodiment of the present invention. Brother 8 Picture Biography θ _ No * is not a two-legged walking machine according to the embodiment of the present invention.

Xi Yue said 丄 J,; ^ ΟΠ / V Xiguan: An example of a link model.

Cyborg II 3 uses a neural network to generate a two-foot walking machine with a motion pattern.苐 1 〇 图 私 』. , On the structural map. τ, neural networks such as generating motion patterns according to the link model 1 1 U) to (e) The graph shows that the combination of neurons that exist in the middle is -4M, Gongfan, and the layered neurons. schematic diagram. Figure 12 is a schematic diagram of the motion model generated by a neural network-like wailing, flying, and two-foot walking.

Figure 6 shows the main parts and Figure 7 is a detailed flowchart. Figure 13 is based on other link models to generate other types of W, Gonglu road structure diagrams of Ning Tong Di m ^ Le chess. Figure 14 is a block diagram of the internal structure of a neuron. Block structure ::. The figure is the two-legged walking robot of the embodiment of the present invention. The three-dimensional parts are jointed with 71 and 72 finger boards. The two-legged walking robot is 4 '41 to 43, 61 to 70 5, 51, 52, 53, and 54. 314618 26 50 200307592 communication interface communication line computer host, NA2, NAn, NB1, NB2, wCn y 73 upper body 81 to 88, 93 neuron 94 computing unit 102 motor

111 CPU

113 RAM 1 15 117 130 ΝΑ PAk, PBk outputs wkl to wl (5, wAn, wBn, xl to xn, u0, ul input signals 0 1 to 0 3 Joint opening angle 80 Signal generator 89 to 92 Adder 100 Three-dimensional part 106 sensor

112 ROM 114 I / O interface 116 Port 120 Two-foot walking robot

Ex 1, Ex2 reaction force NBn, NT1 to NT3 neurons T1, T2, T3 torque force. Weight Output signal 0 1 to < M0 Joint angle 27 314618

Claims (1)

200307592 Scope of patent application: 1. A design method of a two-legged walking robot, which consists of different models of the two-legged walking robot in the various programs for designing the two-legged walking robot, and according to the majority of programs designed by the foregoing model Formed, in which the above-mentioned majority of programs include: a program in the next program that refers to the design data obtained in the previous program. 2. According to the design method of the two-legged walking robot in item 1 of the scope of patent application, most of the foregoing procedures include: using a combination model of three-dimensional parts to represent the two-legged walking robot, and using the first evolutionary calculation method Calculate the physical parameter disc in the combined model of the aforementioned three-dimensional part: the first program of the dynamic mode; and calculate by the second evolutionary calculation method = the second column order of the component placement positions placed in the aforementioned three-dimensional parts. 3: The design procedure of the two-legged walking robot described in item 1 of Shenli's scope includes most of the procedures, which include: the multi-joint model is shown as the two-legged walking robot, and the chains in the aforementioned multi-joint model are as follows: / 2 3 evolutionary calculation method sequence, where, in the next-program ^ = range mode of motion mode. The length and movement of each link 4. Ru Shen's patented two-foot method of Weiwei Chu target sibling, in which the number of objects in the design of the front-line family is based on the package, size, and center of gravity. The physical parameters obtained in the position and inertial force 4 / position of the three-dimensional parts will be set at the target value of the first program method described above. … 疋 L ,,, and 5 describe the second evolutionary calculation 31461S 28 ^ uju / ^ 92 5. For the design method of the robot in the scope of the patent application, the two-leg walk teleportation algorithm of any of the two items of music. / Among them, the above-mentioned evolutionary calculation method is applied. 6. If the patent application scope is the second method, in which the former ^ worker ’s two-foot walking robot is set 7 • If the patent application scope is #the dynamic mode is through a neural network. The road is formed. Method, in which the two-step f robot design method of the aforementioned deities is obtained. The weighting coefficient of Zhou Lu is based on genetic calculations. For example, the 6th method of the scope of patent application, the design method of the two walking robots of the two members, where the neural network system includes: group And as a signal circuit with positive feedback, and as a neuron with 9, ^ ^ into the neuron. 9 · For item 6 of the scope of patent application, the designer of Bengans _ Beizhi's two-legged walking robot ... In the above, the signal generating device of neural network is described. The name j produces periodic messages. 10. A kind of two-foot walking robot, which performs two-foot walking. "/, Read by: Modeling the combination of three-dimensional parts with the combination of three-dimensional parts; 2 :: The first evolutionary calculation method Two of the physical parameters of the three-dimensional parts in the front M and i, and the design method of calculating the weights of the component parts arranged in the above-mentioned various moon beans by the second evolutionary calculation method. 2㈣Formed Π.A A two-legged walking robot is provided with a neural-like network including pairs of neurons that act as a signaling circuit with positive return money, and a signal generation circuit that generates periodic signals. Motion mode generating device. 314618 29 200307592 12. A two-legged walking robot, which is composed of three-dimensional disasters in humans # · Λ m Xunzhuan, two-legged walking machine parts, Γρη three-dimensional parts in two-legged walking robots, as described above epU PU, and the formation of the above-mentioned knowledge access interface, the order that the mainframe of the computer is to use the meter to predict the IMW of the predetermined motion mode will definitely produce a pre-arsenic experience. The weighting coefficient of the road is used to implement the program by adding the road t, so that the neural net-like walking action is performed.-The person pays according to the appropriate exercise mode 13 = two of the 12th scope of the patent application A foot walking robot, among which, the first four two-foot walking load, the crying brain host, the car's communication line or wirelessly communicate with the aforementioned electricity 14 ^ 妾 'and send and receive data through the communication interface. The two-legged walking robot of scope item 12, wherein a plurality of groups of the aforementioned three-dimensional parts and front i 5 遨 spoke access interfaces are arranged in the two-legged walking robot for function dispersion and load dispersion. Item 12 or Item 14: A two-legged walking robot, Ganshan 'The aforementioned three-dimensional parts are formed by a motor and a sensor, respectively, and are generated by the CPU and ROM or RAM in the two-legged walking robot. Turn around as instructed in the previous procedure. 30 14618