JP4453076B2 - Humanoid robot - Google Patents

Description

  The present invention relates to a humanoid robot and a map creation method, and particularly relates to, for example, a humanoid robot that recognizes an environment and performs an action suitable for the environment, and associates an environment, an action, and a sensor output for such a humanoid robot. The present invention relates to a map creation method for creating a map.

For example, a biped walking humanoid robot such as ASIMO (trade name) introduced in Non-Patent Document 1 may fall over and collapse if it cannot recognize the environment. For this reason, for example, it is not possible to appropriately take action and then recognize the environment, so it is necessary to recognize the environment where the robot is placed before the action. In conventional humanoid robots, in addition to the environment, many sensors are provided for recognizing humans and obstacles. Therefore, the outputs of all sensors are used to recognize the environment before acting.
“ASIMO SPECIAL SITE”, Honda Motor Co., Ltd., Internet <URL: http://www.honda.co.jp/ASIMO/>

  In the prior art, many sensors are required for environment recognition, and thus there are problems in terms of cost, wiring, and power consumption.

  Therefore, a main object of the present invention is to associate a humanoid robot capable of recognizing the environment even if the number and types of sensors are small, and to associate the environment, behavior, and sensor output for such a humanoid robot. It is to provide a map creation method for creating a map.

  The invention of claim 1 is a humanoid robot that recognizes the environment, and stores a map in which a sensor, the environment in which the humanoid robot is placed, the behavior of the humanoid robot, and time-series data of the output of the sensor are associated with each other Storage means, first execution means for executing an action for environment recognition, and first measurement means for measuring time series data of an output of a sensor when the action is executed by the first execution means, A humanoid robot comprising recognition means for recognizing an environment based on time-series data output from a sensor measured by a first measurement means and a decision tree based on the environment created based on a map. .

  In the invention of claim 1, the humanoid robot (hereinafter also simply referred to as a robot) recognizes the environment and includes a sensor. The map storage means stores a map in which the environment in which the robot is placed, the behavior of the robot, and the time-series data of the sensor output are associated with each other. Recognize which environment is working. The environment is recognized by the recognition means. That is, the first execution unit of the recognition unit executes an action for environment recognition, and the first measurement unit measures time-series data of the sensor output when the first execution unit executes the action. Then, the recognizing unit recognizes the environment based on the time series data of the sensor output measured by the first measuring unit and the decision tree having the environment created based on the map as a discrimination target. The decision tree defines the order of application of the decision procedure for determining the environment in which the robot exists based on the action performed by the robot and the sensor output at that time. Therefore, the recognition means can recognize the environment by repeating the execution of the action and the determination based on the sensor output at that time according to the decision tree until the environment is determined in the decision tree. Therefore, according to the first aspect of the present invention, since the map in which the time series data of the sensor output is associated with the behavior of the robot and the environment is provided, the environment created based on the map can be obtained even if the number of sensors is small. The environment can be recognized based on the decision tree as the discrimination target.

  The invention of claim 2 is dependent on the invention of claim 1, and the recognizing means further includes a decision tree creating means for creating a decision tree having the environment as a discrimination target based on the map.

  In the invention according to claim 2, the decision tree creating means creates a decision tree having the environment as a discrimination target based on the map. This makes it possible to create a decision tree based on the map each time an attempt is made to recognize the environment. Therefore, according to the invention of claim 2, when performing environment recognition, the first action can be arbitrarily selected.

  The invention of claim 3 is dependent on the invention of claim 1 or 2, wherein the sensor includes an acceleration sensor, the action includes at least one of walking, sleeping, and getting up, and the environment to be recognized is at least Classified by floor hardness or floor slip.

  In the invention of claim 3, the sensor provided in the robot includes an acceleration sensor, the action performed by the robot includes at least one of walking, sleeping, and getting up, and the environment to be recognized is at least a floor Are classified according to their hardness or floor slip. Therefore, according to the invention of claim 3, by measuring the time series data of the output of the acceleration sensor when performing an action involving contact with the floor such as walking, lying down, and getting up, the floor is measured. It is possible to recognize an environment having a difference such as hardness or floor slip (friction).

  The invention according to claim 4 is dependent on any one of claims 1 to 3, and further comprises second execution means for executing an action suitable for the environment recognized by the recognition means.

  In the invention of claim 4, the second execution means executes an action suitable for the environment recognized by the recognition means. Therefore, according to the fourth aspect of the invention, after recognizing the environment, an action suitable for the environment can be executed, so that a desired operation can be performed according to the environment.

  The invention according to claim 5 is dependent on the invention according to claim 4, and includes a second measuring unit that measures time-series data of the output of the sensor when the action is being executed by the second executing unit, and a second measuring unit. Environment change determination means for determining whether the time-series data of the measured sensor output corresponds to the environment recognized by the recognition means and the action executed by the second execution means based on the map When the environment change determination unit determines that the time series data of the sensor output does not correspond to the environment and the action, the recognition unit recognizes the environment.

  In the invention of claim 5, the second measuring means measures the time series data of the output of the sensor when the action is being executed by the second executing means. The environmental change determination means determines whether the time-series data of the sensor output measured by the second measurement means corresponds to the environment recognized by the recognition means and the action executed by the second execution means. Is determined based on the map. Since the sensor output when the action suitable for the recognized environment is executed should be the sensor output associated with the map, the environment change determination unit can change the environment based on the sensor output. Judging. When the environment change determination means determines that the time series data of the sensor output does not correspond to the environment and the action, the environment is recognized by the recognition means because the environment has changed. Therefore, in the invention of claim 5, since the change of the environment can be grasped even after the environment is recognized once, the environment after the change can be recognized.

  The invention of claim 6 creates a map for a humanoid robot having a sensor and recognizing the environment by associating the environment where the humanoid robot is placed, the behavior of the humanoid robot, and the time-series data of the sensor output. (A) preparing a humanoid robot in which a plurality of behaviors are adjusted so as to execute a behavior suitable for each environment in a plurality of environments to be recognized by the humanoid robot, and (b) By using a humanoid robot to perform multiple actions in multiple environments that should be recognized, the time series data of the sensor output when each action is being executed is measured. A mapping that associates actions performed in the environment with time-series data of sensor outputs measured when the actions are executed. It is the creation method.

  According to the invention of claim 6, since a map in which the environment, the action, and the sensor output are associated with each other can be created, the humanoid robot equipped with the sensor can recognize the environment using this map. .

  According to the present invention, since the map in which the time series data of the sensor output, the robot action and the environment are associated with each other is provided, the decision tree for which the environment created based on this map is a discrimination target, The environment can be determined based on the sensor output. Therefore, the environment can be recognized even if there are few sensors. Furthermore, since an action suitable for the recognized environment can be executed, the environment can be smoothly operated.

  The above object, other objects, features and advantages of the present invention will become more apparent from the following detailed description of embodiments with reference to the drawings.

  FIG. 1 shows the appearance of a humanoid robot (hereinafter also simply referred to as “robot”) 10 according to one embodiment of the present invention, and FIG. 2 is an illustrative view showing the robot 10 as a model. Referring to FIG. 2, the robot 10 of this embodiment includes two recumbent arms 12 and 14 that are spaced apart in the vertical direction or the longitudinal direction, and between the upper and lower recumbent surfaces 12 and 14, 16 is formed.

  One yaw shaft joint 18 that forms the degree of freedom of the chest is provided on the upper portion of the body 16, and one pitch shaft joint 20 that forms the degree of freedom of the waist is provided below the body 16. . As is well known, the yaw axis is a vertical rotation axis (Z axis), the pitch axis is one of two axes perpendicular to the yaw axis, for example, the X axis, and will be described later. The roll axis is the other of the two axes perpendicular to the yaw axis, for example, the Y axis. The X axis (pitch axis) is a rotation axis extending in a direction parallel to the paper surface of FIG. 2, and the Y axis (roll axis) is a rotation axis extending in a direction perpendicular to the paper surface of FIG.

  Arms 22 </ b> R and 22 </ b> L are attached to both ends of the upper horizontal arm 12, respectively. These left and right arms 22R and 22L have the same configuration. When it is necessary to distinguish between right and left, “R” indicating right or “L” representing left is added, and when it is not necessary to distinguish, “R” or “L” is not added. In this way, duplicate descriptions are omitted as much as possible. The same applies to the feet described later.

  The arm 22 includes a roll shaft joint 24 that forms a degree of freedom of the shoulder, and is attached to the upper lateral reed 12 by the shoulder joint 24 so as to be rotatable around the roll shaft. In front of the shoulder joint 24, one pitch shaft joint 26, one yaw shaft joint 28, and one roll shaft joint 30 are connected in series. Therefore, in the robot 10 of this embodiment, the arm 22 has 4 degrees of freedom and 8 degrees of freedom on the left and right.

  In addition, legs 32R and 32L are attached to both ends of the lower lying heel 14, respectively. The foot 32 is attached to the lower recumbent joint 14 by one roll shaft joint 34 that forms a degree of freedom of the hip joint. In the lower end direction from the hip joint 32, three pitch shaft joints 36, 38 and 40, which are continuous, And a series connection of one roll shaft joint 42 and one yaw shaft joint 44 is attached. Therefore, in the case of the robot 10 of this embodiment, the foot 32 has 6 degrees of freedom and 12 degrees of freedom on the left and right.

  The electrical configuration of the robot 10 of the embodiment of FIG. 1 is shown in the block diagram of FIG. As shown in FIG. 2, the robot 10 includes a microcomputer or CPU 46 for overall control. The CPU 46 includes a memory 50, a motor control board 52, and a sensor input / output board 54 through a bus 48. Connected.

  Although not shown, the memory 50 includes a ROM and a RAM. In the ROM, the control program and data of the robot 10 are written in advance, and the RAM can be used as a temporary storage memory and a working memory. The memory 50 stores in advance an environment-optimal behavior table as shown in FIG. 4 and a sensor output time series-behavior-environment association map as shown in FIG. It will be described later.

  The sensor input / output board 54 is connected with one biaxial acceleration sensor 56 provided on the body 16 of FIG. 2 and a joint angle sensor 58. The joint angle sensor 58 is shown as one block, but actually includes 22 joint angle sensors that are individually provided for the 22 joints described above and detect the respective joint angles. The biaxial acceleration sensor 56 can detect acceleration with two axes in the front-rear direction and the left-right direction of the robot 10. Using the output of the biaxial acceleration sensor 56, the CPU 46 controls the bipedal walking, reaction rise, and the like of the robot 10. Further, the CPU 46 measures the output time series of the biaxial acceleration sensor 56 and recognizes the environment using the association map stored in the memory 50.

  The motor control board 52 is composed of, for example, a DSP (Digital Signal Processor), and controls each joint shaft motor 60 for each arm, each leg, chest, and waist. In FIG. 3, the joint shaft motor 60 is shown as one block for the sake of simplification, but in reality, one motor (servo motor) is provided for each of the 22 joints described above. It includes 22 motors that control the joint angle.

  FIG. 4 shows an example of a table stored in advance in the memory 50 in which the environment and the optimum behavior are associated with each other. In the robot 10 of this embodiment, the robot 10 is placed, that is, a large number of environments E1, E2,... Ei (i is a predetermined integer) that the robot 10 should recognize are known in advance. The environment is classified by, for example, the reaction force and shape of the floor (or the ground). Specifically, it is classified by the difference in factors such as floor hardness (hard <> soft) and floor slip (friction: small <> large). In this embodiment, i environments are assumed. In the example of FIG. 4, for example, the environment E1 is hard and has low friction, the environment E2 is hard and has moderate friction, and the environment Ei is hard. Is soft and has high friction.

  Further, in this robot 10, for example, all of a plurality of basic actions (j: j is a predetermined integer) such as “walking”, “falling down”, “rebound rising”, etc. are adjusted in advance so as to be optimal in each environment. Action control data for each environment is prepared. In this environment-optimal action table, an action (identifier) that is optimal for the environment is stored in association with each environment. Therefore, using this table, it is possible to grasp which environment is the most suitable action for each action. In the example of FIG. 4, for example, the optimal action for the environment E1 is the action A11 for the “walking” action, the action A12 for the “falling down” action, and the action A1j for the “wake-up reaction” action. It turns out that it is.

  FIG. 5 shows an example of a sensor output time series-behavior-environment association map stored in advance in the memory 50. The association map associates the time-series data of the output of the biaxial acceleration sensor 56, the action performed by the robot 10, and the environment where the robot 10 is placed.

  Here, a method for creating this association map will be described. The present inventors consider that the output of a sensor that detects information of the outside world is closely related to the behavior of the robot and the characteristics of the environment, and allows the robot to recognize the environment and to select a behavior suitable for the environment. Therefore, a map in which “robot behavior”, “time series of sensor outputs”, and “environment” are associated with each other is created.

  When creating a correspondence map, first, the user is optimal for each of a large number (i) of environments E1-Ei to be recognized for all (j) basic actions of the robot 10. Each control data is created as optimum behavior A11-Aij in each environment, and stored in the memory 50. For example, the “walking” behavior is tuned so as to be suitable for operation in each environment, such as changing the way the foot is lowered in a slippery environment and a slippery environment. Thus, in this robot 10, all the actions are adjusted so as to execute actions suitable for each environment in a large number of environments to be recognized.

  Although not shown in FIG. 3, the robot 10 is provided with an external interface (for example, RS232C) so that data can be communicated with an external computer (not shown). The control data of the optimum behavior created on the external computer can be written in the memory 50 of the robot 10.

  Then, by using this tuned robot 10, by executing all the actions A11-Aij in a large number of environments E1-Ei to be recognized, the two-axis acceleration sensor 56 when executing each action Output time-series data is measured and stored in the memory 50. Specifically, in this embodiment, since (j × i) actions are executed in i environments, (j × i × i) sensor output time-series data can be obtained. The sensor output data stored in the memory 50 is transferred to an external computer via the above external interface, for example.

  For example, on an external computer, the environment in which the robot 10 is placed, the action performed by the robot 10 in the environment, and the time-series data of the sensor output when the action is executed are associated with each other. A map as shown in FIG. The created map data is transferred from the external computer to the robot 10 and stored in the memory 50.

  In the above description, the map is created on the external computer. However, the environment, the action, and the sensor output may be associated on the robot 10 as they are.

  In the example of FIG. 5, for example, it can be seen that the sensor output (time-series data) D1, the action A11, and the environments E1, E2, and E5 are associated with each other. This means that when the action A11 is executed in the environments E1, E2, and E5, the same sensor output D1 is obtained. Since the action A11 is tuned to be optimal for the environment E1, the sensor output D1 obtained when the action A11 is executed in the environment E1 can be expressed as an optimum sensor output.

  The robot 10 of this embodiment recognizes the environment by determining in what environment the robot 10 is placed at the beginning or when a change in the environment is detected. When recognizing the environment, a decision tree for recognizing the environment is created based on the sensor output time series-behavior-environment association map as described above. That is, the memory 50 of the robot 10 is provided with a decision tree creation program based on a decision tree automatic generation algorithm such as ID3 or C4.5, for example, and the CPU 46 executes the decision tree creation program to associate Based on the map, a decision tree is created with the environment as a discrimination target. In this embodiment, since a decision tree can be automatically created inside the robot 10, an action to be executed first for environment recognition can be selected at random.

  The decision tree is a technique for determining the input pattern class. In the decision tree, the order of application of each decision procedure is completely determined in advance, and the decision step (classifier) determines which classifier to go to in the next stage (Reference: Junichiro Toriwaki, Television Society) Textbook series 9 “Cognitive engineering: pattern recognition and its application”, Corona, 1993).

  In the decision tree of this embodiment, the application order of the decision procedure for determining the environment in which the robot exists is determined by the action performed by the robot 10 and the sensor output at that time. Therefore, until the environment is determined in the decision tree, the environment can be recognized by repeating the execution of the action and the determination based on the sensor output at that time according to the decision tree.

  Specifically, the robot 10 acts in the unrecognized environment based on the decision tree, and measures a time series of sensor outputs during the behavior. Then, based on the measurement result of the sensor output, an attempt is made to determine the environment. If the environment cannot be recognized, the next action is selected based on the decision tree, and the action, measurement, and determination are repeated again. In this way, the robot 10 can recognize which environment it is in while performing an action based on the decision tree.

  FIG. 6 shows an outline of a part of an example of a decision tree to be created. In the example of FIG. 6, action A11 is set in the first stage. In this embodiment, the behavior of the first stage of the decision tree is determined at random. An example of the flow of environment determination by the decision tree of FIG. 6 will be described. First, when the sensor output D1 is measured as a result of the execution of the action A11, the action A55 is selected and executed in the next stage. The If the sensor output D5 is not measured as a result of the execution of the action A55, the environment is determined, that is, it is recognized that the environment where the robot 10 is placed is the environment E2.

  When the robot 10 recognizes the environment, the robot 10 can execute an action suitable for the recognized environment, for example, based on the environment-optimal action table. Therefore, the robot 10 can operate as desired in accordance with the recognized environment. Specifically, since it is possible to walk, for example, two legs according to the hardness of the floor and the degree of sliding, it is possible to act smoothly without slipping and falling or breaking the balance.

  Further, the robot 10 measures the time series data of the output of the biaxial acceleration sensor 56 when executing an action suitable for the environment, and discriminates a change in the environment. That is, the time series data of the sensor output when the action suitable for the recognized environment is executed should be the optimum sensor output associated with the map. Therefore, the change in the environment is grasped by determining whether the time series data of the sensor output corresponds to the recognized environment and the executed action based on the association map. If the time series data of the sensor output does not correspond to the environment and action, it can be seen that the environment has changed, so the process for recognizing the environment as described above is performed again. Execute. Therefore, in this embodiment, since the change of the environment can be grasped even after the environment is recognized once, the environment after the change can be recognized, and further, the action suitable for the environment after the change can be executed.

  Specifically, the CPU 46 of the robot 10 executes processing according to a flowchart as shown in FIG. In the first step S <b> 1 in FIG. 7, one action is randomly selected from actions registered in advance in the memory 50. Next, in step S3, the decision tree creation program is executed to generate a decision tree having the selected action as the first stage based on the sensor output time series-behavior-environment association map.

  Subsequently, in step S5, the control data of the selected action is read from the memory 50, and the execution of the action is started. In step S <b> 7, the output from the biaxial acceleration sensor 56 during action is measured at a constant cycle via the sensor input / output board 54, and the sensor output time series data during action is stored in the memory 50.

  In step S9, determination based on the decision tree is performed. That is, based on the measured sensor output time-series data, it is determined which decision procedure is to be advanced at the next stage of the decision tree. For example, in the case where the first action A11 in FIG. 6 is executed, it is determined whether the measured sensor output data is the sensor output D1 or another output, and which action is followed next. To decide.

  Subsequently, in step S11, it is determined whether or not the environment has been recognized. That is, the determination in step S9 determines whether the next stage advanced based on the sensor output in the decision tree is to determine the environment.

  If “NO” in the step S11, that is, if the environment has not yet been recognized, in the subsequent step S13, the next action is selected based on the decision tree. That is, the action set as the decision procedure for the next stage advanced in the decision tree is selected. Then, returning to step S5, the execution of the selected action (step S5) and the sensor output measurement during the action (step S7) are processed again, and the determination by the decision tree (step S9) is performed based on the sensor output result. Then, it is determined whether the environment has been recognized (step S11). In this way, the processing from step S5 to step S13 is repeated to recognize the environment based on the decision tree.

  If “YES” in the step S11, that is, if the environment is determined in the decision tree, in the subsequent step S15, the recognized environment is recorded in the memory 50. In step S17, an action suitable for the environment is selected based on, for example, the environment-optimal action table in accordance with an internal program or an external program of the robot 10 or an action command given by the user. For example, when the recognized environment is the environment E1, the action to be executed is selected from the action A11-action A1j that is optimal for the environment E1.

  Subsequently, in step S19, execution of the selected action is started, and in step S21, output time-series data of the biaxial acceleration sensor 56 during action is measured.

  In step S23, it is determined based on the association map whether the measured sensor output data is optimal for the recognized environment and the executed action. That is, in this step S23, it is determined whether or not the environment has changed. Specifically, in step S19, an action adjusted to be optimal for the environment is executed in the recognized environment. Therefore, in step S21, an optimum associated with the environment and the action in the map is executed. The sensor output should be measured. Therefore, if the sensor output measured when the optimal action is executed in the recognized environment is not the sensor output set in the association map, the recognized environment may have changed. I understand.

  Therefore, if “NO” in the step S23, that is, if the environment has changed, the process returns to the step S1 again to repeat the process in order to try to recognize the environment. As a result, the changed environment can be recognized, and the optimum action for the changed environment can be executed thereafter.

  On the other hand, if “YES” in the step S23, that is, if the environment has not changed, in the step S25, the environment is determined in accordance with the internal program or the external program of the robot 10 or the next action command issued from the user. Process the selection of the next action to be performed. In step S27, it is determined whether or not to end the action. If “NO”, that is, if there is an action to be executed, the process returns to step S19 to execute the next action. On the other hand, if “YES” in the step 27, the process is ended.

  According to this embodiment, by looking at the sensor output during action in a long time series, even a very small amount of sensors of one type of the biaxial acceleration sensor 56 could not be known in normal usage. Can be extracted, that is, the environment can be recognized. Therefore, the robot 10 can execute an action suitable for the recognized environment. Further, since the sensor can be recognized even with a very small amount, for example, wiring saving and power saving can be realized. Further, by combining with other sensors, more environmental information can be extracted, and more complex action selection can be performed.

  In the above-described embodiment, when an environmental change is detected in step S23 of FIG. 7 and the environment recognition process is performed again, the process returns to step S1 to randomly select an action. In the example, the environment recognition process may be performed while continuing the action by selecting the action that was executed when the change in the environment was recognized.

  In each of the above-described embodiments, a correspondence map is stored in the memory 50, a first action for environment recognition is arbitrarily selected, and a decision tree is created each time an environment recognition process is executed. However, in another embodiment, a decision tree created in advance from an association map by an external computer or the like may be stored in the memory 50. In this case, since the decision tree is fixed, the environment recognition method is fixed. That is, the robot 10 always performs the same action first when recognizing the environment.

  In each of the above-described embodiments, the robot 10 is provided with the biaxial acceleration sensor 56. Then, in an environment that is classified according to differences in factors such as the hardness of the floor (or the ground) and the slipping of the floor, actions that touch the floor such as walking, lying down, and recoiling are executed, and the sensor output time series during the action is calculated. By making a measurement, a correspondence map was created. However, the type of sensor provided in the robot 10 may be changed as appropriate. However, the environmental features and behaviors to be recognized or discriminated need to be reflected in the sensor output.

  In addition, the degrees of freedom and specific configurations of the above-described embodiments are merely examples, and do not necessarily have 22 degrees of freedom, and the joint shaft motor can be arbitrarily changed. If the robot has a high degree of freedom, such as the biped walking type of the above-described embodiment, the environment that can be discriminated increases such as the difference in floor hardness and sliding (friction). In the case of a robot, it is only possible to grasp whether it is smooth or uneven, and the environment that can be discriminated is reduced.

1 is an external view showing a robot according to an embodiment of the present invention. It is an illustration figure which shows the robot of FIG. 1 Example. It is a block diagram which shows the electrical structure of the robot of FIG. 1 Example. It is an illustration figure which shows an example of the table regarding the optimal action for every environment memorize | stored in the memory of FIG. It is an illustration figure which shows an outline of an example of the map which matched the sensor output time series memorize | stored in the memory of FIG. 3, an action, and an environment. It is an illustration figure which shows the outline of an example of the decision tree produced | generated from a map. It is a flowchart which shows an example of operation | movement of the robot of FIG. 1 Example.

Explanation of symbols

10 ... Robot (Humanoid)
22R, 22L ... arms 32R, 32L ... feet 46 ... CPU
DESCRIPTION OF SYMBOLS 50 ... Memory 52 ... Motor control board 54 ... Sensor input / output board 56 ... Biaxial acceleration sensor 60 ... Joint axis motor

Claims (6)

A humanoid robot that recognizes the environment,
Sensor,
Map storage means for storing a map in which the environment in which the humanoid robot is placed, the action of the humanoid robot, and the time series data of the output of the sensor are associated, and the first execution for executing the action for environment recognition Means and first measurement means for measuring time series data of the sensor output when the action is being executed by the first execution means, and the output of the sensor measured by the first measurement means. A humanoid robot comprising recognition means for recognizing an environment based on time-series data and a decision tree for which the environment created based on the map is a discrimination target.   The humanoid robot according to claim 1, wherein the recognizing unit further includes a decision tree creating unit that creates a decision tree whose environment is a discrimination target based on the map. The sensor includes an acceleration sensor;
The action includes at least one of walking, falling asleep, and getting up;
The humanoid robot according to claim 1 or 2, wherein the environment to be recognized is classified by at least floor hardness or floor slip.   The humanoid robot according to claim 1, further comprising second execution means for executing an action suitable for an environment recognized by the recognition means. Second measurement means for measuring time-series data of the output of the sensor when the action is being executed by the second execution means, and time-series data of the output of the sensor measured by the second measurement means, An environment change determination unit that determines whether the environment corresponds to the environment recognized by the recognition unit and the action executed by the second execution unit based on the map;
The humanoid robot according to claim 4, wherein when the environment change determination unit determines that the time series data of the output of the sensor does not correspond to the environment and the action, the recognition unit recognizes the environment. A map creation method for creating a map that associates an environment in which the humanoid robot is placed, an action of the humanoid robot, and time-series data of the output of the sensor for a humanoid robot that includes a sensor and recognizes the environment Because
(a) preparing the humanoid robot in which a plurality of behaviors are adjusted so as to execute a behavior suitable for each environment in a plurality of environments to be recognized by the humanoid robot;
(b) Using the humanoid robot, by measuring each of the plurality of actions in the plurality of environments to be recognized, to measure time series data of the output of the sensor when executing each action,
(c) A map creation method that associates an environment, an action executed in the environment, and time-series data of the sensor output measured when the action is executed.

Images