JP4649913B2 - Robot apparatus and movement control method of robot apparatus - Google Patents

Description

The present invention relates to a legged mobile robot apparatus that operates autonomously and a movement control method for the robot apparatus, and a robot apparatus and a robot apparatus that can move even in a region where the state of a walking surface such as a floor surface changes greatly. The present invention relates to a movement control method.

  A mechanical device that performs an action similar to that of a human (living body) using an electrical or magnetic action is called a “robot”. Robots started to spread in Japan from the end of the 1960s, but many of them are industrial robots such as manipulators and transfer robots for the purpose of automating and unmanned production work in factories. Met.

  Recently, practical robots that support life as a human partner, that is, support human activities in various situations in daily life such as the living environment, have been developed. Unlike industrial robots, such practical robots have the ability to learn how to adapt themselves to humans with different personalities or to various environments in various aspects of the human living environment. For example, a “pet-type” robot that mimics the body mechanism and movement of a four-legged animal such as a dog or cat, or a body mechanism and movement of a human that walks upright on two legs. Robot devices such as “humanoid” or “humanoid robots” are already in practical use.

  Since these robot devices can perform various operations with an emphasis on entertainment, for example, compared to industrial robots, they may be referred to as entertainment robots. In addition, such a robot apparatus is equipped with imaging means such as a camera, various sensors, etc., to acquire information from the outside, and depending on these external information (external stimuli) and the internal state of itself Can act autonomously.

  By the way, among such robotic devices, legged mobile robotic devices, particularly bipedal walking type robotic devices, for example, are inclined or have obstacles while maintaining the balance of the entire body in an actual environment. Walking on such a floor is not easy. Usually, walking is possible by calculating a control signal to be output to an actuator provided at each joint based on a floor reaction force obtained from a sensor or the like provided on the sole or each joint angle.

  For example, the following Patent Document 1 discloses a technology of a floor shape estimation device for a legged mobile robot device that walks while simultaneously estimating a ground contact surface inclination of a foot (foot) and a height difference between both foot contact surfaces. Has been.

JP 2001-328083 A

  However, in the technique described in Patent Document 1, although smooth walking is possible by estimating the local shape of the floor surface, there is absolutely no information about the floor surface until the foot actually landed. From this point of view, it is only a very short amendment to the walking plan.

  In other words, it is possible to adaptively calculate the correction amount of the trajectory (adaptive operation amount) while landing on the floor, but it is not possible to acquire knowledge of each step about the floor before landing one step. During walking control, it is not possible to automatically determine the optimal control form for the floor, assuming a wider range (of the floor) instead of one step unit, and the adaptive operation amount based on the predetermined control form Need to be calculated. Here, the adaptive operation amount in the above-mentioned patent document 1 or the like is a correction amount corresponding to the inclination of the floor surface, its own posture, or the like with respect to a certain walking control model. Accordingly, there is a problem that even if the change of the continuous control, that is, the control content of the walking control model can be changed slowly, it is difficult to walk if such adaptive operation alone cannot be used.

  For example, when walking on a hard ground surface (floor surface) such as a plate surface, marble, or concrete, and when walking on a soft ground surface such as a carpet or lawn, it is a standard walking control mode, for example, walking. The parameters and coefficients for calculating the adaptive operation amount according to the current floor surface from the control model, or the walking algorithm (walking model) are completely different, and simply calculating the adaptive operation amount without changing the walking control form It is not possible to move to a region where the floor surface state changes greatly, such as from a hard ground surface to a soft ground surface. Therefore, during such movement, the robot apparatus itself cannot determine whether or not walking is possible, and a person manually resets the knowledge about the floor surface. Specifically, for example, according to the floor surface It was necessary to reset parameters for calculating the adaptive operation amount.

The present invention has been proposed in view of such a conventional situation, and by determining the state of the floor on which to walk, the walking control mode is not corrected according to the walking plan for each step, but according to the floor. It is an object of the present invention to provide a robot apparatus and a movement control method for the robot apparatus that enable stable walking even when the floor surface state changes greatly.

To achieve the above object, a robot apparatus of the present invention, in autonomously operable legged mobile robot apparatus, based on information about the current moving surface, the movement of the current from the previously prepared plurality of categories were a moving surface determining means for determining a moving surface category is a category of the surface, to select the predetermined movement control mode in response to the movement plane category is determined by the movement plane determining means, sensor provided on at least the leg portion from the sensor value output from possess a movement controlling means for controlling the movement by calculating the movement control amount based on the movement control mode being selected, the movement control means, movement control different models for each of the categories Or having a movement algorithm, selecting a movement control model or a movement algorithm corresponding to the input moving plane category, and using the sensor value A correction amount or adaptive operation amount in accordance with the moving algorithm from serial movement control model calculated as the movement control amount, the moving surface determining means uses the movement control amount as information relating to the current movement plane.

  In the present invention, unlike the conventional method of calculating the correction amount of the walking model as appropriate during walking, the floor surface state is determined to a predetermined floor surface category prepared in advance, and according to the determination result. By selecting a predetermined walking control form, walking can be controlled according to the walking control form optimal for the floor surface.

  The walking control means calculates a walking control amount based on a selected walking control form from sensor values output from at least sensors provided on the legs, and the walking surface discrimination means includes the walking control The amount can be used as information regarding the current walking surface. For example, when the legs have joints, the angles of the joints are detected, the reaction force from the foot walking surface is detected, The walking control amount converted from the sensor values of the various sensors includes information on the floor surface at that time, and can be used to determine the floor surface category.

  Further, the walking control means has a different walking control model for each category, selects a walking control model corresponding to the input walking plane category, and uses the sensor value to determine from the walking control model. The correction amount is calculated as the walking control amount, or the walking control means converts the sensor value into the walking control amount using a predetermined parameter, and has different parameters for each category. The amount of walking control can be calculated using parameters corresponding to the input walking surface category, and thus, the optimal walking control model or the optimal parameter can be calculated according to the input floor surface category. Since the walking control amount can be calculated by selecting, stable walking can be performed.

  Furthermore, the walking surface discriminating means uses at least a sensor value output from a sensor provided on a leg as information relating to the current floor surface, or has an imaging unit. The walking surface area can be extracted from the input image captured by the imaging means, and the walking surface area can be used as information regarding the current walking surface. The sensor value and the floor surface area are information regarding the current floor surface. The floor category classification can be performed using these.

  Further, the walking surface discriminating means can discriminate a walking surface category by pattern identification of information relating to the current walking surface. For example, the floor surface discriminating means is provided for each category classified according to a predetermined standard. , Having a learned feature amount that has been learned in advance with respect to the feature amount of the information relating to the floor surface, and comparing the feature amount of the information relating to the current floor surface with the learned feature amount, to determine the floor category It is possible to learn after classifying various floors into floor categories considered necessary.

  Further, the plurality of categories are classified according to learned feature values obtained by previously learning feature amounts of information relating to the walking surface, and the walking surface determination means includes the feature amount of information relating to the current walking surface. And the learned feature value can be compared to determine the walking surface category, and the floor surface category may be classified according to the feature value obtained from the floor surface.

  Furthermore, it has an action generation means for generating a stepping action for discriminating the walking surface category, and the walking control means is configured to obtain at least a sensor value output from a sensor provided on a leg during the stepping action. The walking control amount may be calculated based on the selected walking control mode, and the walking surface determination unit may use the walking control amount as information on the walking surface, and the floor category may be determined before walking by the stepping action. Can be determined.

  Further, the walking surface discriminating means can continuously discriminate the walking surface category during walking, and the floor surface state changes greatly by walking by constantly determining the floor surface category during walking. Even so, stable walking is possible.

Movement control method for a robot apparatus according to the present invention is a movement control method for autonomously operable legged mobile robot apparatus, based on information about the current moving surface, from a plurality of categories, which are prepared in advance of the current mobile a moving surface determination step of determining a moving surface category is a category of the surface, to select the predetermined movement control mode in accordance with the discriminated said movement plane category in the movement plane determination step, provided on at least the leg portion from the sensor value output from the sensor, it has a movement control step of controlling the movement by calculating the movement control amount based on the movement control mode being selected, in the movement control step, movement control different for each of the categories Select a movement control model or a movement algorithm corresponding to the input movement plane category, and use the sensor value. A correction amount or adaptive operation amount in accordance with the moving algorithms from the mobile control model calculated as the movement control amount in the above movement plane determination step, using the movement control amount as information relating to the current movement plane.

  According to the present invention, a walking control amount that is a correction amount from a reference walking control model that is calculated according to a current walking surface such as a floor surface, a sensor from sensors provided on a sole, a joint, or the like. The current walking surface category is determined using the value or information about the walking surface such as the walking surface area included in the input image as input, and the walking control mode for walking control is selected according to the determination result and walking In order to control walking, it is possible to control walking after recognizing the state of the walking surface as a walking surface category in advance, especially when walking between areas where the state of the walking surface changes greatly. It is possible to provide a robot apparatus and its walking control method that can perform walking control that is much more stable than conventional methods that control walking while appropriately correcting the movement.

  Hereinafter, specific embodiments to which the present invention is applied will be described in detail with reference to the drawings. This embodiment is applied to a robot apparatus that performs operation control such that a part of the apparatus is brought into contact with an object.

  In the present embodiment, a bipedal walking robot device will be described as an example, but it goes without saying that the present invention can be applied to a robotic device that can be moved by four feet or wheels regardless of the bipedal walking robot device. . Here, a bipedal walking robot apparatus will be described first, and then a walking control method of such a robot apparatus will be described.

(1) Robotic device This humanoid robotic device is a practical robot that supports human activities in various situations in the living environment and other daily lives, depending on the internal state (anger, sadness, joy, fun, etc.) It is an entertainment robot that can express basic actions performed by humans. FIG. 1 is a perspective view showing an overview of the robot apparatus according to the present embodiment.

  As shown in FIG. 1, the robot apparatus 1 includes a head unit 3 connected to a predetermined position of the trunk unit 2, and two left and right arm units 4R / L and two right and left leg units 5R /. L is connected to each other (provided that R and L are suffixes indicating right and left, respectively, and the same applies hereinafter).

  The joint degree-of-freedom configuration of the robot apparatus 1 is schematically shown in FIG. The neck joint that supports the head unit 3 has three degrees of freedom: a neck joint yaw axis 101, a neck joint pitch axis 102, and a neck joint roll axis 103.

  Each arm unit 4R / L constituting the upper limb includes a shoulder joint pitch axis 107, a shoulder joint roll axis 108, an upper arm yaw axis 109, an elbow joint pitch axis 110, a forearm yaw axis 111, and a wrist. A joint pitch axis 112, a wrist joint roll wheel 113, and a hand portion 114 are included. The hand portion 114 is actually a multi-joint / multi-degree-of-freedom structure including a plurality of fingers. However, since the operation of the hand unit 114 has little contribution or influence on the posture control or walking control of the robot apparatus 1, it is assumed in the present specification that the degree of freedom is zero. Therefore, it is assumed that each arm portion has seven degrees of freedom.

  The trunk unit 2 has three degrees of freedom: a trunk pitch axis 104, a trunk roll axis 105, and a trunk yaw axis 106.

  Each leg unit 5R / L constituting the lower limb includes a hip joint yaw axis 115, a hip joint pitch axis 116, a hip joint roll axis 117, a knee joint pitch axis 118, an ankle joint pitch axis 119, and an ankle joint. A roll shaft 120 and a foot 121 are included. In the present specification, the intersection of the hip joint pitch axis 116 and the hip joint roll axis 117 defines the hip joint position of the robot apparatus 1. The human foot 121 is actually a structure including a multi-joint / multi-degree-of-freedom sole, but in the present specification, for the sake of simplicity, the foot of the robot apparatus 1 has zero degrees of freedom. . Accordingly, each leg is configured with 6 degrees of freedom.

  In summary, the robot apparatus 1 as a whole has a total of 3 + 7 × 2 + 3 + 6 × 2 = 32 degrees of freedom. However, the robot device 1 for entertainment is not necessarily limited to 32 degrees of freedom. Needless to say, the degree of freedom, that is, the number of joints, can be increased or decreased as appropriate in accordance with design / production constraints or required specifications.

  Each degree of freedom of the robot apparatus 1 as described above is actually implemented using an actuator. It is preferable that the actuator be small and light in light of demands such as eliminating the extra bulge on the appearance and approximating the shape of a natural human body, and performing posture control on an unstable structure such as biped walking. .

  Such a robot apparatus includes a control system for controlling the operation of the entire robot apparatus, for example, in the trunk unit 2. FIG. 3 is a schematic diagram showing a control system configuration of the robot apparatus 1. As shown in FIG. 3, the control system is a motion that controls the whole body cooperative motion of the robot device 1 such as driving of an actuator 350 and a thought control module 200 that dynamically controls emotion judgment and emotional expression in response to a user input or the like. And a control module 300.

  The thought control module 200 includes a CPU (Central Processing Unit) 211, a RAM (Random Access Memory) 212, a ROM (Read Only Memory) 213, and an external storage device (hard disk / disk This is an independent drive type information processing apparatus that is configured with 214 and the like and can perform self-contained processing within the module.

  This thought control module 200 determines the current emotion and intention of the robot apparatus 1 according to stimuli from the outside such as image data input from the image input apparatus 251 and audio data input from the audio input apparatus 252. Here, the image input device 251 includes a plurality of CCD (Charge Coupled Device) cameras, for example, and the sound input device 252 includes a plurality of microphones, for example.

  In addition, the thought control module 200 issues a command to the motion control module 300 to execute an action or action sequence based on decision making, that is, exercise of the limbs.

  One motion control module 300 includes a CPU 311 for controlling the whole body cooperative motion of the robot apparatus 1, a RAM 312, a ROM 313, an external storage device (hard disk drive, etc.) 314, etc., and performs self-contained processing within the module. It is an independent drive type information processing apparatus that can be performed. Also, the external storage device 314 can store, for example, walking patterns calculated offline, target ZMP trajectories, and other action plans.

  The motion control module 300 includes an actuator 350 that realizes degrees of freedom of joints distributed throughout the body of the robot apparatus 1 shown in FIG. 2, a distance measurement sensor (not shown) that measures the distance to the object, a body Posture sensor 351 that measures the posture and inclination of the trunk unit 2, grounding confirmation sensors 352 and 353 that detect the floor or landing of the left and right soles, a load sensor provided on the sole 121 of the foot 121, a power source such as a battery Various devices such as a power supply control device 354 for managing the network are connected via a bus interface (I / F) 301. Here, the posture sensor 351 is configured by, for example, a combination of an acceleration sensor and a gyro sensor, and the grounding confirmation sensors 352 and 353 are configured by proximity sensors, micro switches, or the like.

  The thought control module 200 and the motion control module 300 are constructed on a common platform, and are interconnected via bus interfaces 201 and 301.

  The motion control module 300 controls the whole body cooperative motion by each actuator 350 in order to embody the action instructed from the thought control module 200. That is, the CPU 311 extracts an operation pattern corresponding to the action instructed from the thought control module 200 from the external storage device 314 or generates an operation pattern internally. Then, the CPU 311 sets a foot movement, a ZMP trajectory, a trunk movement, an upper limb movement, a horizontal waist position, a height, and the like in accordance with the designated movement pattern, and commands for instructing the movement according to these setting contents. The value is transferred to each actuator 350.

  Further, the CPU 311 detects the posture and inclination of the trunk unit 2 of the robot apparatus 1 from the output signal of the posture sensor 351, and each leg unit 5R / L is caused to move freely by the output signals of the grounding confirmation sensors 352 and 353. Alternatively, the whole body cooperative movement of the robot apparatus 1 can be adaptively controlled by detecting whether the robot is standing or standing. Further, the CPU 311 controls the posture and operation of the robot apparatus 1 so that the ZMP position always moves toward the center of the ZMP stable region.

  In addition, the motion control module 300 returns to the thought control module 200 the degree to which the intended behavior determined by the thought control module 200 has been expressed, that is, the processing status. In this way, the robot apparatus 1 can determine its own and surrounding conditions based on the control program and act autonomously.

(2) Walking Control Method Next, a walking control method for walking a legged mobile robot device having a walking function like the robot device 1 described above will be described. This robot apparatus discriminates the state of the walking surface by performing statistical pattern recognition of information related to the walking surface, and selects the optimal walking control mode prepared in advance corresponding to the determination result to perform walking control. Is. Specifically, when the robot apparatus is walking or when it is stepping on the spot, which will be described later, various sensor information, walking control amounts, image information, and the like are acquired as information on the walking surface, and thereby the current walking surface category (hereinafter referred to as the walking surface category). Is called a walking surface category), and a stable walking is realized by selecting a walking control mode according to the walking surface category.

  Here, in this embodiment, the walking surface that is a plane on which the robot apparatus walks is the floor surface, and the category in which the floor surface is categorized is the floor surface category, but the walking surface is not only the floor surface, Similarly, with respect to surfaces such as outdoor ground, lawn, and stairs that can be walked by the robot device, the walking surfaces should be classified into categories in advance, and the walking control mode selected after selection. Needless to say.

  As described above, the conventional robot apparatus is, for example, walking control that is a trajectory correction amount from a predetermined walking control model (motion reproduction) according to various sensor values such as posture during walking and reaction force from the floor surface. By calculating the amount (hereinafter referred to as the adaptive operation amount) and correcting the walking control model, the posture balance of the robot apparatus is maintained and stabilized according to the inclination of the floor surface where the foot part contacts the ground during walking. It performs adaptive control so that it can walk. However, in the walking control by the adaptive operation amount calculated from the sensor value or the like during walking, for example, from a floor surface (flooring) with a plate material or the like, or a hard floor surface such as marble or concrete, If you move in a region where the floor condition changes greatly, such as moving to a soft floor such as a long carpet or lawn, you will not be able to walk. This is because the walking control mode to be used when walking on a hard floor and the walking control mode to be used when walking on a soft floor are completely different.

  Therefore, when walking control is performed by calculating the amount of adaptive operation from various sensor values during walking, it is possible to respond to continuous gradual changes, as in the case of moving from a hard floor surface to a soft floor surface. When the floor surface state changes drastically, it becomes difficult to walk unless the walking control mode is changed.

  Therefore, as a result of intensive studies by the inventors of the present application, based on various sensor values and adaptive operation amounts obtained according to the floor surface, the current floor surface is classified into a floor surface category prepared in advance and according to this floor surface category. Walking by calculating the adaptive operation amount after preparing the walking control form and selecting the optimal walking control form corresponding to the floor category by discriminating the floor category of the current floor It has been found that even if the floor condition changes dramatically when controlled, the legged mobile robot device can be walked well.

  Here, different walking control modes mean that the parameters or coefficients used when the adaptive operation amount is converted or calculated from various sensor values are different, and the walking motion used when walking (walking control model). Or when the walking algorithm is different. The adaptive operation amount indicates a correction value for correcting a deviation from a target control value, for example, a deviation of a rotation angle of each actuator. In the present specification, a series of motion control values in a walking motion such as walking fast and slowly walking (walking control model) and walking motion such as the adaptive operation amount are collectively referred to as a gait.

  Hereinafter, the floor-specific walking control device of the robot apparatus capable of adaptively controlling walking by performing floor surface category determination as to which floor surface category the current floor surface is classified in this way will be described below. .

  FIG. 4 is a block diagram functionally showing only the main parts necessary for the floor-based walking control device of the robot device. The floor-specific walking control device includes a behavior control unit 11 that controls the behavior of the robot device, a floor surface determination unit 12 that recognizes the floor surface and outputs a floor category, and a behavior command and a floor from the behavior control unit 11. A walking control unit 13 that controls walking by selecting a walking control mode based on the floor category from the surface determination unit 12 is provided.

  The behavior control unit 11 selects and executes the behavior of the robot apparatus according to the situation. When the behavior control unit 11 issues an operation command (walking command) D1, the walking control unit 13 operates and executes walking control. Note that the situation of the floor surface, for example, whether there is an obstacle or the like, is also one factor (external stimulus) that affects the behavior of the robot apparatus.

  The floor discriminating unit 12 is an amount by which the body of the robot apparatus senses the state of the floor, such as an adaptive operation amount D2 of the walking control unit 13 and output values (sensor values) D3 of various sensors 360, and an image input means such as a camera. By inputting the image data D4 and the like from 251 and solving the pattern identification problem of the statistical floor category, the floor category of the current floor is selected from the floor categories prepared in advance.

  Here, the sensor 360n includes a potentiometer that detects the rotation angle and position of the actuator 350n provided at each joint, an acceleration sensor that detects the acceleration of the foot (plantar) during walking, and a floor surface provided on the sole. A force sensor or the like for detecting the reaction force from the sensor, and the rotation angle, acceleration, reaction force and the like are input as sensor values.

  In addition, since pattern recognition can be performed as soon as input data such as the adaptive operation amount D2, various sensor values D3, or image data D4 enters, the robot apparatus 1 can always continue to determine the floor category while walking. Is possible. The floor category D5 that is the determination result is output to other elements such as the behavior control unit 11 and the walking control unit 13.

  The walking control unit 13 is activated by the operation command D1, and calculates the correction amount from the walking control model in the walking control form as the adaptive operation amount D2 from the various sensor values D3 as described above according to the predetermined walking control form. The data is output to the floor surface determination unit 12. Then, the floor surface determination unit 12 determines the current floor surface category, receives the floor surface category D5, selects the walking control mode corresponding to the current floor surface state, and selects the selected walking control mode. A gait is generated by calculating an adaptive operation amount as a correction amount from the walking control model in FIG. That is, in order to cause the walking control unit 13 to perform a walking control form corresponding to the floor category in advance, for example, a reference walking model and adaptive walking, various correction amounts from the walking model at that time Coefficients or parameters for conversion or calculation from sensor values are prepared.

  Here, it is assumed that the walking control mode corresponding to the floor category is selected. For example, the floor category A and the floor category B are greatly different in the state of the floor, and the walking algorithm is completely different. When it is necessary to change to a thing, it indicates that the walking algorithm is selected, and the floor category A and the floor category B use the same walking control model to calculate the adaptive operation amount from the sensor value. When it is only necessary to change the parameter, this indicates that the parameter is selected. The adaptive operation amount D2 is a correction amount from, for example, a walking model for adaptively stabilizing the posture during walking according to the shape of the floor surface, specifically, a roll axis, pitch, and the like. A correction amount or the like for correcting a deviation from a target trajectory in the axis and the yaw axis is shown.

  When calculating the adaptive operation amount D2 output to the floor surface determination unit 12 and the adaptive operation amount calculated again at the next walking timing according to the walking algorithm selected according to the floor surface category, various sensor values are used. There are parameters to be described later for conversion into the adaptive operation amount. Here, conventionally, a walking control form such as a predetermined walking algorithm or a predetermined parameter is selected by inputting from the outside in advance, and the adaptive operation amount is calculated according to this walking control form. Thus, in this method, only an adaptive operation amount from a predetermined walking control mode can be converted, and depending on the floor, an appropriate adaptive operation amount may not be calculated unless the walking control mode is changed. . Accordingly, the walking control unit 13 first calculates an adaptive operation amount according to various sensor values from a predetermined walking control form used as a standard, and the floor surface determination unit 12 determines the floor surface category. And this floor surface category D5 is received, and according to this floor surface category D5, a walk control form is selected from the said standard walk control form. Specifically, for example, a walking algorithm corresponding to the floor surface, a walking control model (walking type), or a parameter for calculating an adaptive operation amount is selected, and the adaptive operation amount is determined according to the walking control form determined by the floor surface category. By calculating again, more stable walking can be realized. Note that the change in the walking control mode may include other factors such as the state of the aircraft as well as the floor category. Here, the standard walking control mode for calculating the adaptive operation amount D2 used for discriminating the floor category is intended to be used, for example, in the most standard walking algorithm, for example, at home. In this case, a walking control model or the like for walking on a floor surface such as a plate material may be prepared. In addition, when the floor surface is continuously determined, the adaptive operation amount D2 may be calculated according to the walking control mode corresponding to the floor surface category input at the previous timing by the floor surface determination unit 12.

  In this way, the walking control unit 13 generates a gait by selecting an optimal walking control mode according to the behavior command D1 from the behavior control unit 11 and calculating an adaptive operation amount, and drives the actuator 350n of each joint. Command (control value) D6 to be output.

  Examples of the floor surface include hard floor surfaces (plates, concrete), carpets, tatami mats, rubbers, etc. These various floor surfaces include, for example, an elastic coefficient, a friction coefficient, a rolling coefficient, and a carpet. It can classify | categorize into a floor surface category on the basis of the length of a hair foot, or these combination. For each category of the classified floor surface, a feature amount of input data for discrimination one or more of adaptive operation amount D2, sensor value D3, and image data D4 used in category discrimination is learned in advance, and this It is assumed that the learned feature amount is held in the floor discriminating unit 12 as a learning database for each floor category. In addition, as described above, if the robot apparatus is a surface on which the robot apparatus can be walked, such as an outdoor ground or a lawn, it may be categorized and learned in the same manner.

  Alternatively, the robot apparatus does not learn after previously classifying the floor surface category according to the above-described criteria, but the robot apparatus performs any one or more of the adaptive operation amount D2, the sensor value D3, and the image data D4 on a plurality of types of floor surfaces. Is obtained, and the number of clusters is known from the learning result obtained by classifying similar features in the predetermined range into a single class (category) As a result, the robot apparatus itself may classify the floor category using the learning results obtained by performing clustering. In this case as well, the learning result is similarly stored in the floor discriminating unit 12 as a learning database for each floor category.

  The process of receiving one or more discrimination input data from the adaptive operation amount D2, the sensor value D3, the image data D4, etc., discriminating the floor surface category, and generating an appropriate gait is the motion control described above. This is done in module 300.

  Next, processing operations from the floor discrimination to walking control will be described. FIG. 5 is a flowchart showing the flow of the floor-specific walking control process. As shown in FIG. 5, first, the behavior control unit 11 issues a walking command to the walking control unit 13 (step S1). As a result, the walking control unit 13 is activated, performs an operation for inputting various sensor values D3, and calculates an adaptive operation amount by adaptive control (step S2). The adaptive operation amount D2 at this time is calculated based on the standard walking control mode as described above. Here, as an operation for inputting the various sensor values D3, there are an operation of stepping on the spot as will be described later, a normal walking operation, and the like. Since the optimal walking control mode corresponding to the surface is not selected, walking is performed according to a standard walking control mode prepared in advance. Further, when the floor surface discrimination is continuously performed, the walking is performed according to the walking control mode selected according to the floor category determined in the previous floor surface discrimination.

  Next, during the walking or stepping motion, the floor surface determination unit 12 receives the adaptive operation amount D2 of the walking control unit 13, the sensor value D3 of various sensors such as the foot sensor 360, and the image input device 251 such as a camera. Using the image data D4 and the like as input, pattern recognition of the floor category is performed using these feature quantities (step S3). Then, the floor surface determination unit 12 determines which of the floor surface categories the current floor surface corresponds to, and outputs the floor surface category as a result of the floor surface determination to the walking control unit 13 (step S4). The walking control unit 13 selects an optimal walking control mode according to the supplied floor category and performs walking control (step S5). Specifically, for example, the parameter for calculating the adaptive operation amount is selected again, and the adaptive operation amount is calculated again as the walking control amount. And when continuing a walk (step S6: Yes), the process from step S3, ie, a floor surface category discrimination | determination, is repeated again. Here, the floor category determination may be performed at all times during walking or at a predetermined timing.

  Next, a specific example of a method for calculating the adaptive operation amount D2 used when performing floor category determination will be described. In step S2 described above, it is necessary to calculate the adaptive operation amount D2 by inputting various sensor values relating to the current floor surface state. Although this may be acquired while walking, for example, an operation for discriminating the floor category can be prepared, and the floor can be discriminated by this operation. Here, as such a floor surface determination operation, a method of determining the current floor surface by performing a stepping operation on the spot will be described.

  6 (a) to 6 (d) and FIGS. 7 (a) to 7 (d) are diagrams showing how the robot apparatus 1 performs a stepping motion. First, from the upright state (FIG. 6 (a)), while balancing the aircraft (FIG. 6 (b)), the right foot is raised once (FIG. 6 (c)) and grounded to the floor (FIG. 6 (c)). d)), while also balancing the aircraft (Fig. 7 (a)), raising the left foot (Fig. 7 (b)), grounding it to the floor (Fig. 7 (c)) and returning to the upright state (FIG. 7D).

  By such a series of stepping actions, a sensor value corresponding to the floor surface can be acquired, and the walking control unit 13 calculates an adaptive operation amount D2 based on the sensor value D3. And the floor surface discrimination | determination part 12 discriminate | determines the present category of a floor surface with the sensor value D3 and the adaptive operation amount D4. The stepping motion shown here can be, for example, a walking cycle of 600 msec, a foot-lifting height of 20 mm, and a ratio of both-leg support that indicates the proportion of time that both feet are in contact during a series of stepping motions. . These values are parameters in the stepping motion.

  In addition, although it demonstrates as what acquires a sensor value and adaptive operation amount by this stepping motion etc. and performs floor surface category determination, as mentioned above, you may perform floor surface category determination while walking, and floor surface In the category determination, the floor category may be determined using the image data D4 according to the sensor value and the adaptive operation amount or individually.

  Next, a floor surface category determination method of the floor surface determination unit 12 will be described in detail. As a pattern recognition method that the floor discriminating unit 12 performs, the floor category can be discriminated according to one or more discrimination input data of the adaptive operation amount D2, various sensor values D3, and image data D4. Any pattern recognition method may be used, but here, as a specific example of input data to be input when the floor surface determination unit 12 determines the floor surface category, for example, the following Table 1 is used. An example will be described in which categories are determined by Fisher determination using what is shown.

  Here, the roll axis operation amount and the pitch axis operation amount indicate the difference between the planned trajectory (target trajectory) and the actual trajectory, and can be used as a correction amount to be corrected by the adaptive operation. . Of course, the floor surface category determination may be performed using all these values, or the floor surface determination may be performed using some values selected from these values.

  These values are output for each period of the control loop of the motion control, for example, every 4 msec. For example, the plantar acceleration is input as time series data shown in FIG. FIG. 8 is a graph showing time series values of plantar acceleration in the front direction and the lateral direction, with time on the horizontal axis and acceleration on the vertical axis. In FIG. 8, the broken line indicates data regarding the plantar acceleration in the front direction (traveling direction), and the solid line indicates data regarding the plantar acceleration in the horizontal direction.

  As shown in FIG. 8, the input foot sole acceleration is stored in time series. A feature amount is extracted from the input data obtained as described above. For example, the time series values of the plantar acceleration are Fourier transformed with a predetermined window size (fast Fourier transform: FFT), and the data on the time axis is converted into the data on the frequency axis, thereby converting the data into the feature amount.

  FIG. 9 is a graph showing a Fourier power spectrum obtained by Fourier-transforming time series values of front and lateral plantar accelerations shown in FIG. 8 with frequency components on the horizontal axis and power on the vertical axis. Also in FIG. 9, the broken line indicates data regarding the plantar acceleration in the front direction (traveling direction), and the solid line indicates data regarding the plantar acceleration in the horizontal direction. In FIGS. 8 and 9, for the sake of simplicity, only the input data for the front and lateral sole acceleration are shown in the above input data, but the input used for floor category determination is also used for other data. Similarly, data can be stored in time series, and the time series values can be Fourier transformed to be used as feature quantities.

  If this Fourier power spectrum has d-dimensional information for one value, for example, when four input values are used, it can be regarded as a 4d-dimensional feature vector in total. In the example shown in FIG. 8, d = 35 dimensions. Therefore, since the number of dimensions is large as it is, dimension compression by principal component analysis is performed, and further, it is subjected to linear discriminant analysis according to the Fisher standard. Linear discriminant analysis becomes a multi-class discriminant problem having the number of classes corresponding to the number of floor categories.

  Next, a method for classifying floor categories will be described. As described above, the floor category may learn the feature quantity of the input data for discrimination for each floor category classified according to a predetermined standard, or may learn a large amount of the feature quantity of the input data for discrimination. You may classify | categorize by the similarity etc. of a feature-value. That is, for the classification of the floor category, a required category may be prepared in advance, or it may be actually walked and classified when it is determined that classification is necessary. Further, additional learning may be performed by adding categories as necessary.

Here, a case where the floor category is classified on the human side using, for example, the following floor category classification criteria (features) will be described.
1: Elastic modulus (Young's modulus)
2: Friction coefficient 3: Rolling coefficient 4: Combination of lengths of hairs of carpets and carpets 5: 1 to 4

  That is, if a different floor surface category is prepared for each floor surface having different one or more features, a walking control mode corresponding to the floor surface category can be prepared. This floor category depends on the size of the influence of the robot device, because the features that affect walking differ depending on the structure of the target robot device and the floor that is actually the target of walking. Therefore, it is necessary to select a categorization standard.

  Of the various sensors, adaptive operation amounts, and image data, the feature amount of the input data (determination input data) used for floor category discrimination is learned in advance for each classified floor (floor category). A floor database is stored in the surface determination unit 12 as a learning database, and the floor surface category is determined by comparing with the above-described stepping motion or the feature amount of the input data for determination obtained from the current floor surface during walking. Can do.

  In the present embodiment, a floor category considered to be necessary in a home environment will be described assuming a robot apparatus as a consumer product for home use. In order to classify the floor surface category in the home environment, “4: length of hairs of carpets and carpets” can be used as a classification standard.

Specifically, when the general floor surface in the home environment is defined as “plate material (flooring)” and “carpet” and classified according to the above classification criteria, the following three categories are required.
Floor 1: carpet with long bristle feet (length of bristle feet 6mm or more)
Floor 2: carpet with short hairs (length of hairs less than 6mm)
Floor 3: Flooring (corresponds to 0mm hair length)

  Here, the length of the hairy feet is used as a classification standard because the inventors of the present application have the most influence on the walking performance when the robot apparatus actually walks and conducts the experiment. This is because they have found out. That is, it has been found that the walking control unit 13 needs to prepare different walking control modes (parameters or walking algorithms) for each of these three floor surface categories.

  Next, information (determination input data) used for classifying actual floor surfaces into the above three floor surface categories will be described. Of the above-mentioned input data to be input to the floor surface determination unit 12, floor category determination is performed using one or a plurality of combinations as input data for determination. This input data for discrimination is obtained by processing (pre-processing) the time series values as described above by some method and applying it to the floor category in the subsequent linear discrimination.

  Examples of the pretreatment include the following methods. First, in the case of one-dimensional information, that is, the above-described sensor value and adaptive operation amount are obtained, for example, at the time of the above-described stepping action or walking, the sensor value and adaptation obtained at the time of such operation The operation amount includes information about the floor surface in some form. Examples of the information relating to the floor surface, that is, the feature amount of the floor surface include, for example, a Fourier power spectrum feature or an autocorrelation feature.

  In the case of two-dimensional information, that is, for example, an input image from an image input unit 251 such as a camera, the information is two-dimensional and processing is different because it is necessary to divide the region. It is possible to use the same feature amount as in the one-dimensional case. As a method for obtaining the feature of the Fourier power spectrum as the feature amount of the input image, for example, first, a floor area is separated or extracted from the input image. As a method of extracting the floor area from the input image, a dominant plane is extracted by Hough transform from the binocular stereo parallax value, and coordinate conversion is performed to determine whether or not the floor area is the floor area. There is a method of extracting.

  Next, a two-dimensional Fourier power spectrum of the extracted floor area is obtained. Then, this two-dimensional Fourier power spectrum is developed into a one-dimensional vector by a scanning method. Accordingly, two-dimensional information such as an input image can be handled in the same manner as the above-described one-dimensional floor feature.

  Hereinafter, as a specific example of the floor surface category determination, a method for determining the floor surface category by converting two pieces of input data of the roll axis operation amount and the pitch axis operation amount as the determination input data into feature amounts will be described. 10 and 11 are graphs showing Fourier power spectra obtained by Fourier transforming time series values at a certain point in time for the roll axis operation amount and the pitch axis operation amount, respectively. 10 and 11, the horizontal axis corresponds to the frequency, and the vertical axis corresponds to the power. These two data are handled as one data block as shown in FIG. That is, the horizontal axis (frequency component) of the Fourier power spectrum of the roll axis manipulated variable is converted to, for example, 0 to 50, and the horizontal axis (frequency component) of the Fourier power spectrum of the pitch axis manipulated variable is converted to, for example, 50 to 100. Thus, two pieces of data can be handled as one piece of data. With such an operation, a plurality of feature quantities can be used in combination.

  The Fourier power spectrum obtained by combining the feature amounts of the roll axis operation amount and the pitch axis operation amount obtained in this way is used as an input for linear discriminant analysis, and thereby the floor surface category is discriminated. For example, in the case of a robot apparatus having a walking cycle of 8 msec, this composite Fourier power spectrum can be generated up to 125 times per second. Of course, the linear discriminant analysis may be performed at a predetermined timing and may not be performed every walking cycle.

  Further, the data shown in FIG. 10 can take a vector representation as shown in (1) below.

  Here, as described above, it is necessary for the robot apparatus to learn in advance the feature quantity that is input to the linear discriminant analysis for each floor category. For learning of floor categories, for example, a large amount of vectors x shown in the above equation (1) may be collected by performing the above-mentioned stepping motion on the floors belonging to each floor category. For example, 1250 vectors x can be collected in a 10-second stepping motion. Here, there are three floor surface categories: “flooring”, “short-haired carpet (less than 6 mm long hair)”, “long-haired carpet (more than 6 mm long hair)” In order to explain that the floor category is prepared, the floor category is learned by performing stepping on the spot for 10 seconds on each of the three floors and collecting 1250 vectors x each. .

  When this learning is performed, if there is no difference in the feature amount for each classified floor category, this indicates that it is difficult to determine the floor category. In this case, there are essentially two possibilities. One is when trying to classify categories that are essentially unclassifiable. Specifically, for example, a flat stone and a flat glass are both hard and flat floor surfaces, and there is almost no difference for the robot apparatus and classification is impossible. The other is a case where the feature quantity used for discrimination does not have a sufficient degree of freedom or a low degree of freedom for the discrimination problem that the learning algorithm is trying to solve. For example, it is a case where classification cannot be performed by using only one potentiometer value due to an insufficient amount of information. In any case, if the floor category can be classified according to the feature quantity of the input data for discrimination by learning in advance, it indicates that the category can be actually discriminated using the feature quantity.

  In this embodiment, a Fisher discriminant linear discriminant analysis is performed for this floor category discrimination. This is a technique for performing projection on a space (discriminant space) that minimizes the variance in each category and maximizes the variance between the categories, and performs category discrimination based on the distance in the space.

Here, the floor category is referred to as a floor class, and the k-th class among the K classes is referred to as a class C _k . The linear discriminant analysis is performed by converting the input vector x into the vector y in the discriminant space shown in FIG. 13 by the conversion formula shown in the following formula (2), and looking at the distance between each point (vector) in the space. .

If the distance y between the value y in the discriminant space for the input vector x and the class center in the discriminant space of the average value (class center) of the data of class C _k is shorter than the distance from other class centers, the input vector It is determined that x belongs to class C _k . That is, a short distance indicates that the data values are similar. Hereinafter, in this specification, the average value (class center) of data of class C _k is _denoted as <x _k >, and the class center in the discriminant space is _denoted as <y _k >.

The coefficient matrix A in the conversion formula (2) is obtained as follows. That is, if the class average is <x _k >, the total average is <x _T >, and the intra-class covariance Σ _W and inter-class covariance Σ _B are defined as in the following formulas (3) and (4), respectively, The coefficient matrix A is obtained as a solution to the eigenvalue problem shown in (5) below.

: Average value of data of class Ck (class center) = <x _k >
: Average value of all data = <x _T >

The coefficient matrix A is prepared in advance using the many vectors x described above. When the robot apparatus actually determines the floor surface, the coefficient matrix A prepared in advance, Similarly, category classification of the input data P is performed using the class center <y _k > calculated in advance. That is, the input data P can be classified into a class having the smallest distance L _n between the input data P and the class center <y _k >.

  The floor category as the floor discrimination result obtained in this way is supplied to the walking control unit 13. The walking control unit 13 can change an algorithm for calculating a trajectory such as a roll axis and a pitch axis and a parameter (coefficient) used for the trajectory calculation in accordance with the floor category thus input. Examples of the parameters that can be changed include a stride, a walking cycle (time taken for one step), a ratio of supporting both legs, or a PID gain. Table 2 below shows specific examples of parameter changes.

  In this way, according to the floor category, it becomes possible to select the optimal walking control (adaptive operation amount), such as a parameter for converting from various sensor values, and to select a walking control mode, depending on the state of the floor. By adopting an appropriate walking algorithm, stable walking can be realized even when the floor surface state changes greatly. In addition, by recognizing the floor category, the most suitable walking control mode for stable walking is selected, so that it is possible to bring out other performance such as walking faster than usual.

  Furthermore, by recognizing a floor surface that is not suitable for walking, it is possible to control to stop walking as “cannot walk”.

  Furthermore, the performance of the environmental map can be improved by associating the floor category information with the environmental map. In other words, if the floor category information is stored, the walking control mode such as the walking algorithm is automatically changed and adaptively performed without moving the stepping action described above when moving from the floor to the carpet room, for example. A gait can be generated.

  Furthermore, using the knowledge about the floor surface, it is possible to identify a favorite floor, a disliked floor, a good floor, a poor floor, etc., and provide this as interactive content with the user.

  Next, an embodiment of the present invention in which floor category determination is performed using actual data will be described. Here, for the sake of simplification, only one piece of input data for determination is used. However, as described above, it is a matter of course that a plurality of data may be integrated into one to perform floor surface determination. . In addition, as described above, if the discrimination input data is appropriately selected, it is possible to discriminate the floor category with a single discrimination input data.

In this embodiment, the time series value of the roll axis operation amount during the stepping motion described above is used as the data used for the floor surface discrimination. 14 (a), 15 (a), and 16 (a), the floor surface is a flooring F _A , a carpet F _{B with} a hairy foot length of less than 6 mm, and a hairy foot length of 6 mm or more. schematically shows a carpet _{F C} diagram, FIG. 14 (b), the FIG. 15 (b), the and FIG. 16 (b), the time series values obtained upon stepping action respectively at each floor, Fig. 14 (c ), FIG. 15C, and FIG. 16C are graphs showing the floor surface feature values by Fourier transforming each time series value. FIG. 17 is a diagram for explaining a decision boundary when discriminating these three classes, and FIG. 18 is a schematic diagram showing a state where the robot apparatus performs a stepping operation to perform floor surface discrimination.

First, in order to determine the floor surface category, the feature amount of the roll axis operation amount in each floor surface category is learned. At the time of learning, as shown in FIG. 18, a floor surface to be classified into the floor surface category is prepared, and a large number of feature quantities are secured by causing the robot device to step on the floor surface. That is, as shown in FIGS. 14 to 16, the feature value of each floor surface is obtained from the time series data of the roll axis operation amount of each floor surface. The three-class discriminant problem is projected onto the two-dimensional space shown in FIG. 17 by linear discriminant analysis. As shown in FIG. 17, a distribution of data is obtained by collecting and projecting a plurality of feature amounts on each of the floor surfaces F _{A to} F _C , thereby obtaining a boundary for each floor surface indicated by a solid line in the figure. A decision line can be drawn.

  When the robot apparatus performs walking control according to the state of the floor surface with respect to the two-dimensional space learned in this manner and drawn with the class boundary decision line, the roll axis operation amount is reduced by walking or the above-described stepping action. A time series value is acquired, is converted into a feature value by Fourier transform, and is projected onto a two-dimensional space shown in FIG. Then, by comparing the Euclidean distance from each class center point and classifying the class into the class with the shortest distance, the floor surface category can be determined (class determination).

The walking control unit, three floor category, i.e. flooring F _A, short carpet F _{B haired,} walking control forms are prepared, such as, for example, walking algorithm corresponding to long carpet F _C haired, above When a floor category is input by the method, an adaptive operation amount is calculated according to a walking algorithm corresponding to the floor category, and stable walking can be performed.

  It should be noted that the present invention is not limited to the above-described embodiments, and various modifications can be made without departing from the scope of the present invention. In the above-described embodiment, the floor category is classified into a predetermined category in advance, and the feature amount of the input data for discrimination is learned for each category. A large amount of features (feature vectors) in the input data for discrimination are obtained at, and the data (feature vectors) in the pattern space is similar using the Euclidean distance in the pattern space and similarities based on variance. Classification of floor categories may be performed by performing clustering to divide into such subsets.

1 is a perspective view showing an overview of a robot apparatus according to an embodiment of the present invention. It is a figure which shows typically the joint freedom degree structure which the robot apparatus comprises. It is a schematic diagram which shows the control system structure of the robot apparatus. It is a block diagram which shows functionally only the principal part required for the walk control apparatus classified by floor among the robot apparatus. It is a flowchart which shows the flow of the walk control processing classified by floor surface. (A) thru | or (d) is a figure which shows a mode that the robot apparatus is performing stepping motion. Similarly, (a) to (d) are diagrams illustrating a state in which the robot apparatus 1 is performing a stepping motion, and are diagrams illustrating an operation after FIG. 6 (d). It is a graph which shows the time series value of the plantar acceleration of a front direction and a horizontal direction by taking time on a horizontal axis and taking acceleration on a vertical axis. FIG. 9 is a graph showing a Fourier power spectrum obtained by performing a Fourier transform on the time series values of the front and lateral sole accelerations shown in FIG. 8 with the frequency on the horizontal axis and the power on the vertical axis. It is a graph which shows the Fourier power spectrum characteristic obtained by carrying out the Fourier transform of the time series value in a certain time of the roll axis operation amount. It is a graph which shows the Fourier power spectrum characteristic obtained by carrying out the Fourier transform of the time series value in a certain time of a pitch axis manipulated variable. It is a figure which shows the example which handles two data as one lump of data. It is a figure which shows the discrimination | determination space for projecting the feature-value and discriminating a floor surface category. (A) is a schematic diagram showing a floor surface of the flooring, (b) is a time series value obtained when a stepping action is performed on the floor surface of the flooring, and (c) is a Fourier transform of the obtained time series value. It is the graph which made it the feature-value of the floor surface of a flooring. (A) is a schematic diagram showing the floor of a carpet with a hairy foot length of less than 6 mm, and (b) is obtained when a stepping action is performed on the floor of a carpet with a hairy foot length of less than 6 mm. A time series value, (c), is a graph showing the characteristic amount of the floor surface of a carpet having a length of hair foot of less than 6 mm by Fourier transforming the obtained time series value. (A) is a schematic diagram showing the floor surface of a carpet having a hair length of 6 mm or more, and (b) is obtained when a stepping action is performed on the floor surface of a carpet having a hair foot length of 6 mm or more. A time series value, (c), is a graph showing the characteristic amount of the floor surface of a carpet having a length of a hair foot of 6 mm or more by Fourier-transforming the obtained time series value. It is a figure explaining the class decision boundary in the two-dimensional space at the time of discriminating 3 classes. It is a schematic diagram which shows a mode that the robot apparatus is stepping on and performing floor surface discrimination.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 Robot apparatus, 2 trunk unit, 4R / L arm unit, 104 trunk pitch axis, 105 trunk roll axis, 106 trunk yaw axis, 11 action control part, 12 floor surface discrimination | determination part, 13 walking control part

Claims (13)

In a legged mobile robot that can operate autonomously,
Based on information about the current moving surface, the moving surface determining means for determining a moving surface category is a category of the current moving surface from a plurality of categories, which are prepared in advance,
A predetermined movement control mode is selected according to the moving plane category determined by the moving plane determination unit , and at least based on a selected movement control mode from sensor values output from sensors provided on the legs. calculating a movement control amount to have a movement control means for controlling the movement,
The movement control means has a different movement control model or movement algorithm for each category, selects a movement control model or movement algorithm corresponding to the input moving surface category, and uses the sensor value to move the movement. A correction amount from the control model or an adaptive operation amount according to the movement algorithm is calculated as the movement control amount,
The moving surface determination means is a robot apparatus that uses the movement control amount as information on the current moving surface . The movement control means converts the sensor value into the movement control amount using a predetermined parameter, has a parameter different for each category, and is a parameter corresponding to the input moving surface category. robot apparatus according to claim 1, wherein you calculate the movement control amount using. The moving surface discriminating means, the sensor value output from the sensor provided to at least the leg portions of that use according to claim 1, wherein the information relating to the current movement plane robot apparatus. Having imaging means;
The moving surface discriminating means, a robot apparatus according to claim 1, wherein the extracted the moving surface region moving surface region from an input image captured by the image pickup means to use as information on the current movement plane. The moving surface discriminating means, the robot apparatus to that claim 1, wherein determining the moving surface category by pattern identification information relating to the current movement plane. The moving surface discriminating means has a learned feature amount in which a feature amount of information related to the moving surface is learned in advance for each category classified according to a predetermined criterion, and includes a feature amount of information related to the current moving surface, robot apparatus according to claim 1, wherein you determine the moving surface category by comparing the trained feature amount. The plurality of categories are classified according to learned feature amounts learned in advance with respect to feature amounts of information on moving surfaces,
The movement plane determining means that the current moving surface information of the feature amount and the trained features and robot apparatus to that claim 1, wherein determining the moving surface category by comparing related. Robot apparatus according to claim 1, wherein that having a motion generating means for generating a stepping motion to determine the movement plane category. The movement control means calculates a movement control amount based on a selected movement control form from a sensor value output from at least a sensor provided on a leg during the stepping action,
The movement plane determining means, robotic apparatus of claim 8, wherein to use the movement control amount as information relating to the movement plane. The movement control means calculates a time-series movement control amount based on a selected movement control form from at least a time-series sensor value output from a sensor provided on the leg.
The movement plane determining means, the time the movement control amount of the sequence features using Fourier transform data, the feature quantity of pattern identified by the robot apparatus according to claim 5, wherein you determine moving surface category. The movement plane determining means, and the feature amount data obtained by Fourier transform of the sensor values of the time series output from the sensor provided on at least the legs, claim you determine moving surface category by a pattern identification of the feature quantity 5. The robot apparatus according to 5 . The movement plane determining means, determine the moving surface category continuously during movement claim 1, wherein the robotic device. In a movement control method of a legged mobile robot device that can operate autonomously,
Based on information about the current moving surface, the moving surface determination step of determining a moving surface category is a category of the current moving surface from a plurality of categories, which are prepared in advance,
A predetermined movement control mode is selected according to the moving plane category determined in the moving plane determination step, and at least a sensor value output from a sensor provided on the leg is selected as the selected movement control mode. calculating a movement control amount to have a movement control step of controlling the movement based,
In the movement control step, a movement control model or a movement algorithm that is different for each category is selected, a movement control model or a movement algorithm corresponding to the input movement plane category is selected, and the movement is performed using the sensor value. A correction amount from the control model or an adaptive operation amount according to the movement algorithm is calculated as the movement control amount,
In the moving surface discrimination step, a movement control method for a robot apparatus that uses the movement control amount as information on the current moving surface .

Images