CN117400235A - A method for opening and closing a door by a robot and the robot - Google Patents
A method for opening and closing a door by a robot and the robot Download PDFInfo
- Publication number
- CN117400235A CN117400235A CN202210780246.XA CN202210780246A CN117400235A CN 117400235 A CN117400235 A CN 117400235A CN 202210780246 A CN202210780246 A CN 202210780246A CN 117400235 A CN117400235 A CN 117400235A
- Authority
- CN
- China
- Prior art keywords
- door
- robot
- axis parameter
- interaction force
- parameter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J11/00—Manipulators not otherwise provided for
- B25J11/008—Manipulators for service tasks
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1656—Programme controls characterised by programming, planning systems for manipulators
- B25J9/1661—Programme controls characterised by programming, planning systems for manipulators characterised by task planning, object-oriented languages
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1679—Programme controls characterised by the tasks executed
-
- E—FIXED CONSTRUCTIONS
- E05—LOCKS; KEYS; WINDOW OR DOOR FITTINGS; SAFES
- E05F—DEVICES FOR MOVING WINGS INTO OPEN OR CLOSED POSITION; CHECKS FOR WINGS; WING FITTINGS NOT OTHERWISE PROVIDED FOR, CONCERNED WITH THE FUNCTIONING OF THE WING
- E05F15/00—Power-operated mechanisms for wings
- E05F15/70—Power-operated mechanisms for wings with automatic actuation
Landscapes
- Engineering & Computer Science (AREA)
- Mechanical Engineering (AREA)
- Robotics (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Manipulator (AREA)
Abstract
The application discloses a method and robot of robot switch door, can be applied to artificial intelligence field, and scalable robot is the application in fields such as family service, wisdom commodity circulation, includes: the robot acquires a first interaction force between the robot and the target door plate, inputs the first interaction force into the trained neural network, obtains a first door axis parameter (representing the door axis position of the target door plate) and a second door axis parameter (representing the door axis direction of the target door plate) which are output, and finally adjusts a second interaction force applied to the target door plate according to the first door axis parameter and the second door axis parameter. When the second interaction force reaches the preset condition, the target door plate moves around the door shaft under the action of the second interaction force. The method and the device utilize the neural network to convert the detected interaction force of the robot (such as a mechanical arm) and the door into the door axis parameter, and adjust the interaction force based on the door axis parameter, so that the interaction force in the non-door opening direction in the opening/closing process is ensured to be smaller than a threshold value (such as zero), and the door panel is prevented from being damaged.
Description
Technical Field
The present disclosure relates to the field of robots, and in particular, to a method for opening and closing a door of a robot and a robot.
Background
The robot has the door opening capability, so that the activity space and the working range of the robot can be greatly expanded, and the service function and the use experience of the robot are improved. The configuration and the attribute of the door in actual production and life are diversified, and how to design an algorithm so that the robot can have safe and generalized door opening capability is a challenge in the research of a robot control algorithm.
At present, two main door opening modes of robots exist: 1. the robot door opening method based on vision utilizes images and point cloud data to acquire door parameters such as the position of a door handle, the posture of a door and the like, and then drives a mechanical arm to finish a door opening task. 2. The robot door opening method based on force sense drives the mechanical arm to move towards the direction with the minimum interaction force all the time, the door has only one degree of freedom, and when the mechanical arm moves in the non-door opening direction, the interaction force between the mechanical arm and the door can be very large.
In the mode 1, once the acquired visual information has errors, the interaction force between the robot and the door is very large, and the door is damaged; in the mode 2, the direction with the smallest acting force is searched by a continuous searching mode, and huge step interaction force can be generated between the robot and the door due to the exploring characteristic of the robot, so that larger force can be generated in the direction that the door cannot move, and the door can be damaged.
Disclosure of Invention
The embodiment of the application provides a method for opening and closing a door of a robot and the robot, which utilize a neural network to convert the detected interaction force of the robot (such as a mechanical arm) and the door into a door-axis parameter, and reversely adjust the interaction force based on the door-axis parameter, thereby avoiding the problem that the door-axis is damaged due to huge step interaction force between the robot and the door in the opening/closing process.
Based on this, the embodiment of the application provides the following technical scheme:
in a first aspect, an embodiment of the present application provides a method for opening and closing a door of a robot, which may be used in the field of artificial intelligence, and may be used in the fields of home services, smart logistics, and the like, where the method includes: first, the robot may acquire an interaction force between the robot and the target door panel, which may be referred to as a first interaction force. Then, the robot inputs the acquired first interaction force into the trained neural network to obtain two output door-axis parameters which are respectively recorded as first door-axis parametersAnd a second door axis parameter->Wherein the first door spindle parameter->For characterizing the door axis position of the target door panel, and the second door axis parameter for characterizing the door axis orientation of the target door panel >These two door-axis parameters can be noted +.>After obtaining the first door-axis parameter->And a second door axis parameter->After that, the robot is according to the first door axis parameter +.>And a second door axis parameter->The interaction force applied to the target door panel is adjusted, which may be referred to as a second interaction force.
In the above embodiment of the application, the detected interaction force of the robot (e.g., the mechanical arm) and the door is converted into the door-axis parameter by using the neural network, and the interaction force is reversely adjusted based on the door-axis parameter, so that the problem that the door-axis is damaged due to the fact that huge step interaction force occurs between the robot and the door in the opening/closing process is avoided.
In a possible implementation manner of the first aspect, the method further includes: and the robot repeatedly executes the process until a preset condition is reached by taking the second interaction force as a new first interaction force, so that the target door plate moves around the door shaft under the action of the second interaction force applied last time, and the door opening or closing task is completed.
In the above embodiment of the present application, the robot may periodically adjust the first interaction force based on feedback of the trained neural network until reaching a preset condition, so as to improve accuracy of the neural network in constructing the first interaction force and door axis parameter mapping, and avoid damage to the door panel due to inaccurate door axis parameter estimation.
In a possible implementation manner of the first aspect, the case that the preset condition is reached includes, but is not limited to: (1) the number of times the above-described process is performed reaches a preset number of times. (2) The component of the last applied second interaction force in a non-first direction is smaller than a preset threshold (for example, the interaction force is zero), and the first direction is the direction perpendicular to the door axis.
In the above embodiments of the present application, several cases of reaching the preset condition are specifically described, and the present application has wide adaptability.
In a possible implementation manner of the first aspect, the robot includes an end effector (e.g., a mechanical arm) and a force sensor, and the process of obtaining the first interaction force between the robot and the target door panel may be: the robot applies first interaction force to the target door plate through the end effector, and then the force sensor collects information such as the magnitude and moment of the applied first interaction force.
In the above embodiments of the present application, it is specifically described how the robot collects the first interaction force, which is capable of being implemented.
In a possible implementation manner of the first aspect, the robot is configured to control the first door spindle according to the first door spindle parameterAnd a second door axis parameter->The specific manner of adjusting the second interaction force applied to the target door panel may be: firstly, according to the first door-axis parameter +. >Second door axis parameter->Determining the direction of movement of the end effector>Thereafter, according to the direction of movement->And adjusting a second interaction force applied by the end effector to the target door panel.
In the above embodiments of the present application, it is specifically described that the robot is capable of converting door-axis parameters into actual movements of the end effector.
In a possible implementation manner of the first aspect, if the robot performs a door opening task, a movement direction of an end effector of the robot isWherein the method comprises the steps of，For the first door axis parameter, < >>Is a second axis parameter, so that the end effector of the robot follows +.>And the door opening task is completed.
In the above embodiments of the present application, it is specifically described how the movement direction of the end effector of the robot is determined when the robot performs the door opening task, and flexibility is provided.
In a possible implementation manner of the first aspect, if the robot performs a door closing task, the movement direction of the end effector of the robot isSo that the end effector of the robot is along +.>And the door closing task is completed.
In the above embodiments of the present application, it is specifically described how the movement direction of the end effector of the robot is determined when the robot performs the door closing task, and flexibility is provided.
In a possible implementation manner of the first aspect, the force sensor may be a six-dimensional force sensor. In this case, the first interaction force collected by the robot may be expressed as: [ f x ,f y ,f z ,T x ,T y ,T z ]Wherein f represents the force, T represents the moment, x, y, z are three dimensions of the coordinate system, in this embodiment, the end effector is taken as the origin of the coordinate system.
In the above embodiments of the present application, a specific embodiment of the force sensor is specifically described, which is typical.
In a possible implementation manner of the first aspect, the trained neural network is obtained by training the neural network with a training data set based on a target loss function; the training data set comprises a plurality of training data, and each training data comprises interaction force between the robot and the training door plate. The output of the neural network is a first predicted door axis parameter corresponding to a first real door axis parameter of a first training door panel and a second predicted door axis parameter corresponding to a second real door axis parameter of the first training door panel, and the first training door panel is one of the plurality of training door panels.
In the above embodiment of the present application, it is specifically described how the trained neural network is trained, the mapping from interaction force to the door-axis parameter is constructed by using the neural network, the input of the neural network is the interaction force collected by the force sensor, and the output of the neural network is the position and orientation of the door-axis. And the neural network is trained by using gates with different parameters, so that the accuracy and generalization of the construction mapping of the neural network are ensured.
In a possible implementation manner of the first aspect, the objective loss function includes: and the first n power is the n power of the difference between the first predicted door-axis parameter and the first real door-axis parameter, and the second n power is the n power of the difference between the second predicted door-axis parameter and the second real door-axis parameter, wherein n is more than or equal to 1.
In the above embodiments of the present application, a specific manner of determining the objective loss function is described, which is capable of being implemented.
A second aspect of the embodiments of the present application provides a robot having a function of implementing the method of the first aspect or any one of the possible implementations of the first aspect. The functions can be realized by hardware, and can also be realized by executing corresponding software by hardware. The hardware or software includes one or more modules corresponding to the functions described above.
A third aspect of the embodiments of the present application provides a robot, which may include a memory, a processor, and a bus system, where the memory is configured to store a program, and the processor is configured to invoke the program stored in the memory to perform a method according to the first aspect or any one of possible implementations of the first aspect of the embodiments of the present application.
A fourth aspect of the embodiments of the present application provides a computer readable storage medium having instructions stored therein which, when run on a computer, cause the computer to perform the method of the first aspect or any one of the possible implementations of the first aspect.
A fifth aspect of the embodiments of the present application provides a computer program which, when run on a computer, causes the computer to perform the method of the first aspect or any one of the possible implementations of the first aspect.
A sixth aspect of the embodiments of the present application provides a chip, where the chip includes at least one processor and at least one interface circuit, where the interface circuit is coupled to the processor, and where the at least one interface circuit is configured to perform a transceiver function and send instructions to the at least one processor, where the at least one processor is configured to execute a computer program or instructions, where the at least one processor is configured to implement a method according to the first aspect or any one of the possible implementations of the first aspect, where the function is implemented by hardware, or by software, or by a combination of hardware and software, and where the hardware or software includes one or more modules corresponding to the above functions. In addition, the interface circuit is used for communicating with other modules outside the chip.
Drawings
Fig. 1 is a schematic diagram of a robot control flow provided in an embodiment of the present application;
fig. 2 is a schematic flow chart of a method for opening and closing a door of a robot according to an embodiment of the present application;
FIG. 3 is a schematic diagram of door spindle parameters provided in an embodiment of the present application;
FIG. 4 is a system architecture diagram of a data processing system provided in an embodiment of the present application;
fig. 5 is a schematic flow chart of a training method of a neural network according to an embodiment of the present application;
FIG. 6 is a schematic diagram of a neural network simulation training framework provided in an embodiment of the present application;
fig. 7 is a schematic diagram of a system architecture to which the method for opening and closing a door of a robot according to the embodiment of the present application is applied;
fig. 8 is a schematic diagram of an application scenario of a method for opening and closing a door of a robot according to an embodiment of the present application;
FIG. 9 is a schematic diagram of a network structure of a trained neural network according to an embodiment of the present application;
FIG. 10 is a schematic view of door panels with different parameters according to an embodiment of the present application;
fig. 11 is a schematic structural diagram of a robot according to an embodiment of the present disclosure;
fig. 12 is another schematic structural view of a robot according to an embodiment of the present disclosure;
fig. 13 is a schematic diagram of an exemplary configuration scenario for operation provided in an embodiment of the present application.
Detailed Description
The embodiment of the application provides a method for opening and closing a door of a robot and the robot, which utilize a neural network to convert the detected interaction force of the robot (such as a mechanical arm) and the door into a door-axis parameter, and reversely adjust the interaction force based on the door-axis parameter, thereby avoiding the problem that the door-axis is damaged due to huge step interaction force between the robot and the door in the opening/closing process.
The terms first, second and the like in the description and in the claims of the present application and in the above-described figures, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances and are merely illustrative of the manner in which the embodiments of the application described herein have been described for objects of the same nature. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of elements is not necessarily limited to those elements, but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
In order to better understand the schemes of the embodiments of the present application, related terms and concepts of neural networks and element learning that may be involved in the embodiments of the present application are described below. It should be understood that the related conceptual illustrations may be limited by the specific embodiments of this application, but are not intended to limit the application to that specific embodiment, and that differences between the specific embodiments may exist, and are not specifically defined herein.
(1) Neural network
A neural network may be composed of neural units, and is understood to mean, in particular, a neural network having an input layer, an hidden layer, and an output layer, where in general, the first layer is the input layer, the last layer is the output layer, and the middle layers are all hidden layers. Among them, the neural network with many hidden layers is called deep neural network (deep neural network, DNN). The operation of each layer in the neural network may be expressed mathematicallyTo describe, from the physical level, the operation of each layer in a neural network can be understood as the completion of the transformation of input space into output space (i.e., row space to column space of the matrix) by five operations on the input space (set of input vectors), including: 1. dimension increasing/decreasing; 2. zoom in/out; 3. rotating; 4. translating; 5. "bending". Wherein the operations of 1, 2, 3 are defined by +. >The operation of 4 is completed by "+b", and the operation of 5 is completed by "a ()". The term "spatial" is used herein because the object being classified is not a single thing, but a class of things, and space refers to the collection of all individuals of such things, where W is the weight matrix of the layers of the neural network, each value in the matrix representing the weight value of one neuron of that layer. The matrix W determines the spatial transformation of the input space into the output space described above, i.e. the neural network eachThe W of the layer controls how the space is transformed. The purpose of training the neural network is to finally obtain the weight matrix of all layers of the trained neural network. Thus, the training process of the neural network is essentially a way to learn and control the spatial transformation, and more specifically to learn the weight matrix.
(2) Loss function
In the process of training the neural network, because the output of the neural network is expected to be as close to the value actually expected, the weight matrix of each layer of the neural network can be updated according to the difference between the predicted value of the current network and the actually expected target value (of course, there is usually an initialization process before the first update, that is, the parameters are preconfigured for each layer in the neural network), for example, if the predicted value of the network is higher, the weight matrix is adjusted to be lower than the predicted value, and the adjustment is continuously performed until the neural network can predict the actually expected target value. Thus, it is necessary to define in advance "how to compare the difference between the predicted value and the target value", which is a loss function (loss function) or an objective function (objective function), which are important equations for measuring the difference between the predicted value and the target value. Taking the loss function as an example, the higher the output value (loss) of the loss function is, the larger the difference is, and the training of the neural network becomes the process of reducing the loss as much as possible.
In the training process of the neural network, an error Back Propagation (BP) algorithm can be adopted to correct the size of parameters in an initial neural network model, so that the reconstruction error loss of the neural network model is smaller and smaller. Specifically, the input signal is forwarded until the output is generated with error loss, and the parameters in the initial neural network model are updated by back-propagating the error loss information, so that the error loss converges. The back propagation algorithm is a back propagation motion that dominates the error loss, and aims to obtain parameters of the optimal neural network model, such as a weight matrix.
Embodiments of the present application are described below with reference to the accompanying drawings. As one of ordinary skill in the art can appreciate, with the development of technology and the appearance of new scenes, the technical solutions provided in the embodiments of the present application are applicable to similar technical problems.
The door is a common switching device in life, and has the main function of isolating a physical space, and the robot has the door opening/closing capability and can expand the reachable range of the robot in the space. The robot door opening and closing method provided by the embodiment of the application can be applied to service robots such as inspection robots and cleaning robots, and allows the robots to finish tasks such as entering rooms and cleaning wardrobes. The common door can be abstracted into a rigid body rotating around a fixed rotating shaft, when the position of the rotating shaft is determined, the robot can finish the door opening/closing task only by 1 degree of freedom, but the degree of freedom of an end effector (such as a mechanical arm) of the robot is generally 6 or more, and a large number of redundant degrees of freedom exist in the door opening process of the robot. Based on this, the embodiment of the application converts the interaction force of the mechanical arm detected by the force sensor and the door into door axis parameters (such as the door axis position and the door axis orientation) by using the neural network, and further generates the terminal motion control instruction of the mechanical arm from the door axis parameters, for example, the terminal motion control instruction on the robot (such as the robot control flow shown in fig. 1) can be generated by using the terminal motion controller on the robot, so that the interaction force in the non-door opening direction in the door opening/closing process is ensured to be smaller than the preset threshold, for example, the interaction force in the non-door opening direction is zero.
Referring specifically to fig. 2, fig. 2 is a schematic flow chart of a method for opening and closing a door of a robot according to an embodiment of the present application, and it should be noted that any opening and closing device with a single rotation shaft may be used in the method for opening and closing a door according to an embodiment of the present application, including but not limited to a door panel, a toilet lid, a vehicle charging lid, a vehicle fuel tank lid, a vehicle engine lid, a well lid, a box lid, a dustbin lid, etc., for convenience of explanation, the following embodiments of the present application take the door panel as an example, and the method may specifically include the following steps:
201. a first interaction force between the robot and the target door panel is obtained.
First, the robot may acquire an interaction force between the robot and the target door panel, which may be referred to as a first interaction force. Specifically, in some embodiments of the present application, the robot includes an end effector (e.g., a mechanical arm) and a force sensor, and the robot applies a first interaction force to the target door panel through the end effector, and then the force sensor collects information about the magnitude, moment, and the like of the applied first interaction force.
It should be noted that in some embodiments of the present application, the force sensor may be a six-dimensional force sensor, in which case the first interaction force acquired by the robot may be expressed as: [ f x ,f y ,f z ,T x ,T y ,T z ]Wherein f represents the force, T represents the moment, x, y, z are three dimensions of the coordinate system, in this embodiment, the end effector is taken as the origin of the coordinate system.
202. And inputting the first interaction force into the trained neural network to obtain a first door shaft parameter and a second door shaft parameter which are output, wherein the first door shaft parameter is used for representing the door shaft position of the target door plate, and the second door shaft parameter is used for representing the door shaft orientation of the target door plate.
Then, the robot inputs the acquired first interaction force into the trained neural network to obtain two output door-axis parameters which are respectively recorded as first door-axis parametersAnd a second door axis parameter->Wherein the first door spindle parameter->For characterizing the door axis position of the target door panel, and the second door axis parameter for characterizing the door axis orientation of the target door panel>These two door spindle parameters can be recorded asWhere it is requiredNote that in some embodiments of the present application, the first door spindle parameter +.>And a second door axis parameter->The calibration is performed by using the end effector of the robot as the origin of the coordinate system, and referring to fig. 3, fig. 3 is a schematic diagram of the door spindle parameter according to the embodiment of the present application, wherein the end effector of the robot is used as the origin O of the coordinate system end 。
As an example, when the force sensor is a six-dimensional force sensor, the trained neural network will detect the first interaction force [ f ] detected by the six-dimensional force sensor x ,f y ,f z ,T x ,T y ,T z ]Conversion of door spindle parameters
It should be noted that, in the embodiment of the present application, the neural network is trained by using a machine learning method to obtain the capability of converting the interaction force into the door-axis parameter, the neural network refreshes the prediction of the door-axis parameter by the collected interaction force, and the training goal of the neural network is to improve the accuracy of the door-axis parameter estimation. Referring to fig. 4, fig. 4 is a system architecture diagram of a data processing system provided in an embodiment of the present application, in fig. 4, a data processing system 400 includes a robot 410, a training device 420, a database 430, a client device 440, a data storage system 450, and a data acquisition device 460, where the robot 410 includes a calculation module 411, where the calculation module 411 includes a trained neural network 401 obtained by applying the embodiment of the present application, and the neural network 401 is used to perform operations such as feature extraction on acquired interaction force to obtain predicted door-axis parameters
The data acquisition device 460 is configured to acquire an open-source large-scale data set (i.e., a training data set) required by a user, store the training data set in the database 430, and train the neural network 401 in the computing module 411 based on the training data set maintained in the database 430 by the training device 420. The robot 410 may call data, code, etc. in the data storage system 450, or may store data, instructions, etc. in the data storage system 450. The data storage system 450 may be disposed in the robot 410, or the data storage system 450 may be an external memory with respect to the robot 410. It should be noted that in embodiments of the present application, the data acquisition device 460 may be a first force sensor (e.g., a six-dimensional force sensor).
In fig. 4, the robot 410 may be further configured with an I/O interface 412, and may interact with external devices to input data to the I/O interface 412 through a client device 440. For example, the client device 440 may be a second force sensor disposed on the robot, the first interaction force collected by the second force sensor is input as input data to the computing module 411 of the robot 410, and the computing module 411 performs feature extraction on the input first interaction force to obtain a predicted door-axis parameterThe product forms of the robot 410 and the client device 440 are not limited in this application.
It should be noted that fig. 4 is only a schematic diagram of a system architecture provided in the embodiments of the present application, where the positional relationship between devices, apparatuses, modules, etc. shown in the drawings does not constitute any limitation, for example, in fig. 4, the data storage system 450 is an external memory with respect to the robot 410, and in other cases, the data storage system 450 may be disposed in the robot 410; in fig. 4, the client device 440 is an external device with respect to the robot 410, in other cases, the client device 440 may also be integrated into the robot 410.
It should be further noted that, in some embodiments of the present application, the computing module 411 may be further split into a plurality of sub-modules/sub-units to jointly implement the solutions provided by the embodiments of the present application, which are not limited herein.
On the basis of the data processing system corresponding to fig. 4, the following description describes a neural network training method provided in the embodiment of the present application, with reference to fig. 5, where the method specifically includes the following steps:
step 1, acquiring a training data set, wherein the training data set comprises a plurality of training data, and the training data comprises interaction force applied to a training door plate.
First, a training device obtains a training dataset comprising a plurality of training data, each training data comprising an interaction force between a robot and a training door panel.
It should be noted that, in the embodiment of the present application, the acquisition manner of the training data may include, but is not limited to, the following manners:
(1) Training data is acquired in a virtual environment.
In this case, the random door generation module generates the doors with different door axis positions, angles, masses, door axis friction coefficients and the like (i.e. virtual doors) continuously and randomly on the line, and outputs the real door axis parameters of each virtual doorThen, the robot (i.e. virtual robot) performs door opening or closing task on the generated door, and the interaction force of the robot and the door is continuously and periodically collected through the force sensor (i.e. virtual sensor) as training data of the neural network.
The benefits of this acquisition of training data in a virtual environment are: various types of gates can be generated rapidly, a large amount of interaction force is collected as training data based on the gates, and the completeness of a training data set can greatly improve the training precision of the neural network.
(2) Training data is acquired in a real environment.
In this case, the interaction force between the robot and various doors in real life (namely real doors) is collected as training data of a neural network through a real force sensor pair, and the door axis parameter of each real door is recorded
The benefit of this training data acquisition in real environments is that: the training data come from the real environment, and the neural network prediction obtained by training is high in accuracy.
And step 2, training the neural network by using a training data set according to the target loss function to obtain the trained neural network.
Then training the neural network by using a training data set according to the target loss function to obtain a trained neural network, wherein the output of the neural network is a first predicted door-axis parameterAnd a second predicted door axis parameter->First predicted door-axis parameter->First real door axis parameter +. >Correspondingly, the second predicted door-axis parameter +.>Second real door-axis parameter +.>Correspondingly, the first training door plate is one of a plurality of training door plates.
It should be noted that, in some embodiments of the present application, the objective loss function L may also be expressed asn≥1，Can be called the first power of n, is the first predicted door-axis parameter +.>And the first real door spindle parameter->The difference to the power n, the second power n is the second predicted door-axis parameter +.>And a second real door-axis parameter->To the nth power of the difference, n is greater than or equal to 1. As an example, when n=2, the target loss function L can be expressed as the distance of the predicted door-axis parameter from the actual door-axis parameter, i.e. +.>
It should also be noted that, in some embodiments of the present application, there may be a plurality of ways to determine how far the neural network is trained using the objective loss function, and some termination conditions for ending training the neural network are provided below, including but not limited to:
(1) The adjusted target loss function reaches a preset threshold.
After the objective loss function is configured, a threshold (e.g., 0.03) may be set in advance for the objective loss function, and in the process of performing iterative training on the neural network, whether the value of the objective loss function obtained by the current round of training reaches the threshold is judged after each training is finished, if not, the training is continued, and if the preset threshold is reached, the training is terminated, and then the value of the network parameter of the neural network determined by the current round of training is used as the network parameter value of the neural network which is finally trained.
(2) The adjusted target loss function begins to converge.
After the objective loss function is configured, iterative training can be performed on the neural network, if the difference between the value of the objective loss function obtained by the current round of training and the value of the objective loss function obtained by the previous round of training is within a preset range (for example, 0.01), the objective loss function is considered to be converged, the training can be terminated, and then the value of the network parameter of the neural network determined by the current round of training is used as the network parameter value of the neural network which is finally trained.
(3) Training is carried out for a preset number of times.
In this manner, the iteration number (for example, 1000 times) of training the neural network may be preset, after the objective loss function is configured, the neural network may be iteratively trained, after each round of training is finished, the values of the network parameters of the neural network corresponding to the round of training are stored until the iteration number of training reaches the preset number of times, and then the neural network obtained in each round of training is verified by using the test data, and the value of the network parameter with the best performance is selected as the value of the final network parameter of the neural network.
It should be further noted that, in some embodiments of the present application, because the training data collected in the virtual environment is larger, the present application may also first pretrain the neural network by using the training data collected in the virtual environment to obtain a pretrained neural network, and then adjust the pretrained neural network by using the training data collected in the real environment (the amount of the training data required at this time is not large), thereby obtaining a trained neural network that is finally deployed on the robot.
As an example, the following describes a training process of the neural network, taking a neural network simulation training framework (i.e., training the neural network in a virtual environment, such that the neural network obtains the ability to translate interaction information into portal parameters):
referring specifically to fig. 6, first, doors with different door axis positions, angles, masses, and door axis friction coefficients may be continuously and randomly generated by a random door generation module (i.e., virtualQuasi-door) and output real door-axis parametersThe robot (virtual robot) performs door opening or closing task on the generated door, and continuously and periodically collects interaction force [ f ] of the robot and the door through six-dimensional force sensor (virtual sensor) x ,f y ,f z ,T x ,T y ,T z ]As input to the neural network, the output of the neural network is the predicted door-axis parameter +.>Distance from the actual door pivot parameter using the predicted door pivot parameter +.>Network parameters of the neural network, such as a weight matrix, are modified as a target loss function.
203. And adjusting the second interaction force applied to the target door panel according to the first door axis parameter and the second door axis parameter.
At the time of obtaining the first door spindle parameterAnd a second door axis parameter->After that, the robot is according to the first door axis parameter +.>And a second door axis parameter->The interaction force applied to the target door panel is adjusted, which may be referred to as a second interaction force.
Specifically, in some embodiments of the present application, the robot is based on the first door axis parameterAnd a second door axis parameter->The specific manner of adjusting the second interaction force applied to the target door panel may be: firstly, according to the first door-axis parameter +.>Second door axis parameter->Determining the direction of movement of the end effector>Thereafter, according to the direction of movement->And adjusting a second interaction force applied by the end effector to the target door panel.
204. And taking the second interaction force as a new first interaction force, and repeatedly executing the steps 201 to 203 until a preset condition is reached.
It should be noted that, in some embodiments of the present application, the robot may further use the second interaction force as the new first interaction force, and repeatedly execute the steps 201 to 203 until reaching the preset condition, so that the target door panel moves around the door shaft under the action of the second interaction force applied last time, that is, the door shaft parameter is converted into the actual motion of the end effector of the robot, so as to complete the door opening or door closing task.
It should be noted that, in the embodiment of the present application, the case where the preset condition is reached includes, but is not limited to:
(1) The number of times steps 201 to 203 are performed reaches a preset number of times m.
This is the predicted door-axis parameter for the trained neural network outputAs close as possibleTrue door spindle parameters. And outputting predicted door-axis parameters corresponding to each time the first interaction force is input into the trained neural network. The currently derived predicted door spindle parameters in turn adjust the next applied interaction force. And (5) circulating until the predicted times m are reached.
It should be noted here that each time the interaction force is applied and the collection of the interaction force is performed with a fixed period, for example, the collection period may be 1000Hz, so that even if the interaction force applied for the first time is not in the door opening direction, the period of applying the interaction force is short and can be regarded as the interaction force applied instantaneously, so that the time of the interaction force acting on the target door panel is short and the damage of the door panel is not caused.
(2) The component force of the second interaction force applied last time in the non-first direction is smaller than a preset threshold value, and the first direction is the direction perpendicular to the door shaft.
The situation is to ensure that the interaction force between the end effector of the robot and the door is as perpendicular to the door axis as possible, so that the interaction force of the end effector of the robot in the non-door opening/closing direction is smaller than a preset threshold (for example, the interaction force is zero), and the robot can finish the door opening or closing task with the smallest interaction force as possible.
In some embodiments of the present application, if the robot performs a door opening task, the movement direction of the end effector of the robot isWherein (1)>For the first door axis parameter, < >>Is a second axis parameter, so that the end effector of the robot follows +.>Is moved in the direction of the door to finish the door opening task; if the robot executesIs a door closing task, the movement direction of the end effector of the robot is +.>So that the end effector of the robot is along +.>And the door closing task is completed.
In the above embodiment of the application, the detected interaction force of the robot (e.g., the mechanical arm) and the door is converted into the door-axis parameter by using the neural network, and the interaction force is reversely adjusted based on the door-axis parameter, so that the problem that the door-axis is damaged due to the fact that huge step interaction force occurs between the robot and the door in the opening/closing process is avoided.
As shown in fig. 7, a system architecture applied to the method for opening and closing a door of a robot provided in this embodiment of the present application may include a robot 201 including a neural network 202, a force sensor (e.g., a six-dimensional force sensor) 203, an end motion controller 204, and an end effector 205, where the robot 201 interacts with an environment (i.e., a door panel) 206 to complete a task of opening or closing the door. The robot 201 collects interaction force between the current robot 201 and the environment 206 by using the force sensor 203, and takes the collected force as input of the neural network 202, the output of the neural network 202 is a first door axis parameter and a second door axis parameter, the collected interaction force in each current period corresponds to the door axis parameter output once, the output door axis parameter is used for adjusting the next interaction force in turn, the output door axis parameter is circularly reciprocated, and the output door axis parameter is converted into actual motion of the end effector 205 through the end motion controller 204.
Since the method for opening and closing the door of the robot provided in the embodiments of the present application can be used in the fields of home service, smart logistics, etc., a typical application scenario of landing to a product will be described below as a specific example.
The embodiment of the application can be used for a home service robot door. Referring specifically to fig. 8, the home service robot 3001 is composed of a robot arm 3002, a six-dimensional force sensor 3005, and a mobile chassis 3003. By the method described herein, the home service robot 3001 is able to open the door 3004.
The robot 3001 collects interaction force between the mechanical arm 3002 and the door 3004 through the six-dimensional force sensor 3005 at the end of the mechanical arm 3002, and uses six-dimensional force sensor data of the last 50 steps (other steps may be taken, and the 50 steps are only schematic)Inputting trained neural network to obtain door axis parameters +.>The neural network structure used in the embodiment of the present application may be as shown in fig. 9, where the neural network is composed of two parts of a fully connected neural network and a convolutional neural network, the fully connected neural network converts a 300-dimensional input interaction force time sequence into 256-dimensional data through a LeakyReLU layer and a Dropout layer, and then the 256-dimensional data enters the convolutional neural network through the data, the convolutional neural network is a one-dimensional convolutional neural network Conv1D, and parameters (32,32,5,1) of the convolutional neural network are respectively represented by an input channel number, an output channel number, a convolutional kernel size and a movement step length, and finally, the output of the convolutional neural network is converted into a portal parameter through a 96×6 linear neural network.
The training process of the neural network may be implemented in a simulator. Door panels with different parameters can be generated in the simulator, as shown in fig. 10, and accurate values of door shaft parameters in the simulationIt is known that the neural network outputs an estimated value of the door-axis parameter via the input six-dimensional force sensor data +. >By means of the exact value +.>And estimate +.>L of (2) 2 And correcting the neural network parameters by the distance, so that the neural network learns the mapping relation between the interaction force information acquired by the six-dimensional force sensor and the door axis parameters.
In the above embodiment of the present application, a mapping from interaction force to door-axis position and orientation is constructed by using a neural network, and the neural network is trained by simulating doors with randomly generated parameters, so that the network learns the mapping relationship of interaction force to door-axis position and orientation.
It should be understood that the above-described door opening of the home service robot is only one specific scenario where the method for opening and closing a door of the robot provided in the embodiments of the present application is applied, and the method provided in the embodiments of the present application is not limited to the above scenario, and can be applied to any scenario where the door opening and closing is required, which is not repeated herein.
In order to better implement the above-described solutions of the embodiments of the present application on the basis of the above-described embodiments, a related apparatus for implementing the above-described solutions is also provided below. Referring specifically to fig. 11, fig. 11 is a schematic structural diagram of a robot provided in an embodiment of the present application, and the robot 1100 may specifically include: the device comprises an acquisition module 1101, a conversion module 1102 and an adjustment module 1103, wherein the acquisition module 1101 is used for acquiring a first interaction force between a robot and a target door plate; the conversion module 1102 is configured to input the first interaction force into the trained neural network to obtain an output first door axis parameter and a second door axis parameter, where the first door axis parameter is used to represent a door axis position of the target door panel, and the second door axis parameter is used to represent a door axis orientation of the target door panel; the adjusting module 1103 is configured to adjust a second interaction force applied to the target door panel according to the first door axis parameter and the second door axis parameter.
In one possible design, robot 1100 further includes a cycle trigger module 1104 for: and triggering the acquisition module 1101, the conversion module 1102 and the adjustment module 1103 to repeatedly execute the respective steps until a preset condition is reached, so that the target door panel moves around the door shaft under the action of the second interaction force applied last time.
In one possible design, the reaching the preset condition includes: the number of times of executing the above-mentioned process reaches the preset number of times; or, the component force of the last applied second interaction force in the non-first direction is smaller than a preset threshold, and the first direction is the direction perpendicular to the door axis.
In one possible design, the acquisition module 1101 includes: an end effector block 11011 and a force sensing block 11012; wherein the end effector 11011 is configured to apply a first interaction force to the target door panel; the force sensing module 11012 is configured to acquire the first interaction force.
In one possible design, the adjustment module 1103 is specifically configured to: determining a direction of motion of the end effector 11011 based on the first door axis parameter and the second door axis parameter; based on the direction of motion, a second interaction force applied by the end effector 11011 to the target door panel is adjusted.
In one possible design, in the case where the robot 1100 performs a door opening task, the movement direction isWherein (1)>For the first door-axis parameter, +.>Is the second door axis parameter.
In one possible design, in the case where the robot 1100 performs a door closing task, the direction of motion isWherein (1)>For the first door-axis parameter, +.>Is the second door axis parameter.
In one possible design, the force sensing module includes: a six-dimensional force sensor.
In one possible design, the trained neural network is based on a target loss function, and is obtained by training the neural network by using a training data set; the training data set includes a plurality of training data including interactive forces applied to a training door panel; the output of the neural network is a first predicted door axis parameter corresponding to a first real door axis parameter of a first training door panel and a second predicted door axis parameter corresponding to a second real door axis parameter of the first training door panel, the first training door panel being one of the plurality of training door panels.
In one possible design, the objective loss function includes: and the first n power is the n power of the difference between the first predicted door-axis parameter and the first real door-axis parameter, and the second n power is the n power of the difference between the second predicted door-axis parameter and the second real door-axis parameter, wherein n is more than or equal to 1.
Referring to fig. 12, fig. 12 is another schematic structural diagram of a robot provided in this embodiment of the present application, on which each module of the robot 1100 described in the corresponding embodiment of fig. 11 may be disposed, for implementing the functions of the robot 1100 in the corresponding embodiment of fig. 11, specifically, the robot 1200 is implemented by one or more servers, where the robot 1200 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (central processing units, CPU) 1222 and a memory 1232, and one or more storage media 1230 (such as one or more mass storage devices) storing application programs 1242 or data 1244. Wherein memory 1232 and storage medium 1230 can be transitory or persistent. The program stored on the storage medium 1230 may include one or more modules (not shown), each of which may include a series of instruction operations in the robot 1200. Still further, the central processor 1222 may be configured to communicate with the storage medium 1230, executing a series of instruction operations on the robot 1200 in the storage medium 1230.
The robot 1200 may also include one or more power supplies 1226, one or more wired or wireless network interfaces 1250, one or more input/output interfaces 1258, and/or one or more operating systems 1241, such as Windows Server, mac OS XTM, unixTM, linuxTM, freeBSDTM, and the like.
In the embodiment of the application, the central processor 1222 is configured to execute the cover pulling method executed by the robot in the corresponding embodiment of fig. 1. For example, a central processor 1222 for: first, an interaction force between the robot and the target door panel, which may be referred to as a first interaction force, is acquired. Then, the acquired first interaction force is input into a trained neural network to obtain two output door-axis parameters which are respectively recorded as first door-axis parametersAnd a second door axis parameter->Wherein the first door spindle parameter->For characterizing the door axis position of the target door panel, and the second door axis parameter for characterizing the door axis orientation of the target door panel>These two door-axis parameters can be noted +.>Finally, according to the first door-axis parameter +.>And a second door axis parameter->The interaction force applied to the target door panel is adjusted, which may be referred to as a second interaction force.
It should be noted that, the specific manner in which the cpu 1222 executes the above steps is based on the same concept as that of the method embodiment corresponding to fig. 2 in the present application, and the technical effects thereof are the same as those of the embodiment in the present application, and the specific details can be found in the description of the method embodiment shown in the foregoing application, which is not repeated herein.
It should be further noted that, as shown in fig. 13, the typical configuration scenario of the operation of the present invention may be that the force sensor and the trained neural network interact with each other through a bus and a processor, and the result of the operation of the processor is fed back to the corresponding system. The force sensor is responsible for collecting interaction force between the robot and the environment (namely the door plate), and the end effector is responsible for carrying out force interaction with the environment and can be composed of mechanical structures such as a mechanical arm and the like.
There is also provided in an embodiment of the present application a computer-readable storage medium having stored therein a program for performing signal processing, which when run on a computer, causes the computer to perform the steps described in the embodiment shown in fig. 2.
The robot and the like provided in the embodiment of the application may specifically include a chip, where the chip includes: a processing unit, which may be, for example, a processor, and a communication unit, which may be, for example, an input/output interface, pins or circuitry, etc. The processing unit may execute the computer-executable instructions stored in the storage unit to cause the chip within the robot to perform the steps described in the embodiment shown in fig. 2 above.
Optionally, the storage unit is a storage unit in the chip, such as a register, a cache, etc., and the storage unit may also be a storage unit in the wireless access device side located outside the chip, such as a read-only memory (ROM) or other type of static storage device that may store static information and instructions, a random access memory (random access memory, RAM), etc.
It should be further noted that the above-described apparatus embodiments are merely illustrative, and that the units described as separate units may or may not be physically separate, and that units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the drawings of the embodiment of the device provided by the application, the connection relation between the modules represents that the modules have communication connection therebetween, and can be specifically implemented as one or more communication buses or signal lines.
From the above description of the embodiments, it will be apparent to those skilled in the art that the present application may be implemented by means of software plus necessary general purpose hardware, or of course may be implemented by dedicated hardware including application specific integrated circuits, dedicated CPUs, dedicated memories, dedicated components and the like. Generally, functions performed by computer programs can be easily implemented by corresponding hardware, and specific hardware structures for implementing the same functions can be varied, such as analog circuits, digital circuits, or dedicated circuits. However, a software program implementation is a preferred embodiment in many cases for the present application. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a readable storage medium, such as a floppy disk, a usb disk, a removable hard disk, a ROM, a RAM, a magnetic disk or an optical disk of a computer, etc., including several instructions for causing a computer device (which may be a personal computer or a network device, etc.) to execute the method described in the embodiments of the present application.
In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product.
The computer program product includes one or more computer instructions. When loaded and executed on a computer, produces a flow or function in accordance with embodiments of the present application, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website, computer, or data center to another website, computer, or data center by a wired (e.g., coaxial cable, fiber optic, digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer readable storage medium may be any available medium that can be stored by a computer or a data storage device such as a data center that contains an integration of one or more available media. The usable medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a Solid State Disk (SSD)), or the like.
Claims (24)
1. A method for opening and closing a door of a robot, comprising:
acquiring a first interaction force between the robot and a target door plate;
inputting the first interaction force into the trained neural network to obtain a first door axis parameter and a second door axis parameter which are output, wherein the first door axis parameter is used for representing the door axis position of the target door plate, and the second door axis parameter is used for representing the door axis orientation of the target door plate;
and adjusting a second interaction force applied to the target door panel according to the first door axis parameter and the second door axis parameter.
2. The method according to claim 1, wherein the method further comprises:
and taking the second interaction force as a new first interaction force, and repeatedly executing the process until a preset condition is reached, so that the target door plate moves around a door shaft under the action of the second interaction force applied last time.
3. The method of claim 2, wherein the reaching a preset condition comprises:
the times of executing the process reach the preset times;
or alternatively, the first and second heat exchangers may be,
the component force of the second interaction force applied last time in the non-first direction is smaller than a preset threshold value, and the first direction is the direction perpendicular to the door shaft.
4. The method of any of claims 1-3, wherein the robot includes an end effector and a force sensor thereon, and wherein the acquiring a first interaction force between the robot and the target door panel includes:
applying a first interactive force to the target door panel via the end effector;
and acquiring the first interaction force through the force sensor.
5. The method of claim 4, wherein the adjusting the second interaction force applied to the target door panel according to the first door axis parameter and the second door axis parameter comprises:
determining a direction of motion of the end effector from the first door axis parameter and the second door axis parameter;
and adjusting a second interaction force applied by the end effector to the target door plate according to the movement direction.
6. The method of claim 5, wherein in the event that the robot performs a door opening task, the direction of movement isWherein (1)>For the first door spindle parameter, +.>Is the second door axis parameter.
7. The method according to claim 5, wherein in case the robot performs a door closing task, the movement direction is Wherein (1)>For the first door spindle parameter, +.>Is the second door axis parameter.
8. The method of any one of claims 4-7, wherein the force sensor comprises:
a six-dimensional force sensor.
9. The method according to any one of claims 1-8, wherein the trained neural network is obtained by training the neural network with a training dataset based on a target loss function;
the training data set includes a plurality of training data including interactive forces applied to a training door panel;
the output of the neural network is a first predicted door axis parameter and a second predicted door axis parameter, the first predicted door axis parameter corresponds to a first real door axis parameter of a first training door panel, the second predicted door axis parameter corresponds to a second real door axis parameter of the first training door panel, and the first training door panel is one of a plurality of training door panels.
10. The method of claim 9, wherein the objective loss function comprises:
and the first n power is the n power of the difference between the first predicted door-axis parameter and the first real door-axis parameter, and the second n power is the n power of the difference between the second predicted door-axis parameter and the second real door-axis parameter, wherein n is more than or equal to 1.
11. A robot, comprising:
the acquisition module is used for acquiring a first interaction force between the robot and the target door plate;
the conversion module is used for inputting the first interaction force into the trained neural network to obtain a first door axis parameter and a second door axis parameter which are output, wherein the first door axis parameter is used for representing the door axis position of the target door plate, and the second door axis parameter is used for representing the door axis orientation of the target door plate;
and the adjusting module is used for adjusting the second interaction force applied to the target door plate according to the first door shaft parameter and the second door shaft parameter.
12. The robot of claim 11, further comprising a cycle trigger module for:
and taking the second interaction force as a new first interaction force, triggering the acquisition module, the conversion module and the adjustment module to repeatedly execute the respective steps until a preset condition is reached, so that the target door plate moves around a door shaft under the action of the second interaction force applied last time.
13. The robot of claim 12, wherein the reaching the preset condition comprises:
The times of executing the process reach the preset times;
or alternatively, the first and second heat exchangers may be,
the component force of the second interaction force applied last time in the non-first direction is smaller than a preset threshold value, and the first direction is the direction perpendicular to the door shaft.
14. The robot of any one of claims 11-13, wherein the acquisition module comprises:
an end effector module and a force sensing module;
the end execution module is used for applying a first interaction force to the target door plate;
the force sensing module is used for acquiring the first interaction force.
15. The robot of claim 14, wherein the adjustment module is specifically configured to:
determining a movement direction of the end effector module according to the first door axis parameter and the second door axis parameter;
and adjusting a second interaction force applied by the end execution module to the target door plate according to the movement direction.
16. The robot of claim 15, wherein in the event that the robot performs a door opening task, the direction of movement isWherein (1)>For the first door spindle parameter, +.>Is the second door axis parameter.
17. The robot of claim 15, wherein in the event that the robot performs a door closing task, the direction of motion isWherein (1)>For the first door spindle parameter, +.>Is the second door axis parameter.
18. The robot of any one of claims 14-17, wherein the force sensing module comprises:
a six-dimensional force sensor.
19. The robot of any one of claims 11-18, wherein the trained neural network is trained using a training dataset based on a target loss function;
the training data set includes a plurality of training data including interactive forces applied to a training door panel;
the output of the neural network is a first predicted door axis parameter and a second predicted door axis parameter, the first predicted door axis parameter corresponds to a first real door axis parameter of a first training door panel, the second predicted door axis parameter corresponds to a second real door axis parameter of the first training door panel, and the first training door panel is one of a plurality of training door panels.
20. The robot of claim 19, wherein the objective loss function comprises:
And the first n power is the n power of the difference between the first predicted door-axis parameter and the first real door-axis parameter, and the second n power is the n power of the difference between the second predicted door-axis parameter and the second real door-axis parameter, wherein n is more than or equal to 1.
21. A robot comprising a processor and a memory, said processor being coupled to said memory, characterized in that,
the memory is used for storing programs;
the processor for executing a program in the memory, causing the robot to perform the method of any one of claims 1-10.
22. A computer readable storage medium comprising a program which, when run on a computer, causes the computer to perform the method of any of claims 1-10.
23. A computer program product containing instructions which, when run on a computer, cause the computer to perform the method of any of claims 1-10.
24. A chip comprising a processor and a data interface, the processor reading instructions stored on a memory via the data interface, performing the method of any of claims 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210780246.XA CN117400235A (en) | 2022-07-04 | 2022-07-04 | A method for opening and closing a door by a robot and the robot |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210780246.XA CN117400235A (en) | 2022-07-04 | 2022-07-04 | A method for opening and closing a door by a robot and the robot |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117400235A true CN117400235A (en) | 2024-01-16 |
Family
ID=89485770
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210780246.XA Pending CN117400235A (en) | 2022-07-04 | 2022-07-04 | A method for opening and closing a door by a robot and the robot |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117400235A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118848997A (en) * | 2024-09-26 | 2024-10-29 | 国网(山东)电动汽车服务有限公司 | A method for accurately controlling the opening and closing of the charging cover by a robotic arm based on deep reinforcement learning |
-
2022
- 2022-07-04 CN CN202210780246.XA patent/CN117400235A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118848997A (en) * | 2024-09-26 | 2024-10-29 | 国网(山东)电动汽车服务有限公司 | A method for accurately controlling the opening and closing of the charging cover by a robotic arm based on deep reinforcement learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Su et al. | An incremental learning framework for human-like redundancy optimization of anthropomorphic manipulators | |
| JP7301034B2 (en) | System and Method for Policy Optimization Using Quasi-Newton Trust Region Method | |
| EP4384359A1 (en) | A robot system for anomaly detection | |
| JP7375587B2 (en) | Trajectory generation device, multi-link system, and trajectory generation method | |
| CN120917450A (en) | Reduced order modeling and control of high-dimensional physical systems using neural network models | |
| CN116079747A (en) | Robot cross-body control method, system, computer equipment and storage medium | |
| CN119238499B (en) | A robot control method, device, electronic device, and storage medium | |
| Jia et al. | Improved dynamic parameter identification method relying on proprioception for manipulators | |
| Limtrakul et al. | Supervised learning based on the self-organizing maps for forward kinematic modeling of Stewart platform | |
| Tanwani et al. | Generalizing robot imitation learning with invariant hidden semi-Markov models | |
| Patel | Robotic hand-eye coordination fusion | |
| CN117400235A (en) | A method for opening and closing a door by a robot and the robot | |
| Scholz et al. | Learning non-holonomic object models for mobile manipulation | |
| CN117609673B (en) | Forward solution method of six-degree-of-freedom parallel mechanism based on physical information neural network | |
| Su et al. | Incremental learning with balanced update on receptive fields for multi-sensor data fusion | |
| Fananás-Anaya et al. | Dynamical system simulation with attention and recurrent neural networks | |
| US20240051124A1 (en) | Simulation of robotics devices using a neural network systems and methods | |
| US12541962B2 (en) | Training autoencoders for generating latent representations | |
| Chen et al. | Open multimode process monitoring: a dual memory-based continual dictionary learning method | |
| Xiong et al. | Parameter identification method for collaborative robot joint using a cuckoo search-bp neural network approach | |
| Huang et al. | A Gradient‐Based Recurrent Neural Network for Visual Servoing of Robot Manipulators with Acceleration Command | |
| WO2023057518A1 (en) | Demonstration-driven reinforcement learning | |
| JPWO2018084164A1 (en) | Motion transition apparatus, motion transition method, and non-transitory computer-readable medium storing motion transition program | |
| Younes et al. | Toward faster reinforcement learning for robotics: using Gaussian processes | |
| Yassin et al. | Comparison between NARX parameter estimation methods with Binary Particle Swarm Optimization-based structure selection method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |