Semantic positioning method, device and robot based on probabilistic ontology

Abstract

The present invention provides a semantic positioning method, device and robot based on probabilistic ontology, the method comprising: based on the position context information of the robot's operating object and human common sense, performing probabilistic ontology knowledge expression and constructing a multi-entity Bayesian network model; based on the entity Bayesian network model, performing multiple instantiations, combining into a target situation Bayesian network, and inferring a first probability value; in the case where the robot has target historical task data, inferring a second probability value based on the target historical task data and the target situation Bayesian network; based on the first probability value and the second probability value, using a weight factor for calculation, obtaining the positioning result of the robot's operating object, so that the robot performs the task according to the positioning result of the operating object and records the historical task data. The semantic positioning method, device and robot based on probabilistic ontology provided by the present invention can improve the accuracy of semantic positioning.

Description

Technical Field

[0001] The present invention relates to the technical field of location services, and in particular to a semantic positioning method, device and robot based on probabilistic ontology. Background Art

[0002] When performing daily tasks, intelligent service robots first need to have the ability to obtain the target objects in the task, so searching for specific objects is a key part of the service robot to complete daily tasks. Benefiting from the rapid development of sensor technology and perception algorithms, the perception ability of robots has been significantly improved in recent years. However, the placement and status of objects in the human living environment are messy. If robots can assist in searching with common sense knowledge and accumulated experience like humans, it will greatly reduce the search space of object detection and recognition algorithms.

[0003] Ontology is a model that effectively represents the concept hierarchy and semantics, and has become an important tool for robots to utilize knowledge. Tenorth et al. proposed a knowledge processing system, KnowRob, which uses ontology as a knowledge carrier and stores the semantic location of objects by defining object classes and location attributes. KnowRob also supports abstract rule definition, combining the attribute knowledge of objects with semantic environment maps to achieve object location reasoning. Schuster et al. proposed a semantic similarity model based on ontology. If the robot knows the location of some objects, it can obtain the location of new objects based on the semantic similarity with known objects.

[0004] It can be seen that both methods in the prior art use ontology as the basis for knowledge representation and reasoning, which can provide robots with rich semantic information about object positions. However, due to the limitation of network ontology language based on logical description, it can only represent the deterministic position information of robots. In actual applications, there are many factors that affect the position of objects. There is always uncertainty in the position of objects operated by service robots, so the semantic positioning accuracy is low. Summary of the invention

[0005] The present invention provides a semantic positioning method, device and robot based on probabilistic ontology, which are used to solve the defect of low semantic positioning accuracy in the prior art, realize semantic positioning by combining human common sense and the reference value of historical tasks, and improve the semantic positioning accuracy.

[0006] The present invention provides a semantic positioning method based on probabilistic ontology, comprising:

[0007] Based on the position context information of the robot's operating objects and human common sense, probabilistic ontology knowledge expression is performed to build a multi-entity Bayesian network model;

[0008] Based on the entity Bayesian network model, multiple instantiations are performed to combine into a target situation Bayesian network, and a first probability value is inferred;

[0009] In the case where the robot has target historical task data, inferring a second probability value based on the target historical task data and the target situation Bayesian network;

[0010] Based on the first probability value and the second probability value, a weight factor is used for calculation to obtain a positioning result of the operation object of the robot, so that the robot performs a task according to the positioning result of the operation object and records historical task data;

[0011] The target historical task data has a corresponding relationship with the position context information of the operation object of the robot, and the position context information includes one or more of object type, temperature condition and user preference.

[0012] According to a semantic positioning method based on probabilistic ontology provided by the present invention, after the first probability value is inferred, it also includes: when the robot does not have target historical task data, using the first probability value as the positioning result of the robot's operation object, so that the robot performs the task according to the positioning result of the operation object and records the historical task data.

[0013] According to a semantic positioning method based on probabilistic ontology provided by the present invention, the probabilistic ontology knowledge expression is performed based on the position context information of the robot's operation object and human common sense, and a multi-entity Bayesian network model is constructed, including:

[0014] Construct a probabilistic ontology model based on the position context information of the robot's operating object;

[0015] Design corresponding local probabilities for each target element in the probabilistic ontology model according to the human common sense, generate entity fragments, and merge the entity fragments into the multi-entity Bayesian network model;

[0016] The target elements include at least the storage location of the operation object, the type of the operation object, the area of ​​the operation object and user information.

[0017] According to a semantic positioning method based on probability ontology provided by the present invention, the second probability value is inferred based on the target historical task data and the target situation Bayesian network, including:

[0018] Based on the target historical task data, using Dirichlet distribution, calculate the probability distribution of each node in the target situation Bayesian network, and update the target situation Bayesian network;

[0019] Based on the updated target situation Bayesian network, a standard Bayesian inference algorithm is used to obtain the second probability value.

[0020] According to a semantic positioning method based on probabilistic ontology provided by the present invention, the calculation formula of the weight factor is:

[0021]

[0022] Among them, α t+1 is the weight factor at the t+1th moment, α t is the weight factor at the tth moment, n t+1 is the number of target historical task data at the t+1th moment, n t is the number of the target historical task data at the tth moment, d is the weight factor attenuation coefficient, and m is the attenuation speed.

[0023] The present invention also provides a semantic positioning device based on probabilistic ontology, comprising:

[0024] The model building module is used to express probabilistic ontology knowledge and build a multi-entity Bayesian network model based on the position context information of the robot's operating objects and human common sense;

[0025] A first probability acquisition module is used to perform multiple instantiations based on the entity Bayesian network model, combine into a target situation Bayesian network, and infer a first probability value;

[0026] A second probability acquisition module is used to infer a second probability value based on the target historical task data and the target situation Bayesian network when the robot has target historical task data;

[0027] A first semantic positioning module, configured to obtain a positioning result of an operation object of the robot by using a weight factor for calculation based on the first probability value and the second probability value, so that the robot performs a task according to the positioning result of the operation object and records historical task data;

[0028] The target historical task data has a corresponding relationship with the position context information of the operation object of the robot, and the position context information includes one or more of object type, temperature condition and user preference.

[0029] The present invention also provides a robot, comprising the semantic positioning device based on probabilistic ontology knowledge as described above.

[0030] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-mentioned semantic positioning methods based on probabilistic ontology are implemented.

[0031] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described semantic positioning methods based on probabilistic ontology.

[0032] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of any of the above-mentioned semantic positioning methods based on probabilistic ontology are implemented.

[0033] The semantic positioning method, device and robot based on probabilistic ontology provided by the present invention are based on probabilistic ontology knowledge expression, and the uncertainty of the object storage location is standardizedly described and modeled, and a multi-entity Bayesian network model is obtained. Through multiple instantiations and generalizations, a target situation Bayesian network is formed, and human common sense and target historical tasks are used as samples respectively. The first probability value and the second probability value are inferred in the target situation Bayesian network, and the first probability value and the second probability value are dynamically fused through the weight factor to obtain the positioning result and assist the robot in recognizing the operation object in semantic positioning. It can combine the reference value of human common sense and historical tasks to improve the accuracy of semantic positioning, and then make decisions to guide the robot to perform tasks on the operation object in an unstructured human living environment, thereby improving the efficiency of task execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 It is a flow chart of the semantic positioning method based on probabilistic ontology provided by the present invention;

[0036] Figure 2 It is a schematic diagram of the structure of the target situation Bayesian network provided by the present invention;

[0037] Figure 3 It is a structural schematic diagram of the probabilistic ontology model provided by the present invention;

[0038] Figure 4 It is a structural schematic diagram of a semantic positioning device based on probabilistic ontology provided by the present invention;

[0039] Figure 5 It is a structural schematic diagram of the robot provided by the present invention;

[0040] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] Figure 1 FIG. 1 is a flow chart of the semantic positioning method based on probabilistic ontology provided by the present invention. Figure 1 As shown, the semantic positioning method based on probabilistic ontology provided by an embodiment of the present invention includes: step 101, constructing a multi-entity Bayesian network model based on the position context information of the robot's operation object and human common sense.

[0043] The location context information includes one or more of object type, temperature condition and user preference.

[0044] It should be noted that the execution subject of the semantic localization method based on probabilistic ontology provided in the embodiment of the present invention is a semantic localization device based on probabilistic ontology. The semantic localization device based on probabilistic ontology can be integrated into an independent chip or processor, or can be used as a part of the chip or processor of the robot on which it is running, for executing the semantic localization method based on probabilistic ontology.

[0045] Optionally, the application scenario of the semantic positioning method based on probabilistic ontology can be that when the robot is executing a task and the operation object is not within the detection range of the sensor, the storage location of the operation object can be inferred through relevant information about the task of the operation object and the semantic relationship of the object position in the current environment, so as to provide a decision for the robot to find the operation object.

[0046] Optionally, the application scenario of the semantic positioning method based on probabilistic ontology can also be that the intelligent service robot can obtain the feature information of objects in the current environment through the acquisition device and identify it. After confirming the type of object, it determines it as the operation object, and judges the original storage location of the object through the semantic relationship of the object position in the current environment, and performs the task of placing the object back to its original position through the stored location information.

[0047] Before step 101, the user or manager of the semantic positioning device based on probabilistic ontology knowledge needs to extract the location context information of the robot's operation object according to the robot's service scenario.

[0048] The location context information of the operation object refers to the factor information that affects the original storage location of the object. The embodiment of the present invention does not specifically limit this.

[0049] Optionally, the location context information of the operation object may be an object type, and different types of objects correspond to different storage locations. For example, frozen food is usually placed in a refrigerator, and condiments are usually placed in a cupboard.

[0050] Optionally, the location context information of the operation object may be a temperature condition. Different temperature conditions in the area where the operation object is located may also affect the positioning of the storage location. For example, if the operation object is a frozen food, if the temperature in its current area is high or normal temperature, that is, it does not meet the temperature condition corresponding to the item, then the storage location is positioned as a place with low temperature conditions. If the temperature in its current area is low temperature, which meets the temperature condition corresponding to the item, then the storage location may not need to be changed.

[0051] Optionally, the location context information of the operation object may be a user preference. The preferences of the user to whom the operation object belongs are different, which may also affect the location of the storage location. For example, if the operation object is a drink, the original storage location of the item may be a dining table or a refrigerator according to different user preferences.

[0052] Human common sense refers to the quantitative description of the various intrinsic connections between operation objects and their corresponding storage locations based on human behavioral habits.

[0053] The embodiments of the present invention do not specifically limit the description of human common sense.

[0054] For example, human common sense may be to quantify the intrinsic relationship between the operation object, the storage location, and the context information of each location into a probability value.

[0055] For example, in human common sense, based on the intrinsic connection between object type and storage location, it is generally believed that the probability that food needs to be placed on the dining table is 0.2, the probability of being stored in the refrigerator is 0.6, and the probability of being stored in other locations is 1-0.2-0.6=0.2.

[0056] For another example, in human common sense, there is a certain intrinsic connection between object type and user preference. For beverages, it is generally determined that the probability that the user tends to refrigerate them is 0.6, while the probability of not specifying a preference condition is 1-0.6=0.4.

[0057] Probabilistic ontology is a standardized knowledge representation used to characterize knowledge related to application domains. It includes entity types, entity attributes, relationships between entities, events that occur with entities, and knowledge uncertainty. In view of the limitations of Web Ontology Language (OWL) in expressing uncertain information, Probabilistic Web Ontology Language (PR-OWL) is selected to describe the probabilistic ontology model of the semantic location of robot-operated objects.

[0058] Multi-entity Bayesian networks, as the underlying logical basis of PR-OWL, are a first-order language that specifies probabilistic knowledge bases as parameterized fragments of Bayesian networks. It unifies ordinary Bayesian networks with classical first-order logic.

[0059] Specifically, in step 101, in a semantic positioning device based on probabilistic ontology, probabilistic ontology knowledge expression is performed according to the robot's operating object, storage location, and contextual information that affects the storage location of the object. Each entity node in the ontology knowledge is set in the form of Bayesian probability according to human common sense, and a multi-entity Bayesian network model is constructed.

[0060] The multi-entity Bayesian network model is based on human common sense, defines the semantic relationship between objects, storage locations, and location context information, and combines the probabilities of various semantic relationships with the first-order logic of the corresponding semantic relationships. The multi-entity Bayesian network model is used to provide a logically consistent representation of uncertainty factors and to reason about semantic locations under uncertainty.

[0061] Step 102: Based on the entity Bayesian network model, multiple instantiations are performed to combine into a target situation Bayesian network, and a first probability value is inferred.

[0062] Specifically, in step 102, the multi-entity Bayesian network model constructed in step 101 is instantiated multiple times for each Bayesian network corresponding to the entity, and combined according to the principle of the probabilistic graphical model to generate a target situation Bayesian network. The network is then combined with evidence (i.e., prior knowledge) and inferred using a standard Bayesian inference algorithm to obtain a first probability value.

[0063] The first probability value refers to the probability value of the operation object being in each storage location after the judgment and processing in step 102, and is the robot's prediction value of the location where the operation object should be placed based on human common sense.

[0064] The embodiment of the present invention does not specifically limit the construction method of the target situation Bayesian network.

[0065] Preferably, the target situation Bayesian network can be constructed through the bottom-up method proposed by Mahoney and Laskey.

[0066] For example, Figure 2 Schematic diagram of the target situation Bayesian network provided by the present invention. Figure 2 As shown in the figure, the target situation Bayesian network contains the user's gender, age group and preference information, the type of operation object and the ambient temperature condition information. These information nodes are used as factors affecting the final position of the operation object. Using the structure of the target situation Bayesian network and the conditional probability table of each node, the standard Bayesian algorithm is used for reasoning after given evidence to obtain the position probability distribution.

[0067] The standard Bayesian inference algorithm used in the embodiment of the present invention includes, but is not limited to, a variable elimination algorithm (VE), a junction tree algorithm (JT), a Monte Carlo algorithm, or a variational method (Variational Methods) algorithm.

[0068] Step 103: When the robot has target historical task data, a second probability value is inferred based on the target historical task data and the target situation Bayesian network.

[0069] Among them, the target historical task data has a corresponding relationship with the position context information of the robot's operation object.

[0070] It should be noted that, before step 103, the semantic positioning device based on probabilistic ontology knowledge needs to obtain historical task data from a local database set inside the robot.

[0071] Among them, if the historical task data is empty, it means that the semantic positioning task performed on the current operation object is the first task in the life cycle or usage cycle of the robot, and there is no historical data for reference for this semantic positioning task. Therefore, the storage location corresponding to the first probability value is used as the target of this task.

[0072] If the historical task data is not empty, it means that there is historical data available for reference for this semantic positioning task. In order to ensure the consistency of this semantic positioning task, the historical task data is filtered according to the known information in this semantic positioning task. There are two types of filtering results: successful filtering and failed filtering, where:

[0073] Failure in screening means that the historical task data does not have data that is consistent with the current semantic positioning task. Therefore, the historical task data has little reference value for this task.

[0074] The screening is successful, which means that the historical task data contains target historical task data that is consistent with the current semantic positioning task.

[0075] The target historical task data is consistent with one or more element information in the position context information of the robot for the operation object this time. Among them, the more element information consistent with the position context information, the closer the target historical task data that can be extracted is to the semantic positioning task performed by the robot for the operation object this time, and the greater the reference value of the target historical task data for this semantic positioning task.

[0076] Specifically, in step 103, when the target historical task data is screened out, the historical task execution data is used as a Bayesian parameter estimation sample, and the structure of the target situation Bayesian network is used for reasoning to calculate the second probability value.

[0077] The first probability value refers to the probability value of the operation object being in each storage location after the judgment and processing in step 103, and is the robot's prediction value of the location where the operation object should be placed based on consistent historical task data.

[0078] Step 104: Based on the first probability value and the second probability value, a weight factor is used for calculation to obtain a positioning result of the robot's operation object, so that the robot performs the task according to the positioning result of the operation object and records historical task data.

[0079] The weight factor refers to the weight value corresponding to the first probability value or the second probability value. Since two variables are involved in the calculation, the sum of the weight factor corresponding to the first probability value and the weight factor corresponding to the second probability value is 1.

[0080] Preferably, the weight factor is the weight value corresponding to the first probability value, and the weight value corresponding to the second probability value is the difference between 1 and the weight factor.

[0081] Specifically, in step 104, the semantic positioning device based on probabilistic ontology knowledge calculates a weighted sum of the first probability value and the second probability value in combination with their corresponding weight factors to obtain the robot's positioning result of the operation object this time, and sends the positioning result to the robot's control system, so that the robot's control system, after receiving the positioning result, establishes a task for the purpose of carrying the operation object to the placement position indicated by the positioning result, controls the various components of the robot to perform the task, and records the operation object and the corresponding positioning result as historical task data in the local database after the task is completed.

[0082] The positioning result refers to the probability calculation value obtained by weighted summation of the first probability value and the second probability value. The storage location corresponding to the probability value is the target location obtained by the robot for positioning the operation object this time.

[0083] The process of obtaining the positioning result depends on the data content of the first probability value and the second probability value, and the embodiment of the present invention does not specifically limit the process of obtaining the positioning result.

[0084] Optionally, the first probability value and the second probability value may be a set of probability values, which are respectively probability values ​​[X 11 ,X 21 ,…,X i1 ], and the probability value of the operation object inferred based on the historical task in each storage location [X 12 ,X 22 ,…,X i2 ]. Then the probability calculation value of the operation object in each storage location [Y1,Y2,…,Y i ] can be expressed as:

[0085] a 11 *X 11 +a 12 *X 12 =Y1

[0086] a 21 *X 21 +a 22 *X 22 =Y2

[0087]

[0088] a i1 *X i1 +a i2 *X i2 =Y i

[0089] Among them, [a 11 ,a 21 ,…,a i1 ] is the weight factor corresponding to the first probability value, [a 12 ,a 22 ,…,a i2 ] is the weight factor corresponding to the second probability value, and i is the number of child nodes under the storage location.

[0090] The embodiment of the present invention is based on the probabilistic ontology knowledge expression, and performs a standardized description and modeling of the uncertainty of the object storage location, obtains a multi-entity Bayesian network model, and combines it into a target situation Bayesian network through multiple instantiations and generalizations. Human common sense and target historical tasks are used as samples, and the first probability value and the second probability value are inferred in the target situation Bayesian network. The first probability value and the second probability value are dynamically fused through the weight factor to obtain the positioning result and assist the robot in recognizing the semantic positioning of the operation object. It can combine the reference value of human common sense and historical tasks to improve the accuracy of semantic positioning, and then make decisions to guide the robot to perform tasks on the operation object in an unstructured human living environment, thereby improving the efficiency of task execution.

[0091] On the basis of any of the above embodiments, after inferring the first probability value, it also includes: when the robot does not have target historical task data, using the first probability value as the positioning result of the robot's operation object, so that the robot performs the task according to the positioning result of the operation object and records the historical task data.

[0092] Specifically, in the absence of historical task data or in the absence of target historical task data, the semantic positioning device based on probabilistic ontology knowledge determines the positioning result of the robot's operating object according to the first probability value, and sends the positioning result to the robot's control system, so that the robot's control system, after receiving the positioning result, establishes a task for the purpose of carrying the operating object to the placement position indicated by the positioning result, and controls the various components of the robot to perform the task, and after the task is completed, the operating object and the corresponding positioning result are recorded as historical task data in the local database.

[0093] The positioning result refers to the actual storage location of the operation object.

[0094] It can be understood that, in this case, the weight factor corresponding to the first probability value is 1, and the weight factor corresponding to the second probability value is 0.

[0095] In the absence of a target historical task, the embodiment of the present invention uses human common sense and target historical tasks as samples, infers a first probability value in the target situation Bayesian network as a positioning result, and assists the robot in recognizing the semantic positioning of the operation object. It can combine human common sense to express probabilistic ontology knowledge, improve the accuracy of semantic positioning, and then make decisions to guide the robot to perform tasks on the operation object in an unstructured human living environment, thereby improving the efficiency of task execution.

[0096] On the basis of any of the above embodiments, based on the position context information of the robot's operating object and human common sense, probabilistic ontology knowledge expression is performed to construct a multi-entity Bayesian network model, including: constructing a probabilistic ontology model based on the position context information of the robot's operating object.

[0097] Specifically, in step 101, the semantic positioning device based on probabilistic ontology knowledge extracts the context information of the robot operation object location (such as the type of object, temperature conditions and user preferences) according to the robot service scenario, and constructs a probabilistic ontology model using the probabilistic network ontology language to form an ontology knowledge expression of the object storage location.

[0098] The embodiment of the present invention does not specifically limit the probabilistic ontology model.

[0099] For example, Figure 3 Schematic diagram of the structure of the probabilistic ontology model provided by the present invention. Figure 3 As shown in Figure 2, taking the location of edible objects in a human home environment as an example, the semantic localization of the robot's manipulation objects depends on several elements of available contextual information, such as the type of object, temperature conditions, and user preferences. Figure 3 The probabilistic ontology representation contains related entities, entity attributes and relationships between entities. It is formed based on OWL ontology through concept mapping, attribute mapping, constraint mapping and probability information.

[0100] The "operation object" node may be classified according to the item type and divided into multiple sub-nodes, including but not limited to drinks, food and other items.

[0101] The "storage location" node and the "operation object" node have a first-order logical relationship of "stored in". The "storage location" node can be classified according to space and divided into multiple child nodes, including but not limited to refrigerators, dining tables, cabinets and other locations.

[0102] The "region" node and the "operation object" node have a first-order logical relationship of "region". The "region" node can be classified according to temperature conditions and divided into multiple sub-nodes, including but not limited to high temperature, low temperature and normal temperature.

[0103] The "user" node and the "operation object" node have a first-order logical relationship of "owning". The "region" node can be classified according to the requirements and divided into multiple sub-nodes, including but not limited to age group, gender and preference. Among them, the "age group" sub-node can be refined into second-order sub-nodes including but not limited to youth, middle-aged and elderly, the "gender" sub-node can be refined into second-order sub-nodes including male and female, and the "preference" sub-node can be refined into second-order sub-nodes including but not limited to refrigeration and no requirements.

[0104] For each target element in the probabilistic ontology model, the corresponding local probability is designed according to human common sense to generate entity fragments, and then the entity fragments are merged into a multi-entity Bayesian network model.

[0105] The target elements include at least the storage location of the operation object, the type of the operation object, the area of ​​the operation object and user information.

[0106] It should be noted that the target element refers to a node with a high correlation in the probabilistic ontology model in which the location context information of the operation object is located.

[0107] Optionally, the target element may be a storage location of the operation object, that is, a “storage location” node having a “stored in” entity relationship with an “operation object” node.

[0108] Optionally, the target element may be the type of the operation object, ie, the "operation object" node itself.

[0109] Optionally, the target element may be the region of the operation object, that is, an "region" node having an entity relationship of "region" with the "operation object" node.

[0110] Optionally, the target element may be user information. The embodiment of the present invention does not specifically limit the user information. The user information includes but is not limited to the "user" node, and the "age group" sub-node, "gender" sub-node and "preference" sub-node under the node.

[0111] Specifically, each target element in the probabilistic ontology model is taken as a separate entity, and the local probability is designed according to human common sense to describe the uncertainty information of the target element and other target elements, forming an entity fragment corresponding to the target element, and then merging the entity fragments into a multi-entity Bayesian network model.

[0112] The embodiment of the present invention does not specifically limit the construction tool of the multi-entity Bayesian network model. Exemplarily, the semantic positioning device based on probabilistic ontology knowledge can use the tool UnBBayes to build a multi-entity Bayesian network model, which provides a graphical user interface and Java language program interface for modeling multi-entity Bayesian networks and performing reasoning.

[0113] The embodiment of the present invention obtains a probabilistic ontology model based on location context information, establishes entity fragments based on human common sense through each target element in the probabilistic ontology model, and merges each entity fragment into a multi-entity Bayesian network model probability. It can improve the generalization degree of human life scenes, improve the accuracy of semantic positioning, and then make decisions to guide robots to perform tasks on operating objects in unstructured human living environments, thereby improving task execution efficiency.

[0114] Based on any of the above embodiments, a second probability value is inferred based on the target historical task data and the target situation Bayesian network, including: based on the target historical task data, using Dirichlet distribution, calculating the probability distribution of each node in the target situation Bayesian network, and updating the target situation Bayesian network.

[0115] Specifically, in step 103, the robot's target historical task data is used as a parameter learning sample, and a Bayesian parameter estimation method is used to calculate the local probability of each node state, and the obtained local probability distribution is used to update the structure of the target situation Bayesian network.

[0116] Preferably, the target historical task data set is used as a parameter learning sample, a uniform distribution is used to set the prior distribution of the probability, and then the Dirichlet distribution is used to calculate the posterior distribution of the probability.

[0117] The target task data includes the name of the random variable of each node in the target situation Bayesian network and the corresponding state of the random variable in this task. The embodiment of the present invention does not specifically limit the form of the target task data, and Table 1 can be used as an example:

[0118] Table 1 Schematic diagram of robot's historical task data

[0119] Age Group gender Preference Operation Object area Storage Location youth female refrigeration Drinks high temperature refrigerator middle aged male none food high temperature refrigerator middle aged female refrigeration Drinks high temperature refrigerator elderly male none food Low temperature dining table elderly female refrigeration food Low temperature cupboard … … … … … …

[0120] It is understandable that after the robot completes a task, it also needs to record the task according to the structure shown in Table 1.

[0121] Based on the updated target situation Bayesian network, a standard Bayesian inference algorithm is used to obtain a second probability value.

[0122] Specifically, the updated target situation Bayesian network is inferred using a standard Bayesian inference algorithm to obtain a second probability value.

[0123] The standard Bayesian inference algorithm used in the embodiment of the present invention includes, but is not limited to, a variable elimination algorithm (VE), a junction tree algorithm (JT), a Monte Carlo algorithm, or a variational method (Variational Methods) algorithm.

[0124] The embodiment of the present invention uses the target historical tasks as samples, infers the second probability value in the target situation Bayesian network, and then dynamically fuses the first probability value and the second probability value through the weight factor to obtain the positioning result and assist the robot in recognizing the semantic positioning of the operation object. It can combine human common sense and the reference value of historical tasks to improve the accuracy of semantic positioning, and then make decisions to guide the robot to perform tasks on the operation object in an unstructured human living environment, thereby improving the efficiency of task execution.

[0125] Based on any of the above embodiments, the calculation formula of the weight factor is:

[0126]

[0127] Among them, α t+1 is the weight factor at the t+1th moment, α t is the weight factor at the tth moment, n t+1 is the number of target historical task data at time t+1, n t is the number of target historical task data at the tth moment, d is the weight factor attenuation coefficient, and m is the attenuation speed.

[0128] Specifically, before step 104, a weight factor is set.

[0129] The embodiment of the present invention does not specifically limit the setting of the weight factor.

[0130] Exemplarily, the weight factor may be set autonomously by a user of the robot.

[0131] Exemplarily, the weight factor may be set accordingly according to different value ranges of the first probability value.

[0132] Preferably, the weight factor is determined by the weight factor at the previous moment, the weight factor attenuation coefficient, the quantity change value of the target historical task data, and the attenuation speed, and its mathematical expression is as follows:

[0133]

[0134] Among them, α t+1 is the weight factor at the t+1th moment, α t is the weight factor at the tth moment, and the value range of the weight factor at any moment is [0,1].

[0135] The embodiment of the present invention does not specifically limit the form of expression at any time.

[0136] Exemplarily, each moment may be any time point in the life cycle or usage cycle of the robot.

[0137] Exemplarily, each moment may correspond to any time interval within the life cycle or usage cycle of the robot.

[0138] Therefore, at the initial moment of the robot's life cycle or usage cycle, since the robot has no historical tasks in the cycle, the corresponding weight factor is 1.

[0139] m is the decay rate, and its value range is greater than or equal to 1. The decay rate is used to control the update frequency of the weight factor.

[0140] n t+1 is the number of target historical task data at time t+1, n t is the number of target historical task data at the tth moment, d is the weight factor attenuation coefficient, and both the attenuation coefficient d and the attenuation speed m are set manually.

[0141] Therefore, the weight factor can be dynamically adjusted according to the number of samples. The larger the amount of target historical sample data that is consistent with the current task, the smaller the weight factor, the weaker the influence of human common sense on the position of the robot's operating object, and the stronger the influence of the robot's task experience on the position.

[0142] The embodiment of the present invention is based on setting a dynamically adjusted weight factor, and then dynamically merges the first probability value and the second probability value to obtain a positioning result and assist the robot in recognizing the semantic positioning of the operation object. The weight can be dynamically adjusted according to the number of samples to improve the accuracy of semantic positioning, and then the robot can make decisions to guide the robot to perform tasks on the operation object in an unstructured human living environment, thereby improving the efficiency of task execution.

[0143] Figure 4 is a schematic diagram of the structure of the semantic positioning device based on probabilistic ontology provided by the present invention. Figure 4 As shown, the semantic positioning device based on probabilistic ontology knowledge provided by the embodiment of the present invention includes: a model building module 410, a first probability acquisition module 420, a second probability acquisition module 430 and a first semantic positioning module 440, wherein:

[0144] The model building module 410 is used to express probabilistic ontology knowledge and build a multi-entity Bayesian network model based on the position context information of the robot's operating object and human common sense.

[0145] The first probability acquisition module 420 is used to perform multiple instantiations based on the entity Bayesian network model, combine into a target situation Bayesian network, and infer a first probability value.

[0146] The second probability acquisition module 430 is used to infer a second probability value based on the target historical task data and the target situation Bayesian network when the robot has target historical task data.

[0147] The first semantic positioning module 440 is used to obtain the positioning result of the robot's operating object based on the first probability value and the second probability value by using a weight factor for calculation, so that the robot performs the task according to the positioning result of the operating object and records historical task data.

[0148] The target historical task data has a corresponding relationship with the position context information of the robot's operation object, and the position context information includes one or more of object type, temperature condition and user preference.

[0149] Specifically, the model building module 410 , the first probability acquisition module 420 , the second probability acquisition module 430 and the first semantic positioning module 440 are electrically linked in sequence.

[0150] The model building module 410 expresses probabilistic ontology knowledge according to the robot's operating objects, storage locations, and contextual information that affects the storage locations of objects, and sets each entity node in the ontology knowledge in the form of Bayesian probability based on human common sense to construct a multi-entity Bayesian network model.

[0151] The first probability acquisition module 420 instantiates the Bayesian network corresponding to each entity of the multi-entity Bayesian network model constructed by the model construction module 410 multiple times, combines them according to the principles of the probabilistic graphical model, generates a target situation Bayesian network, and combines evidence (i.e., prior knowledge corresponding to human common sense) with a standard Bayesian inference algorithm to perform reasoning to obtain a first probability value.

[0152] When the target historical task data is screened out, the second probability acquisition module 430 uses the historical task execution data as a Bayesian parameter estimation sample, performs reasoning using the structure of the target situation Bayesian network, and calculates a second probability value.

[0153] The first semantic positioning module 440 performs a weighted sum calculation on the first probability value and the second probability value in combination with their corresponding weight factors to obtain the positioning result of the robot for the operation object this time, and sends the positioning result to the control system of the robot, so that the control system of the robot, after receiving the positioning result, establishes a task for the purpose of carrying the operation object to the placement position indicated by the positioning result, controls the various components of the robot to perform the task, and records the operation object and the corresponding positioning result as historical task data in the local database after the task is completed.

[0154] Optionally, the device further includes a second semantic positioning module, wherein:

[0155] The second semantic positioning module is used to use the first probability value as the positioning result of the robot's operation object when there is no target historical task data for the robot, so that the robot performs the task according to the positioning result of the operation object and records the historical task data.

[0156] Optionally, the model building module 410 includes a probabilistic ontology expression unit and a model building unit, wherein:

[0157] The probabilistic ontology expression unit is used to construct a probabilistic ontology model based on the position context information of the robot's operation object.

[0158] The model building unit is used to design the corresponding local probabilities of each target element in the probabilistic ontology model according to human common sense, generate entity fragments, and merge the entity fragments into a multi-entity Bayesian network model.

[0159] The target elements include at least the storage location of the operation object, the type of the operation object, the area of ​​the operation object and user information.

[0160] Optionally, the second probability acquisition module 430 includes a network updating unit and a second probability acquisition unit, wherein:

[0161] The network updating unit is used to calculate the probability distribution of each node in the target situation Bayesian network based on the target historical task data and adopt Dirichlet distribution, and update the target situation Bayesian network.

[0162] The second probability acquisition unit is used to acquire a second probability value based on the updated target situation Bayesian network and using a standard Bayesian reasoning algorithm.

[0163] Optionally, the weight factor is calculated as:

[0164]

[0165] Among them, α t+1 is the weight factor at the t+1th moment, α t is the weight factor at the tth moment, n t+1 is the number of target historical task data at time t+1, n t is the number of target historical task data at the tth moment, d is the weight factor attenuation coefficient, and m is the attenuation speed.

[0166] The semantic localization device based on probabilistic ontology knowledge provided in an embodiment of the present invention is used to execute the above-mentioned semantic localization method based on probabilistic ontology knowledge of the present invention. Its implementation method is consistent with the implementation method of the semantic localization method based on probabilistic ontology knowledge provided by the present invention, and can achieve the same beneficial effects, which will not be repeated here.

[0167] The embodiment of the present invention is based on the probabilistic ontology knowledge expression, and performs a standardized description and modeling of the uncertainty of the object storage location, obtains a multi-entity Bayesian network model, and combines it into a target situation Bayesian network through multiple instantiations and generalizations. Human common sense and target historical tasks are used as samples, and the first probability value and the second probability value are inferred in the target situation Bayesian network. The first probability value and the second probability value are dynamically fused through the weight factor to obtain the positioning result and assist the robot in recognizing the semantic positioning of the operation object. It can combine the reference value of human common sense and historical tasks to improve the accuracy of semantic positioning, and then make decisions to guide the robot to perform tasks on the operation object in an unstructured human living environment, thereby improving the efficiency of task execution.

[0168] Figure 5 is a schematic diagram of the structure of the robot provided by the present invention. Based on any of the above embodiments, Figure 5 As shown, the robot 500 provided by the embodiment of the present invention includes a semantic positioning device 510 based on probabilistic ontology knowledge.

[0169] Specifically, the manager or user of the robot 500 may set a semantic positioning device 510 based on probabilistic ontology knowledge in the robot 500 according to different business requirements, so that the robot can perform tasks according to the corresponding business requirements.

[0170] Preferably, a plurality of semantic localization devices 510 based on probabilistic ontology knowledge in different business scenarios may be provided in the robot 500, so that the robot can perform a variety of tasks in a plurality of business scenarios.

[0171] The embodiment of the present invention is based on the probabilistic ontology knowledge expression, and performs a standardized description and modeling of the uncertainty of the object storage location, obtains a multi-entity Bayesian network model, and combines it into a target situation Bayesian network through multiple instantiations and generalizations. Human common sense and target historical tasks are used as samples, and the first probability value and the second probability value are inferred in the target situation Bayesian network. The first probability value and the second probability value are dynamically fused through the weight factor to obtain the positioning result and assist the robot in recognizing the semantic positioning of the operation object. It can combine the reference value of human common sense and historical tasks to improve the accuracy of semantic positioning, and then make decisions to guide the robot to perform tasks on the operation object in an unstructured human living environment, thereby improving the efficiency of task execution.

[0172] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6As shown, the electronic device may include: a processor (processor) 610, a communication interface (Communications Interface) 620, a memory (memory) 630 and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call the logic instructions in the memory 630 to execute a semantic positioning method based on probabilistic ontology knowledge, which includes: based on the position context information of the robot's operating object and human common sense, expressing probabilistic ontology knowledge and constructing a multi-entity Bayesian network model; based on the entity Bayesian network model, performing multiple instantiations to combine into a target situation Bayesian network, and inferring a first probability value; when the robot has target historical task data, inferring a second probability value based on the target historical task data and the target situation Bayesian network; based on the first probability value and the second probability value, using a weight factor for calculation to obtain the positioning result of the robot's operating object, so that the robot performs the task according to the positioning result of the operating object and records the historical task data; wherein the target historical task data has a corresponding relationship with the position context information of the robot's operating object, and the position context information includes one or more of object type, temperature conditions and user preferences.

[0173] In addition, the logic instructions in the above-mentioned memory 730 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0174] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the semantic positioning method based on probabilistic ontology knowledge provided by the above-mentioned methods, the method including: based on the position context information of the robot's operating object and human common sense, expressing probabilistic ontology knowledge and constructing a multi-entity Bayesian network model; based on the entity Bayesian network model, performing multiple instantiations to combine into a target situation Bayesian network, and inferring a first probability value; in the case where the robot has target historical task data, inferring a second probability value based on the target historical task data and the target situation Bayesian network; based on the first probability value and the second probability value, using a weight factor for calculation to obtain the positioning result of the robot's operating object, so that the robot performs the task according to the positioning result of the operating object and records the historical task data; wherein the target historical task data has a corresponding relationship with the position context information of the robot's operating object, and the position context information includes one or more of object type, temperature conditions and user preferences.

[0175] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the semantic positioning method based on probabilistic ontology knowledge provided by the above-mentioned methods, the method comprising: based on the position context information of the robot's operating object and human common sense, expressing probabilistic ontology knowledge and constructing a multi-entity Bayesian network model; based on the entity Bayesian network model, performing multiple instantiations to combine into a target situation Bayesian network, and inferring a first probability value; in the case where the robot has target historical task data, inferring a second probability value based on the target historical task data and the target situation Bayesian network; based on the first probability value and the second probability value, using a weight factor for calculation to obtain the positioning result of the robot's operating object, so that the robot performs the task according to the positioning result of the operating object and records the historical task data; wherein the target historical task data has a corresponding relationship with the position context information of the robot's operating object, and the position context information includes one or more of object type, temperature conditions and user preferences.

[0176] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0177] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A semantic positioning method based on probabilistic ontology knowledge, characterized in that: include: Based on the position context information of the robot's operating objects and human common sense, probabilistic ontology knowledge expression is performed to build a multi-entity Bayesian network model; Based on the entity Bayesian network model, multiple instantiations are performed to combine into a target situation Bayesian network, and a first probability value is inferred; In the case where the robot has target historical task data, inferring a second probability value based on the target historical task data and the target situation Bayesian network; Based on the first probability value and the second probability value, a weight factor is used for calculation to obtain a positioning result of the operation object of the robot, so that the robot performs a task according to the positioning result of the operation object and records historical task data; The target historical task data has a corresponding relationship with the position context information of the operation object of the robot, and the position context information includes one or more of object type, temperature condition and user preference.

2. The semantic positioning method based on probabilistic ontology knowledge according to claim 1 is characterized in that: After the first probability value is inferred, the method further includes: when there is no target historical task data for the robot, using the first probability value as the positioning result of the robot's operation object, so that the robot performs the task according to the positioning result of the operation object and records the historical task data.

3. The semantic positioning method based on probabilistic ontology knowledge according to claim 1 is characterized in that: The method of expressing probabilistic ontology knowledge based on the position context information of the robot's operating object and human common sense and constructing a multi-entity Bayesian network model includes: Construct a probabilistic ontology model based on the position context information of the robot's operating object; Design corresponding local probabilities for each target element in the probabilistic ontology model according to the human common sense, generate entity fragments, and merge the entity fragments into the multi-entity Bayesian network model; The target elements include at least the storage location of the operation object, the type of the operation object, the area of ​​the operation object and user information.

4. The semantic positioning method based on probabilistic ontology knowledge according to claim 1 is characterized in that: The inferring a second probability value based on the target historical task data and the target situation Bayesian network includes: Based on the target historical task data, the probability distribution of each node in the target situation Bayesian network is calculated using Dirichlet distribution, and the target situation Bayesian network is updated; Based on the updated target situation Bayesian network, a standard Bayesian inference algorithm is used to obtain the second probability value.

5. The semantic positioning method based on probabilistic ontology knowledge according to claim 1 is characterized in that: The calculation formula of the weight factor is: Among them, α t+1 is the weight factor at the t+1th moment, α t is the weight factor at the tth moment, n t+1 is the number of target historical task data at the t+1th moment, n t is the number of the target historical task data at the tth moment, d is the weight factor attenuation coefficient, and m is the attenuation speed.

6. A semantic positioning device based on probabilistic ontology knowledge, characterized in that: include: The model building module is used to express probabilistic ontology knowledge and build a multi-entity Bayesian network model based on the position context information of the robot's operating objects and human common sense; A first probability acquisition module is used to perform multiple instantiations based on the entity Bayesian network model, combine into a target situation Bayesian network, and infer a first probability value; A second probability acquisition module is used to infer a second probability value based on the target historical task data and the target situation Bayesian network when the robot has target historical task data; A first semantic positioning module, configured to obtain a positioning result of an operation object of the robot by using a weight factor for calculation based on the first probability value and the second probability value, so that the robot performs a task according to the positioning result of the operation object and records historical task data; The target historical task data has a corresponding relationship with the position context information of the operation object of the robot, and the position context information includes one or more of object type, temperature condition and user preference.

7. A robot, characterized in that: It includes the semantic positioning device based on probabilistic ontology knowledge as described in claim 6.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the semantic positioning method based on probabilistic ontology knowledge as described in any one of claims 1 to 5 are implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the semantic positioning method based on probabilistic ontology knowledge as claimed in any one of claims 1 to 5 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the semantic positioning method based on probabilistic ontology knowledge as claimed in any one of claims 1 to 5 are implemented.

Images