CN113723750A - Method and system for constructing robot humanoid operation action knowledge base - Google Patents
Method and system for constructing robot humanoid operation action knowledge base Download PDFInfo
- Publication number
- CN113723750A CN113723750A CN202110821417.4A CN202110821417A CN113723750A CN 113723750 A CN113723750 A CN 113723750A CN 202110821417 A CN202110821417 A CN 202110821417A CN 113723750 A CN113723750 A CN 113723750A
- Authority
- CN
- China
- Prior art keywords
- task
- planning
- tasks
- knowledge
- robot
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06311—Scheduling, planning or task assignment for a person or group
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/008—Artificial life, i.e. computing arrangements simulating life based on physical entities controlled by simulated intelligence so as to replicate intelligent life forms, e.g. based on robots replicating pets or humans in their appearance or behaviour
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Entrepreneurship & Innovation (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Educational Administration (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Game Theory and Decision Science (AREA)
- Robotics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Manipulator (AREA)
Abstract
The invention provides a method and a system for constructing a robot humanoid operation action knowledge base, which relate to the technical field of epidemic prevention robots, and the method comprises the following steps: step S1: carrying out universal solution on the inverse programming problem of the robot by adopting an inverse RRT algorithm; step S2: decomposing the tasks of the robot, establishing a task instruction set, extracting key task feature knowledge in the task instruction set, and listing feature knowledge required by each task; step S3: modeling the multitask planning of the robot by using a tabu search algorithm according to the characteristic knowledge, thereby realizing multitask corresponding planning of the heterogeneous robot; step S4: and constructing a subtask offline knowledge base based on manual and priori knowledge through multi-task corresponding planning. The invention can solve the problem that the inverse kinematics problem of the traditional humanoid epidemic prevention robot is difficult to solve directly, and improve the rapid cooperative operation capability of multiple robots in a complex environment.
Description
Technical Field
The invention relates to the technical field of epidemic prevention robots, in particular to a method and a system for constructing a robot humanoid operation action knowledge base.
Background
With the development of epidemic situations, the pressure of medical resources and public health resources becomes greater and greater, the problem of insufficient manpower and the problem of high risk of infection of workers become more and more serious, and a measure for replacing manpower is urgently needed. The tabu search algorithm is a meta-heuristic (random) random search algorithm that starts from an initial feasible solution by selecting a series of specific search directions (moves) as heuristics, and selecting the move that achieves the most change in the value of a specific objective function.
Robots are often used to perform dangerous, repetitive tasks to relieve labor. Can just meet the requirements for epidemic prevention. Therefore, epidemic prevention and control work by adopting epidemic prevention robots is considered in more and more places. There is a higher demand for response to such an emergency. The occurrence of epidemic situation can cause serious harm to society, and meanwhile, corresponding secondary disasters can possibly occur at any time; and because the precursor of the emergency is insufficient, the occurrence and evolution process have complexity which is equivalent, and the future situation of the epidemic situation is difficult to predict, the traditional ' prediction-response ' decision mode is difficult to completely meet the requirement, so the situation-response ' decision mode is required for responding the epidemic situation, the situation is comprehensively analyzed, and a reasonable decision is made.
From the above, a task planning method suitable for the current epidemic situation prevention and control needs to be adopted. In addition, the traditional epidemic prevention robot mainly adopts a wheel type mobile robot, which cannot well realize epidemic prevention work which can be carried out by human beings, and tasks of the humanoid epidemic prevention robot planner comprise solving a transformation equation, carrying out inverse kinematics solution, interpolation operation and the like. For the planning problem of the bionic robot, the bionic robot is a redundant robot, the inverse kinematics problem of the bionic robot is a typical time-varying and nonlinear problem, is described by an underdetermined and transcendental equation, is very difficult to solve, and has infinite solutions. Different constraints also exist for different mission plans, such as gait optimization plans. At present, there is no unified solving method in the academic world, and developing an efficient, unified, accurate and optimal solver is a key point for researching the subject. For the cooperation of heterogeneous multi-class robots, under different non-structural environments and complex tasks, multiple optimization indexes and different types of constraints must be integrated, and the inverse kinematics problem is a huge challenge.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a method and a system for constructing a knowledge base of humanoid operation actions of a robot, and aims to establish the knowledge base of humanoid operation actions of an epidemic prevention robot.
According to the method and the system for constructing the robot humanoid operation action knowledge base provided by the invention, the scheme is as follows:
in a first aspect, a method for constructing a robot humanoid operation action knowledge base is provided, and the method comprises the following steps:
carrying out universal solution on the inverse programming problem of the robot by adopting an inverse RRT algorithm;
decomposing the tasks of the robot, establishing a task instruction set, extracting key task feature knowledge in the task instruction set, and listing feature knowledge required by each task;
modeling the multitask planning of the robot by using a tabu search algorithm according to the characteristic knowledge, thereby realizing multitask corresponding planning of the heterogeneous robot;
and constructing a subtask offline knowledge base based on manual and priori knowledge through multi-task corresponding planning.
Preferably, the performing the general solution by using the RRT algorithm includes:
defining a starting point xinitRandomly scattering a point in the environment to obtain a point xrandJudgment of xrandWhether or not within the barrier region;
if xrandNot in the barrier area, x is connectedinitAnd xrandThus obtaining a connecting line L and judging whether the L is in the obstacle area;
if L is not entirely inside the obstacle, then along L, from xinitTo xrandIs moved a certain distance to obtain a new point xnewThen xinitAnd xnewAnd the line segments between them form the simplest tree;
expanding the tree, continuously repeating the process on the basis of the initial point, scattering points in the environment, and obtaining points x which are not in the area of the obstaclerandThen find a departure x on the existing treerandNearest point xnearConnecting two points, along which the line, if it is not in the region of the obstacle, runs from xnearTo xrandMoving a certain distance to obtain a new point xnewThis point is added to an already existing tree;
the above process is repeated until the target point or a point near it is added to the tree.
Preferably, the task decomposition of the robot comprises non-contact scale body temperature screening and card punching; detecting and reminding wearing a mask; monitoring the density of the people stream; intercepting the passerby and the passerby to pass; epidemic prevention broadcasting; epidemic situation propaganda and diagnosis guide; the remote doctor consults the task.
Preferably, the constructing of the subtask offline knowledge base based on manual and prior knowledge includes:
programming an HTN by using a hierarchical task network, decomposing the tasks step by step according to domain knowledge, solving an action sequence by using a sub-action decomposition method with fixed tasks as the domain knowledge for programming, and performing the tasks;
and establishing a hierarchical task planning model, describing the state, initial network and field of the subtasks, decomposing the initial non-original tasks into some non-original tasks or operations, and then continuously decomposing the non-original tasks according to a method in the planning field until the completion of the decomposition of the subtasks, thereby finally forming a sub-network off-line knowledge base.
Preferably, the hierarchical task network planning HTN process includes:
step S4-3: finding corresponding initial planning states and tasks in the problem domain and the knowledge domain: s, T, D;
wherein s represents an initial planning state, T represents an initial task list, D represents knowledge domain information, the initial planning list P is enabled to be empty, and a task T is selected from the current task list T;
step S4-4: judging whether t is an atomic task, if so, entering step S4-5, otherwise, entering step S4-6;
step S4-5: if t is an atomic task, selecting an operation o for completing the task t in the current state from an operator set in the knowledge domain D, if the operation o exists, adding the operation o into a task planning list, updating the current network state, completing the decomposition of the task t, and deleting the task t from a task network;
step S4-6: if the t is not an atomic task, selecting a task m decomposed in the current state from a method set of the knowledge domain D, if the method m exists, decomposing the task t into subtasks n meeting the method m, and replacing the task t in the task network with n;
step S4-7: and repeating the step S4-4, the step S4-5 and the step S4-6 until all tasks in the task list T are executed, namely all tasks are decomposed into atomic tasks, outputting a planning list P after the whole planning is finished, and finishing the construction of a knowledge base, thereby constructing the self-network based on the multiple tasks and carrying out the transfer learning training.
In a second aspect, a system for constructing a knowledge base of actions of a robot humanoid operation is provided, the system comprising:
module M1: carrying out universal solution on the inverse programming problem of the robot by adopting an inverse RRT algorithm;
module M2: decomposing the tasks of the robot, establishing a task instruction set, extracting key task feature knowledge in the task instruction set, and listing feature knowledge required by each task;
module M3: modeling the multitask planning of the robot by using a tabu search algorithm according to the characteristic knowledge, thereby realizing multitask corresponding planning of the heterogeneous robot;
module M4: and constructing a subtask offline knowledge base based on manual and priori knowledge through multi-task corresponding planning.
Preferably, the module M1 includes:
defining a starting point xinitIn the environmentRandomly scattering one point to obtain a point xrandJudgment of xrandWhether or not within the barrier region;
if xrandNot in the barrier area, x is connectedinitAnd xrandThus obtaining a connecting line L and judging whether the L is in the obstacle area;
if L is not entirely inside the obstacle, then along L, from xinitTo xrandIs moved a certain distance to obtain a new point xnewThen xinitAnd xnewAnd the line segments between them form the simplest tree;
expanding the tree, continuously repeating the process on the basis of the initial point, scattering points in the environment, and obtaining points x which are not in the area of the obstaclerandThen find a departure x on the existing treerandNearest point xnearConnecting two points, along which the line, if it is not in the region of the obstacle, runs from xnearTo xrandMoving a certain distance to obtain a new point xnewThis point is added to an already existing tree;
the above process is repeated until the target point or a point near it is added to the tree.
Preferably, the task decomposition of the robot in the module M2 comprises non-contact scale body temperature screening and card punching; detecting and reminding wearing a mask; monitoring the density of the people stream; intercepting the passerby and the passerby to pass; epidemic prevention broadcasting; epidemic situation propaganda and diagnosis guide; the remote doctor consults the task.
Preferably, the module M4 includes:
module M4-1: programming an HTN by using a hierarchical task network, decomposing the tasks step by step according to domain knowledge, solving an action sequence by using a sub-action decomposition method with fixed tasks as the domain knowledge for programming, and performing the tasks;
module M4-2: and establishing a hierarchical task planning model, describing the state, initial network and field of the subtasks, decomposing the initial non-original tasks into some non-original tasks or operations, and then continuously decomposing the non-original tasks according to a method in the planning field until the completion of the decomposition of the subtasks, thereby finally forming a sub-network off-line knowledge base.
Preferably, the hierarchical task network planning HTN process includes:
module M4-3: finding corresponding initial planning states and tasks in the problem domain and the knowledge domain: s, T, D;
wherein s represents an initial planning state, T represents an initial task list, D represents knowledge domain information, the initial planning list P is enabled to be empty, and a task T is selected from the current task list T;
module M4-4: judging whether t is an atomic task, if so, entering step S4-5, otherwise, entering step S4-6;
module M4-5: if t is an atomic task, selecting an operation o for completing the task t in the current state from an operator set in the knowledge domain D, if the operation o exists, adding the operation o into a task planning list, updating the current network state, completing the decomposition of the task t, and deleting the task t from a task network;
module M4-6: if the t is not an atomic task, selecting a task m decomposed in the current state from a method set of the knowledge domain D, if the method m exists, decomposing the task t into subtasks n meeting the method m, and replacing the task t in the task network with n;
module M4-7: and repeating the module M4-4, the module M4-5 and the module M4-6 until all tasks in the task list T are executed, namely all tasks are decomposed into atomic tasks, outputting a planning list P after the whole planning is finished, and finishing the construction of a knowledge base, thereby constructing a self-network based on multiple tasks and carrying out transfer learning training.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention provides a method for constructing a robot humanoid operation action knowledge base by taking the serious coronavirus COVID-19 epidemic situation in the world as the background, and solves the problem that the traditional humanoid robot inverse kinematics problem is difficult to solve directly. An efficient, unified, accurate and optimal inverse kinematics solver is developed;
2. the invention provides a general planning method aiming at the problems that the traditional heterogeneous robot motion planning method is usually modeled and planned according to a specific bionic robot and needs to be re-planned according to models of the bionic robot when the bionic robot is transplanted to other heterogeneous robots, and the general planning method has universality of the heterogeneous robot;
3. the invention takes the hierarchical task network planning as the basis, gradually decomposes the task according to the domain knowledge, and uses the sub-action decomposition method with fixed task as the planning domain knowledge to solve the action sequence and carry out the task. The multi-objective hybrid optimization planning capability is provided, and the operation action knowledge base corresponding to the tasks and the configurations is established, so that the action knowledge base not only is a multi-robot motion track generated by hybrid intelligent planning, but also has learning and reasoning capabilities, and the rapid cooperative operation capability of multiple robots in a complex environment can be improved.
Drawings
Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings:
FIG. 1 is a system overview framework;
FIG. 2 is a flow chart of the inverse RRT algorithm calculation;
FIG. 3 is a flow chart of the calculation of the tabu search algorithm;
FIG. 4 is a flow chart of a hierarchical mission network planning algorithm;
FIG. 5 is an exploded view of a non-contact scale body temperature screening and punch-card job without a heat generator at an entrance to a public place;
FIG. 6 is an exploded view of a non-contact scale body temperature screening and punch-card job with a fever sufferer at an entrance to a public place;
FIG. 7 is an exploded view of a mask wearing detection and reminder task in the case where all masks are worn at the entrance of a public place;
FIG. 8 is an exploded view of a mask wearing detection and reminder task in the absence of a mask at an entrance of a public place;
FIG. 9 is an exploded view of a traffic density detection task at an entrance of a public place where traffic density is normal;
fig. 10 is an exploded view of the people stream density detection task in the case of excessive people stream density at the entrance of a public place.
Detailed Description
The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that it would be obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit of the invention. All falling within the scope of the present invention.
The embodiment of the invention provides a method for constructing a robot humanoid operation action knowledge base, which is shown in figure 1 and specifically comprises the following steps:
and carrying out universal solution on the inverse planning problem of the humanoid robot by adopting an inverse RRT algorithm.
Describing the target state of the heterogeneous robot, modeling by using a tabu search algorithm, establishing a model for changing the task state, and establishing a unified multi-task planning method.
And describing the target state and the newly added target state of each task on the basis of the model.
And establishing a task library scheme evaluation function, and describing a time sequence for optimizing the task decision of the humanoid robot, such as minimizing task completion time and the like.
And opening up a task library environment interface for describing a new task environment state, connecting the new task environment state into a library in a plug-in mode, and establishing an operation information task instruction set of the robot, the operation workpiece state and the operation environment by combining the original basic integral operation task.
And extracting key task feature knowledge in the task instruction set and listing feature knowledge required by each task.
And determining the overall actions of the robot in the task library, which are required to be completed by completing the operation task, and solving the actions by a task planning method according to the state change of each subtask.
And (3) adopting hierarchical task network planning, decomposing the tasks step by step according to the domain knowledge, and solving the action sequence by taking the sub-action decomposition method with fixed tasks as the planning domain knowledge to carry out the tasks.
And establishing a hierarchical task planning model, describing the state, initial network and field of the subtasks, decomposing the initial non-original tasks into some non-original tasks or operations, and then continuously decomposing the non-original tasks according to a method in the planning field until the completion of the decomposition of the subtasks, thereby finally forming a sub-network off-line knowledge base.
And constructing and researching a multi-task sub-network according to the independent task unit after the overall task planning and the subtask decomposition.
As shown in fig. 1, a method for constructing a robot humanoid operation action knowledge base includes a human body action model, a heterogeneous robot model, a decision planner, a knowledge base, a safety controller humanoid and a robot execution part. The main fractions were analyzed separately as follows:
modeling human dynamics: the dynamic analysis of human body is the key and the basis of the humanoid robot, and the part explores the biomechanical characteristics of human body based on the anatomical structures of important joints of human body such as shoulders, elbows, waists, hips, knees, ankles and the like and the motion mechanism of the human body. Simplifying the model of human motion into a skeleton model, and establishing a human dynamics model suitable for a complex non-structural environment. This model should take into account the task requirements of fast jobs. And the simulation and experimental tests verify the motion mechanism and biological characteristics of the human body adaptive to the non-structural environment.
Modeling robot dynamics: and establishing a corresponding humanoid robot kinematics and dynamics model according to the human body biomechanics model. The model can make the robot more vivid and more fit with the characteristics of human beings, thereby better interacting with people naturally. Different from the traditional simplified model, such as a 12-bar model, the model combines the movements of important joint arms of a head, a shoulder, an elbow, a waist, a hip, a knee, an ankle and the like of a human body for the first time to form a brand-new humanoid robot dynamics model. On the basis, the coordination task requirement of rapid operation of the multi-humanoid robot in the non-structural environment is analyzed, and a scheme for realizing rapid operation of the humanoid robot through joint drive control and action track planning matching research is provided.
Establishing a multi-robot global optimization mapping relation: on the basis of the two models, a complex characteristic mapping relation between human dynamics and robot dynamics is further constructed, and human knowledge is converted into a robot working space. On the basis of the mapping relation, under the condition of interaction between multi-robot collaborative interaction and a non-structural environment, a set of complex mapping relation capable of realizing global optimization needs to be established. The establishment of the mapping scheme can not only well complete the cooperative operation task, but also realize obstacle avoidance, mutual avoidance among robots and the like.
A decision planner: the planner can not only well complete the end operation task, avoid obstacles and avoid robots, but also dynamically coordinate tasks, distribute loads, decompose motions, operate flexible large objects and the like, and realize the optimization under a given criterion in a global sense.
Solving an inverse problem: the tasks of the planner include solving a transformation equation, performing inverse kinematics solution, interpolation operation and the like. For the planning problem of the bionic robot, the bionic robot is a redundant robot, the inverse kinematics problem of the bionic robot is a typical time-varying and nonlinear problem, is described by an underdetermined and transcendental equation, is very difficult to solve, and has infinite solutions. Different constraints also exist for different mission plans, such as gait optimization plans. At present, no unified solving method exists in the academic world, and the development of an efficient, unified, accurate and optimal solver is a research key point of the method. For the cooperation of heterogeneous multi-class robots, under different non-structural environments and complex tasks, multiple optimization indexes and different types of constraints must be integrated, and the inverse kinematics problem is a huge challenge of the method.
And (3) corresponding planning of heterogeneous multitasks: because the robot system has different configurations and various categories, the motion planning method is often modeled and planned according to a specific bionic robot, and when the robot system is transplanted to other heterogeneous robots, the robot system needs to be re-planned according to models of the robots. Therefore, it is desirable that the knowledge base should include heterogeneous and multitasking correspondence rules.
Rule matching of actions: in order to realize rapid and safe operation, multi-objective hybrid optimization planning capability needs to be provided, and an operation action knowledge base corresponding to tasks and configurations is established, so that the action knowledge base not only is a multi-robot motion track generated by hybrid intelligent planning, but also has learning and reasoning capabilities, and the rapid cooperative operation capability of multiple robots in a complex environment is improved.
The inverse RRT algorithm calculates an inverse kinematics flow:
as shown in fig. 2, after the generalized inverse of the robot is obtained, the inverse kinematics solution is performed on the motion of the epidemic prevention robot by using an inverse RRT planning method. For the RRT method, there are the following steps:
first in the environment we have a starting point, defined as xinitThen we randomly scatter a point in the environment to get the point xrandIf x israndNot in the barrier area, x is connectedinitAnd xrandWe get a line L, along which we follow from x if L is not entirely inside the obstacleinitTo xrandIs moved a certain distance to obtain a new point xnewThen xinitAnd xnewAnd the line segments between them constitute the simplest tree.
And (3) expanding the tree: on the basis of the starting, continuously repeating the steps, scattering points in the environment, and obtaining a point x of an obstacle-free arearandThen find a departure x on the existing treerandNearest point xnearConnecting two points, along which line, if there is no obstacle, from xnearTo xrandMoving a certain distance to obtain a new point xnewThis point is added to the already existing tree.
The above process is repeated until the target point or a point near it is added to the tree, at which point we can find a path from the starting point to the target point on the tree.
Modeling process of tabu search algorithm:
as shown in fig. 3, a tabu search method is used for modeling multi-task planning of the robot, so that multi-task corresponding planning of the heterogeneous robot is achieved. Some of the contents of the tabu search algorithm are explained as follows:
neighborhood: for the combinatorial optimization problem, given any feasible solution x, x ∈ D, D is the domain of decision variables, for one mapping on D: n x ∈ D → N (x) ∈ 2(D) where 2(D) represents the set of all subsets of D, N (x) becomes a neighborhood of x, y ∈ N (x) is called a neighbor of x.
Candidate set: the candidate set generally consists of neighbors in the neighborhood, and all the neighbors of a certain solution can be used as the candidate set, can also be extracted through optimal extraction, and can also be extracted randomly.
Tabu table: the tabu table includes tabu objects and tabu lengths, and since repeated steps need to be avoided in each search for the current solution, certain elements are put in the tabu table, which will not be considered in the next search, and these tabu objects are the tabu objects.
Taboo length: the number of the most taboo objects that can be accepted by the taboo table may be too large, which may result in a long time consumption or stop of the algorithm, and too small may result in repeated searches.
Evaluation function: used for evaluating the quality of the current solution.
Privileged rules: in the taboo search algorithm, one element of the candidate set is prohibited from being searched in one step of iteration, but the evaluation function is improved if the element is prohibited, so that a privileged rule needs to be set, and when the condition is met, the element jumps out of the taboo list.
Termination rule: generally, when the local optimal solutions obtained by two iterations do not change any more, or the evaluation functions of the two optimal solutions do not differ much, or the iteration is stopped after n iterations, the third method is usually selected.
Hierarchical task network planning process:
as shown in fig. 4, the goal of the hierarchical network of tasks planning HTN is to obtain a planning list, i.e., a knowledge base, that contains only atomic tasks. The whole process is as follows:
step 1: finding corresponding initial planning states and tasks in the problem domain and the knowledge domain: s, T, D, wherein s represents the initial planning state, T represents the initial task list, D represents the knowledge domain information, the initial planning list P is enabled to be empty, and one task T is selected from the current task list T.
Step 2: and judging whether t is an atomic task, if so, entering a step 3, and otherwise, entering a step 4.
And step 3: if t is an atomic task, selecting an operation o which can complete the task t in the current state from an operator set in the knowledge domain D, if the operation o exists, adding the operation o into a task planning list, updating the current network state, completing the decomposition of the task t, and deleting the task t from a task network.
And 4, step 4: and if the t is not an atomic task, selecting a task m which can be decomposed in the current state from the method set of the knowledge domain D, and if the method m exists, decomposing the task t into a subtask n meeting the method m and replacing the task t in the task network with the subtask n.
And 5: and (5) repeating the steps (2), (3) and (4) until all tasks in the task list T are executed, namely all tasks are decomposed into atomic tasks, and after the whole planning is finished, outputting a planning list P and finishing the construction of a knowledge base.
And (3) constructing a subtask network aiming at each task:
we can first get a task list of the robot, for which the tasks are: non-contact scale body temperature screening and card punching, including single body temperature detection and multi-person body temperature simultaneous detection; detecting and reminding wearing a mask; monitoring the density of the people stream; intercepting the passerby and the passerby to pass; epidemic prevention broadcasting; epidemic situation propaganda and diagnosis guide; the remote doctor consults these seven tasks.
The task of the robot is decomposed into key task feature knowledge. The following decomposition is carried out: non-contact scale body temperature screening and card punching: infrared thermal imaging techniques; wear gauze mask and detect and remind: face recognition technology, target detection technology; and (3) people stream density detection: a deep learning technique; intercepting pedestrians and letting pedestrians pass: a motion planning technique; epidemic prevention broadcasting: a voice technique; epidemic situation propaganda and diagnosis guide: visualization technology and human-computer interaction technology; remote doctor consultation: real-time communication technology.
After getting the key technology, we can decompose each task using HTN.
Task one: non-contact scale body temperature screening and card punching. The limited environment is an entrance of a public place, and the influence is to prevent the heat-generating people from entering the public place. As shown in fig. 5, wherein 1 represents a non-atomic task; is there a Indicating whether a task exists; 2 represents the execution sequence of tasks on the same layer of nodes; 3 is an atomic task; | A Representing an atomic task. The uppermost node body temperature detection (. By combining the key technology of body temperature detection and the limitation of a robot executing mechanism, when the robot executes a body temperature detection task, thermal imaging is firstly used for body temperature detection, then a normal body temperature person is released, a heater is alarmed, and the robot is forbidden to enter a public place. The task is thus broken down into: | A Prevent traffic action! Thermography detects body temperature, normothermia (! Prevent traffic action and! Thermal imaging to detect body temperature has been an atomic task and can be performed directly by a robot. When there is no fever patient, the second task can be decomposed according to the HTN idea into: | A Send normal information! Permission to pass action! And allowing three sub-processes to pass, namely sending normal information and allowing the normothermic person to pass, namely an atomic task without decomposition.
At this time, the HTN decomposition is completed when the detection result is no heat generator, and the atomic task is: the method comprises the steps of preventing passing action, carrying out thermal imaging body temperature detection, sending normal information, allowing the passing action and allowing the passing.
As shown in fig. 6, when the detection result is a person with a heat, the task is directly decomposed into four atomic tasks: the passing action is prevented, the thermal imaging body temperature detection is carried out, the alarm information is issued, and the passing of the heat generator is forbidden.
And a second task: and (5) detecting and reminding by wearing the mask. The environment is limited to an entrance of a public place, and the influence is that people who do not wear the mask can be prevented from entering the public place. As shown in fig. 7, in combination with the key technology of mask wearing detection and the limitation of the robot executing mechanism, the robot should perform body temperature detection by using a deep learning method when performing a mask wearing detection task, and then release the person wearing the mask, alarm the person not wearing the mask, and forbid the person from entering the public. Namely, the task is divided into: | A Prevent traffic action! The mask wearing condition and the mask wearing state are detected through deep learning (is normal).
Wherein! Prevent traffic action and! The deep learning detection of the wearing condition of the mask is an atomic task. When the detection result is that the mask is worn completely, HTN decomposition is carried out on the mask worn (normal), and three atomic tasks are obtained: | A Send normal information! Allow a traffic action and! And (5) normally detecting. At this time, the HTN decomposition is completed when the detection result is that all the masks are worn, and the atomic tasks are as follows: and (4) carrying out deep learning to detect the wearing condition of the mask, sending normal information and allowing the user to pass.
As shown in fig. 8, when the detection result shows that there is no mask, the four atomic tasks are directly decomposed: the method has the advantages of preventing traffic movement, performing deep learning to detect the wearing condition of the mask, issuing alarm information and forbidding the persons who do not wear the mask to pass.
And a third task: and (5) detecting the density of the stream of people. The limited environment is an entrance of a public place, and the influence causes epidemic prevention and control problems for preventing a large amount of gathering of personnel. As shown in fig. 9, in combination with the key technology of people stream density detection and the limitation of the robot executing mechanism, the robot should perform body temperature detection by using a deep learning method when performing a people stream density detection task, and then release a normal people stream density area, warn a place with excessive people stream density, and evacuate people. Namely, the task is divided into: | A Prevent traffic action! MCNN detects density, people stream density (.
Wherein! Prevent traffic action and! MCNN detection density has been an atomic task. When the detection result is that the human stream density is normal, performing HTN decomposition on the human stream density (: | A Send normal information! Allow a traffic action and! And (5) normally detecting. At this time, the HTN decomposition is completed when the detection result is that the density of the human stream is normal, and the atomic tasks are as follows: and carrying out MCNN detection on people stream density, sending normal information and carrying out normal detection.
As shown in fig. 10, when the detection result is that the density of the human stream is too large, the direct decomposition is into five atomic tasks: and stopping the passing action, carrying out MCNN to detect the people flow density, sending people flow density warning information, carrying out crowd evacuation action, and carrying out crowd evacuation.
And a fourth task: and blocking pedestrians and letting pedestrians to pass. And performing action planning of intercepting the passerby and passing the passerby on the robot through the obtained inverse kinematics method and the modeling model. The environment is limited to the entrance of the public place and the inside of the public place, and the influence is to intercept or release pedestrians. The atomic task, namely the task itself, can be directly realized by the obtained kinematics planning method.
And a fifth task: and (5) epidemic prevention broadcasting. The limited environment is at the entrance of the public place and inside the public place, and the influence is the situation of epidemic situation prevention and control. The atomic task, i.e. the task itself, can be directly implemented by means of speech technology.
And a sixth task: epidemic situation propaganda and diagnosis guide. The limited environment is at the entrance of a public place and inside the public place, the influence is for propagandizing epidemic situation prevention and control conditions, and suspected epidemic situation personnel are guided to see a doctor. The atomic task is the task itself, and is directly realized through a visualization technology and a human-computer interaction technology.
And a seventh task: and the remote doctor consults. The limited environment is the entrance of the public place, the inside of the public place and the inside of the hospital isolation area, and the influence is to help the suspected epidemic situation personnel and the disease confirming personnel to contact the doctor to diagnose the self condition. The atomic task is the task itself and is directly realized through a real-time communication technology.
Constructing a robot action knowledge base:
the method comprises the steps of decomposing various action tasks of the robot into atomic tasks which can be directly executed by performing hierarchical task network planning on the tasks, and establishing a sub-network offline knowledge base of the robot, thereby constructing a self-network based on multiple tasks and performing transfer learning training.
The embodiment of the invention provides a method for constructing a robot humanoid operation action knowledge base, which solves the problem that the inverse kinematics problem of the traditional humanoid robot is difficult to directly solve. An efficient, unified, accurate and optimal inverse kinematics solver is developed; the invention provides a general planning method aiming at the problems that the traditional heterogeneous robot motion planning method is usually modeled and planned according to a specific bionic robot and needs to be re-planned according to models of the bionic robot when the bionic robot is transplanted to other heterogeneous robots, and the general planning method has universality of the heterogeneous robot; the invention takes the hierarchical task network planning as the basis, gradually decomposes the task according to the domain knowledge, and uses the sub-action decomposition method with fixed task as the planning domain knowledge to solve the action sequence and carry out the task. The multi-objective hybrid optimization planning capability is provided, and the operation action knowledge base corresponding to the tasks and the configurations is established, so that the action knowledge base not only is a multi-robot motion track generated by hybrid intelligent planning, but also has learning and reasoning capabilities, and the rapid cooperative operation capability of multiple robots in a complex environment can be improved.
Those skilled in the art will appreciate that, in addition to implementing the system and its various devices, modules, units provided by the present invention as pure computer readable program code, the system and its various devices, modules, units provided by the present invention can be fully implemented by logically programming method steps in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Therefore, the system and various devices, modules and units thereof provided by the invention can be regarded as a hardware component, and the devices, modules and units included in the system for realizing various functions can also be regarded as structures in the hardware component; means, modules, units for performing the various functions may also be regarded as structures within both software modules and hardware components for performing the method.
The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. The embodiments and features of the embodiments of the present application may be combined with each other arbitrarily without conflict.
Claims (10)
1. A method for constructing a robot humanoid operation action knowledge base is characterized by comprising the following steps:
step S1: carrying out universal solution on the inverse programming problem of the robot by adopting an inverse RRT algorithm;
step S2: decomposing the tasks of the robot, establishing a task instruction set, extracting key task feature knowledge in the task instruction set, and listing feature knowledge required by each task;
step S3: modeling the multitask planning of the robot by using a tabu search algorithm according to the characteristic knowledge, thereby realizing multitask corresponding planning of the heterogeneous robot;
step S4: and constructing a subtask offline knowledge base based on manual and priori knowledge through multi-task corresponding planning.
2. The method of claim 1, wherein the step S1 of performing the universal solution using the RRT algorithm comprises:
step S1-1: defining a starting point xinitRandomly scattering a point in the environment to obtain a point xrandJudgment of xrandWhether or not within the barrier region;
step S1-2: if xrandNot in the barrier area, x is connectedinitAnd xrandThus obtaining a connecting line L and judging whether the L is in the obstacle area;
step S1-3: if L is not entirely inside the obstacle, then along L, from xinitTo xrandIs moved a certain distance to obtain a new point xnewThen xinitAnd xnewAnd the line segments between them form the simplest tree;
step S1-4: expanding the tree, continuously repeating the process on the basis of the initial point, scattering points in the environment to obtainPoint x not in the area of the obstaclerandThen find a departure x on the existing treerandNearest point xnearConnecting two points, along which the line, if it is not in the region of the obstacle, runs from xnearTo xrandMoving a certain distance to obtain a new point xnewThis point is added to an already existing tree;
step S1-5: the above process is repeated until the target point or a point near it is added to the tree.
3. The method as claimed in claim 1, wherein the task decomposition of the robot in the step S2 includes non-contact scale body temperature screening and card punching; detecting and reminding wearing a mask; monitoring the density of the people stream; intercepting the passerby and the passerby to pass; epidemic prevention broadcasting; epidemic situation propaganda and diagnosis guide; the remote doctor consults the task.
4. The method according to claim 1, wherein the step S4 includes:
step S4-1: programming an HTN by using a hierarchical task network, decomposing the tasks step by step according to domain knowledge, solving an action sequence by using a sub-action decomposition method with fixed tasks as the domain knowledge for programming, and performing the tasks;
step S4-2: and establishing a hierarchical task planning model, describing the state, initial network and field of the subtasks, decomposing the initial non-original tasks into some non-original tasks or operations, and then continuously decomposing the non-original tasks according to a method in the planning field until the completion of the decomposition of the subtasks, thereby finally forming a sub-network off-line knowledge base.
5. The method of claim 4, wherein the hierarchical mission network planning (HTN) process comprises:
step S4-3: finding corresponding initial planning states and tasks in the problem domain and the knowledge domain: s, T, D;
wherein s represents an initial planning state, T represents an initial task list, D represents knowledge domain information, the initial planning list P is enabled to be empty, and a task T is selected from the current task list T;
step S4-4: judging whether t is an atomic task, if so, entering step S4-5, otherwise, entering step S4-6;
step S4-5: if t is an atomic task, selecting an operation o for completing the task t in the current state from an operator set in the knowledge domain D, if the operation o exists, adding the operation o into a task planning list, updating the current network state, completing the decomposition of the task t, and deleting the task t from a task network;
step S4-6: if the t is not an atomic task, selecting a task m decomposed in the current state from a method set of the knowledge domain D, if the method m exists, decomposing the task t into subtasks n meeting the method m, and replacing the task t in the task network with n;
step S4-7: and repeating the step S4-4, the step S4-5 and the step S4-6 until all tasks in the task list T are executed, namely all tasks are decomposed into atomic tasks, outputting a planning list P after the whole planning is finished, and finishing the construction of a knowledge base, thereby constructing the self-network based on the multiple tasks and carrying out the transfer learning training.
6. A system for constructing a knowledge base of actions of a robot humanoid operation is characterized by comprising:
module M1: carrying out universal solution on the inverse programming problem of the robot by adopting an inverse RRT algorithm;
module M2: decomposing the tasks of the robot, establishing a task instruction set, extracting key task feature knowledge in the task instruction set, and listing feature knowledge required by each task;
module M3: modeling the multitask planning of the robot by using a tabu search algorithm according to the characteristic knowledge, thereby realizing multitask corresponding planning of the heterogeneous robot;
module M4: and constructing a subtask offline knowledge base based on manual and priori knowledge through multi-task corresponding planning.
7. The system according to claim 6, characterized in that said module M1 comprises:
defining a starting point xinitRandomly scattering a point in the environment to obtain a point xrandJudgment of xrandWhether or not within the barrier region;
if xrandNot in the barrier area, x is connectedinitAnd xrandThus obtaining a connecting line L and judging whether the L is in the obstacle area;
if L is not entirely inside the obstacle, then along L, from xinitTo xrandIs moved a certain distance to obtain a new point xnewThen xinitAnd xnewAnd the line segments between them form the simplest tree;
expanding the tree, continuously repeating the process on the basis of the initial point, scattering points in the environment, and obtaining points x which are not in the area of the obstaclerandThen find a departure x on the existing treerandNearest point xnearConnecting two points, along which the line, if it is not in the region of the obstacle, runs from xnearTo xrandMoving a certain distance to obtain a new point xnewThis point is added to an already existing tree;
the above process is repeated until the target point or a point near it is added to the tree.
8. The system according to claim 6, wherein the task decomposition of the robot in the module M2 includes non-contact scale body temperature screening and card punching; detecting and reminding wearing a mask; monitoring the density of the people stream; intercepting the passerby and the passerby to pass; epidemic prevention broadcasting; epidemic situation propaganda and diagnosis guide; the remote doctor consults the task.
9. The system according to claim 6, characterized in that said module M4 comprises:
module M4-1: programming an HTN by using a hierarchical task network, decomposing the tasks step by step according to domain knowledge, solving an action sequence by using a sub-action decomposition method with fixed tasks as the domain knowledge for programming, and performing the tasks;
module M4-2: and establishing a hierarchical task planning model, describing the state, initial network and field of the subtasks, decomposing the initial non-original tasks into some non-original tasks or operations, and then continuously decomposing the non-original tasks according to a method in the planning field until the completion of the decomposition of the subtasks, thereby finally forming a sub-network off-line knowledge base.
10. The system of claim 9, wherein the hierarchical mission network planning (HTN) process comprises:
module M4-3: finding corresponding initial planning states and tasks in the problem domain and the knowledge domain: s, T, D;
wherein s represents an initial planning state, T represents an initial task list, D represents knowledge domain information, the initial planning list P is enabled to be empty, and a task T is selected from the current task list T;
module M4-4: judging whether t is an atomic task, if so, entering step S4-5, otherwise, entering step S4-6;
module M4-5: if t is an atomic task, selecting an operation o for completing the task t in the current state from an operator set in the knowledge domain D, if the operation o exists, adding the operation o into a task planning list, updating the current network state, completing the decomposition of the task t, and deleting the task t from a task network;
module M4-6: if the t is not an atomic task, selecting a task m decomposed in the current state from a method set of the knowledge domain D, if the method m exists, decomposing the task t into subtasks n meeting the method m, and replacing the task t in the task network with n;
module M4-7: and repeating the module M4-4, the module M4-5 and the module M4-6 until all tasks in the task list T are executed, namely all tasks are decomposed into atomic tasks, outputting a planning list P after the whole planning is finished, and finishing the construction of a knowledge base, thereby constructing a self-network based on multiple tasks and carrying out transfer learning training.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110821417.4A CN113723750A (en) | 2021-07-20 | 2021-07-20 | Method and system for constructing robot humanoid operation action knowledge base |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110821417.4A CN113723750A (en) | 2021-07-20 | 2021-07-20 | Method and system for constructing robot humanoid operation action knowledge base |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113723750A true CN113723750A (en) | 2021-11-30 |
Family
ID=78673598
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110821417.4A Pending CN113723750A (en) | 2021-07-20 | 2021-07-20 | Method and system for constructing robot humanoid operation action knowledge base |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113723750A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114492606A (en) * | 2022-01-13 | 2022-05-13 | 河北工业职业技术学院 | Robot skill transfer learning system based on fuzzy experience and atomic division recombination |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108896052A (en) * | 2018-09-20 | 2018-11-27 | 鲁东大学 | A kind of mobile robot smooth paths planing method under the environment based on DYNAMIC COMPLEX |
| CN110497403A (en) * | 2019-08-05 | 2019-11-26 | 上海大学 | A kind of manipulator motion planning method improving two-way RRT algorithm |
| CN112231489A (en) * | 2020-10-19 | 2021-01-15 | 中国科学技术大学 | Knowledge learning and transferring method and system for epidemic prevention robot |
| CN112231490A (en) * | 2020-10-19 | 2021-01-15 | 中国科学技术大学 | Epidemic prevention robot task characteristic knowledge base construction method, system and medium |
| CN113050640A (en) * | 2021-03-18 | 2021-06-29 | 北京航空航天大学 | Industrial robot path planning method and system based on generation of countermeasure network |
-
2021
- 2021-07-20 CN CN202110821417.4A patent/CN113723750A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108896052A (en) * | 2018-09-20 | 2018-11-27 | 鲁东大学 | A kind of mobile robot smooth paths planing method under the environment based on DYNAMIC COMPLEX |
| CN110497403A (en) * | 2019-08-05 | 2019-11-26 | 上海大学 | A kind of manipulator motion planning method improving two-way RRT algorithm |
| CN112231489A (en) * | 2020-10-19 | 2021-01-15 | 中国科学技术大学 | Knowledge learning and transferring method and system for epidemic prevention robot |
| CN112231490A (en) * | 2020-10-19 | 2021-01-15 | 中国科学技术大学 | Epidemic prevention robot task characteristic knowledge base construction method, system and medium |
| CN113050640A (en) * | 2021-03-18 | 2021-06-29 | 北京航空航天大学 | Industrial robot path planning method and system based on generation of countermeasure network |
Non-Patent Citations (2)
| Title |
|---|
| 宋金泽;戴斌;单恩忠;贺汉根;: "一种改进的RRT路径规划算法" * |
| 成怡;王?;修春波;: "一种改进RRT算法在路径规划中的应用研究" * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114492606A (en) * | 2022-01-13 | 2022-05-13 | 河北工业职业技术学院 | Robot skill transfer learning system based on fuzzy experience and atomic division recombination |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zacharaki et al. | Safety bounds in human robot interaction: A survey | |
| Ingrand et al. | Deliberation for autonomous robots: A survey | |
| Zhou et al. | Fuzzy Petri nets and industrial applications: a review | |
| Szafir et al. | Connecting human-robot interaction and data visualization | |
| Bensaci et al. | STPA and Bowtie risk analysis study for centralized and hierarchical control architectures comparison | |
| CN110398967A (en) | A kind of multirobot collaboration trace information processing method using discretization method | |
| CN112231490A (en) | Epidemic prevention robot task characteristic knowledge base construction method, system and medium | |
| Tsalatsanis et al. | Dynamic task allocation in cooperative robot teams | |
| Huang et al. | Speedup techniques for multiobjective integer programs in designing optimal and structurally simple supervisors of AMS | |
| CN113723750A (en) | Method and system for constructing robot humanoid operation action knowledge base | |
| Giallanza et al. | Occupational health and safety issues in human-robot collaboration: State of the art and open challenges | |
| Tehrani et al. | Review of human-in-the-loop cyber-physical systems (HiLCPS): The current status from human perspective | |
| Sun et al. | Dynamic human systems risk prognosis and control of lifting operations during prefabricated building construction | |
| Askarpour et al. | Formal methods in designing critical cyber-physical systems | |
| Navarro et al. | Temporal bounded reasoning in a dynamic case based planning agent for industrial environments | |
| Tont et al. | Markov approach of adaptive task assignment for robotic system in non-stationary environments | |
| Berger et al. | A generic architecture to design Cyber-Physical and Human Systems | |
| Spector et al. | The Use of Supervenience in Dynamic-world Planning. | |
| Li et al. | Enabling Improved Learning Capability of Industrial Robots with Knowledge Graph Towards Intelligent Digital Twins | |
| Sakthivel et al. | A KBES for integrated engineering | |
| Subramanian et al. | A Study on seamless information sharing between robots in identifying the optimal path: An agent based approach | |
| Ngo et al. | Implementation of the Mathematical Model for Service Robot to Avoid Obstacles and Human | |
| Tont et al. | Fuzzy petri net-based approach in modelling simultaneous task assignment for robotic system | |
| Zhang | Learning human actions on-demand based on graph theory | |
| Ding et al. | VR-based simulation on material handling remote operation for engineering machine |
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 |