CN116985107A - Multi-mechanical-arm space robot on-orbit service task planning method - Google Patents

Abstract

The invention discloses an on-orbit service task planning method for a multi-mechanical-arm space robot, which is suitable for space robots with autonomous decision requirements on-orbit service in a complex space environment and realizes on-orbit service autonomous task planning. Dividing an autonomous task rule into three steps, firstly establishing a multi-mechanical arm space robot kinematic model to obtain the position of each mechanical arm and the pose of an end effector as state information; secondly, a task planner based on a layered task network is designed for the task execution of the robot, and field knowledge and problem description of different tasks are written as the input of the task planner according to different tasks, so that the on-orbit service task is decomposed into executable atomic tasks; and finally, optimally distributing the atomic tasks so that the space robot can maximize the task execution efficiency within the capacity range of the space robot, and completing the on-orbit service. The invention provides support for future space on-orbit service in China.

Description

Multi-mechanical-arm space robot on-orbit service task planning method

Technical Field

The invention belongs to the technical field of intelligent planning of on-orbit service of a spacecraft, and particularly relates to an autonomous task decomposition and planning method of a multi-mechanical-arm on-orbit service space robot.

Background

The spacecraft is high in cost and difficult to maintain after being in orbit, and once the spacecraft fails, huge loss is caused; in addition, most spacecraft are not designed to take into account the need for on-orbit upgrades. However, the on-orbit service operation by astronauts is costly and risky because of the complex and variable and dangerous environment in space, which can create a potential hazard for the life of astronauts. Therefore, the application of the space robot to the space unmanned autonomous on-orbit service has great advantages and benefits.

In the process of replacing or assisting an astronaut to finish various complex on-orbit operation tasks by using a space robot, how to ensure the effectiveness of the space robot task execution process and improve the planning efficiency of the space robot task execution process becomes a problem to be solved when the space operation task is finished by the current space robot. However, the current on-orbit task planning of the multi-mechanical-arm space robot mainly relies on artificial setting of an action sequence thereof to execute an on-orbit task, and the on-orbit service robot also needs to control the movement of the mechanical arm to complete a preset operation in a service process after capturing a target spacecraft, so that the artificial setting mode cannot meet task requirements more and more along with continuous deep space exploration and improvement of the complexity of the mechanical arm operation task. In addition, accurate information of an operation environment is difficult to acquire, resources such as energy sources and computing capacity are limited, and factors such as low space-earth communication bandwidth and long time delay cause that the traditional control mode of ground teleoperation cannot meet real-time requirements, so that new requirements are provided for autonomous decision making capability of the mechanical arm in a complex environment.

Disclosure of Invention

The invention aims to: the invention aims to provide an autonomous task decomposition and planning method for a multi-mechanical arm on-orbit service robot. The task planning method utilizes a hierarchical task network planner to decompose a typical on-orbit service task into atomic tasks which can be executed by an end effector of the mechanical arm, and then the decomposed atomic tasks are completed by the mechanical arm, so that the decomposition and planning of the whole on-orbit service task are completed, and the on-orbit service autonomous task planning requirement of the multi-mechanical-arm space robot aiming at on-orbit service is met.

The technical scheme is as follows: in order to achieve the above purpose, the present invention adopts the following technical scheme;

(1) For a multi-arm space robot, building its kinematic model to obtain a mapping matrix from joint space to working space:

(1) For a multi-arm space robot, its kinematic model is a mapping from joint space to working space:

wherein ,a is a mapping matrix from the joint space of the mechanical arm to the working space between any coordinate system i and an adjacent coordinate system i-1 _i Is along X _i From axis Z _i Move to Z _i+1 Is a distance of (2); alpha _i-1 Is wound around X _i-1 From axis Z _i-1 Rotate to Z _i Is a function of the angle of (2); d, d _i Is along Z _i Axis from X _i-1 Move to X _i Is a distance of (2); θ _i To be around Z _i Axis from X _i-1 Rotate to X _i Is a function of the angle of (2);

(2) Establishing an on-orbit service task planner model based on a layered task network

The hierarchical task network (Hierarchical task network, HTN) planner consists of five sets of definitions: state), task Network (TN), domain description (domain description), planning problem (HTN) node.

Status: state S is a symbolic description of the world state at execution time of the plan, where S ₀ The initial state is the world initial state before planning has not been executed. World state is defined as a set of conjunctive expressions containing true or false binary values, defined as:

S＝p ₁ ^p ₂ ^...^p _n

where p is a predicate expression containing true or false binary values, which determines whether the world state is true or not.

Task network TN: the task network TN is represented by a pair of tuples comprising tasks T and constraints C, wherein the tasks comprise atomic tasks (private tasks) and Compound tasks (Compound tasks). The task network doublet is defined as:

T _N ＝(T，C)

wherein T_N In order to achieve the object, the task network is a task network, wherein T can be an atomic task or a composite task, the composite task is a complex task which can be decomposed into atomic tasks through a decomposition method (method), and the atomic task is a task which cannot be decomposed and can be directly executed by an end effector of a mechanical arm. C specifies the constraints of the current task in the planning process.

Description of the field: the domain knowledge D is defined in HTN as a binary group containing operators and decomposition methods, and is the core of HTN planning. The domain description is defined as follows:

D＝(O，M)

wherein O represents a finite set of operators, is an action template describing an atomic task, and M is a finite set of task decomposition methods. Any operator O e O can be described as the following triplet:

o＝(a，Pre(a)，Eff(a))

where a represents an atomic task that can be performed by an operator, pre (a) represents a precondition required for performing the task, and ef (a) represents an effect on a state space after performing the task. For any decomposition method mε M, the following is described:

m＝(T _c ，Pre(T _c )，Sub(T _c ))

wherein T_c For the composite task to be decomposed, pre (T _c ) Representing preconditions required for decomposing the task, sub (T _c ) Representing subtasks generated after the composite task is decomposed, wherein the tasks can be composite tasks or atomic tasks.

HTN programming problem: a problem with HTN planning is a four-tuple comprising an initial state space, an initial task network, domain knowledge, and a target state:

P _HTN ＝(S ₀ ，TN ₀ ，D，G)

wherein ,P_HIN For HTN planning problem, S ₀ In an initial state, TN ₀ For the initial task network, D is the domain description and G is the target state. Planning problem P for any HTN _HTN If the solution exists, an action sequence is used as a solution of the problem, and the action sequence is the solution of the planning problem, so that the tasks in all task networks can be completed.

HTN task node: the HTN task node is composed of child nodes, a prize value R at the current node, a node entry number N, and an accumulated prize value Q. HTN nodes are defined as a five-tuple:

node＝(n_c，n_p，N，R，Q)

the node is a current task node, n_c is used for storing child nodes, n_p is used for storing parent nodes, and if the current node is a root node, the node is empty. N is the number of node entries, R is the immediate prize value at the node, and Q is the cumulative prize value at the node. The node entering times N are used for calculating UCT values, wherein the UCT values are the basis for node selection in the node selection stage to enter the simulation stage, and are defined as follows:

wherein Q_max (n_c) is the maximum value of the cumulative prize value of all the child nodes under the node n_c, N (node) is the access times of the parent node of the node n_c, N (n_c) is the access times of the node n_c, and c is the search factor, which is a constant.

2. The method for planning on-orbit service tasks of the multi-mechanical arm space robot according to claim 1, which mainly comprises the following parts:

multi-mechanical arm space robot: an atomic task sequence or an action sequence and the like which are responsible for executing the output of the HTN planner;

HTN task planner: a planner composed of various decomposition methods, operators and corresponding domain description files;

3. the on-orbit service mission planning method of a multi-arm space robot according to claim 1, comprising the steps of:

(3-1)target planning problem input, current state information S ₀ Input and task network TN input;

(3-2) extracting an initial task t from the task network TN, taking the task as a root task node, and recording the node times N (t).

(3-3) judging whether the task t is an atomic task;

and (3-4) if yes, selecting any operator O epsilon O meeting the current precondition to be executed by the mechanical arm, updating the current state, and deleting the atomic task t from the task network TN. If not, entering the next step;

(3-5) node extension: selecting any decomposition method meeting the current precondition of the task t to decompose, taking the obtained subtask t 'as a child node of the task t, and recording the node entering times N (t');

(3-6) judging whether t' is an atomic task, if not, entering a simulation stage: randomly selecting a decomposable method t 'to decompose the method t' and giving a prize value until the decomposition is stopped for an atomic task (3-7); if it is (3-7);

(3-7) backtracking all the passed task nodes from the atomic task node t' to the task node t where the task node t is currently located, and accumulating the obtained reward value to obtain a Q value;

(3-8) judging whether all the decomposition methods of the task node t are executed, if so, entering (3-9), otherwise, entering (3-5);

(3-9) the current node has no extensible task node, at which point the selection phase is entered: calculating UCT values of all the expansion task nodes t', arranging the UCT values according to descending order, selecting the task node with the maximum value as the current task node, replacing the task t in the task network TN, and returning to (3-2);

and (3-10) judging whether the task network TN is empty, if so, planning to complete and output an action sequence, otherwise, reselecting the task in the task network TN.

The beneficial effects are that: compared with the existing task planning scheme, the autonomous task decomposition and planning method of the multi-mechanical arm space robot is based on a kinematic model of the multi-mechanical arm space robot, and the task planning process is subdivided into three steps: question input, task decomposition, and task execution. The method and the system realize the effectiveness of the task execution process of the space robot, can complete the autonomous decomposition and planning of the space robot, get rid of an on-orbit service mode of manually setting an action sequence to execute on-orbit tasks, and enable the space robot with multiple mechanical arms to execute the autonomous tasks in a complex environment. At present, the on-orbit service of the multi-arm space robot is concentrated on the planning of on-orbit resources at home and abroad, and the multi-arm on-orbit service task planning research of the multi-arm on-orbit service robot is not carried out yet.

Drawings

FIG. 1 is a functional architecture diagram of the present invention;

FIG. 2 is a workflow diagram of the present invention;

Detailed Description

The invention is further described below with reference to the accompanying drawings.

As shown in fig. 1, the functional architecture of the present invention is first composed of two major parts: a multi-arm space robot and an HTN mission planner. And (3) establishing a kinematic model of the space robot to obtain a mapping from a joint space to a working space, and establishing a coordinate system conversion matrix of adjacent joints of the mechanical arm required by kinematics to prepare for executing a subsequently solved action sequence. The HTN task planner performs task decomposition and planning on the mechanical arm to output a motion sequence which can be directly executed by the mechanical arm. The input of the task planner is the field description and the problem description of the task to be planned, an initial task network is established for the task, the task is decomposed by means of a given decomposition method to obtain an atomic task, the task is added into the task network, the initial task is deleted from the task network, and the task planning scheme action sequence is output to be executed by the mechanical arm until the task is decomposed into an atomic task which cannot be decomposed.

As shown in fig. 2, the working steps of the invention are as follows:

(1) Target planning problem input, current state information S ₀ Input and task network TN input;

(2) Extracting an initial task t from a task network TN, taking the task as a root task node, recording the node times N (t), and judging whether the task t is an atomic task or not;

(3) If yes, any operator O epsilon O meeting the current precondition is selected to be executed by the mechanical arm, the current state is updated, and the atomic task t is deleted from the task network TN. If not, entering the next step;

(4) Node expansion: selecting any decomposition method meeting the current precondition of the task t to decompose, taking the obtained subtask t 'as a child node of the task t, and recording the node entering times N (t');

(5) Judging whether t' is an atomic task, if not, entering a simulation stage: randomly selecting a decomposable method t 'to decompose the method t' and giving a reward value until the decomposition into atomic tasks stops, and entering (6); if so, directly entering (6);

(6) Backtracking all the passed task nodes from the atomic task node t' until the current task node t, and accumulating the obtained reward value to obtain a Q value;

(7) Judging whether all the decomposition methods of the task node t are executed, if so, entering (8), otherwise, entering (4);

(8) The current node has no extensible task node, and the selection phase is entered at this time: calculating UCT values of all the expansion task nodes t', selecting the task node with the maximum value to replace the task t in the task network TN, and returning to the step (2);

(9) And judging whether the task network TN is empty, if so, planning to complete and output an action sequence, otherwise, reselecting the task in the task network TN.

Claims (3)

1. An on-orbit service task planning method for a multi-mechanical arm space robot is characterized by comprising the following steps:

(1) For a multi-arm space robot, its kinematic model is a mapping from joint space to working space:

wherein ,a is a mapping matrix from the joint space of the mechanical arm to the working space between any coordinate system i and an adjacent coordinate system i-1 _i Is along X _i From axis Z _i Move to Z _i+1 Is a distance of (2); alpha _i-1 Is wound around X _i-1 From axis Z _i-1 Rotate to Z _i Is a function of the angle of (2); d, d _i Is along Z _i Axis from X _i-1 Move to X _i Is a distance of (2); θ _i To be around Z _i Axis from X _i-1 Rotate to X _i Is a function of the angle of (2);

(2) Establishing an on-orbit service task planner model based on a layered task network

The hierarchical task network (Hierarchical task network, HTN) planner consists of five sets of definitions: state), task Network (TN), domain description (domain description), planning problem (HTN) node.

Status: state S is a symbolic description of the world state at execution time of the plan, where S ₀ The initial state is the world initial state before planning has not been executed. World state is defined as a set of conjunctive expressions containing true or false binary values, defined as:

S＝p ₁ ∧p ₂ ∧...∧p _n

where p is a predicate expression containing true or false binary values, which determines whether the world state is true or not.

Task network TN: the task network TN is represented by a pair of tuples comprising tasks T and constraints C, wherein the tasks comprise atomic tasks (private tasks) and Compound tasks (Compound tasks). The task network doublet is defined as:

T _N ＝(T，C)

wherein T_N In order to achieve the object, the task network is a task network, wherein T can be an atomic task or a composite task, the composite task is a complex task which can be decomposed into atomic tasks through a decomposition method (method), and the atomic task is a task which cannot be decomposed and can be directly executed by an end effector of a mechanical arm. C designates the current task in the planning processIs a constraint of (a).

Description of the field: the domain knowledge D is defined in HTN as a binary group containing operators and decomposition methods, and is the core of HTN planning. The domain description is defined as follows:

D＝(O，M)

wherein O represents a finite set of operators, is an action template describing an atomic task, and M is a finite set of task decomposition methods. Any operator O e O can be described as the following triplet:

o＝(a，Pre(a)，Eff(a))

where a represents an atomic task that can be performed by an operator, pre (a) represents a precondition required for performing the task, and ef (a) represents an effect on a state space after performing the task. For any decomposition method mε M, the following is described:

m＝(T _c ，Pre(T _c )，Sub(T _c ))

wherein T_c For the composite task to be decomposed, pre (T _c ) Representing preconditions required for decomposing the task, sub (T _c ) Representing subtasks generated after the composite task is decomposed, wherein the tasks can be composite tasks or atomic tasks.

HTN programming problem: a problem with HTN planning is a four-tuple comprising an initial state space, an initial task network, domain knowledge, and a target state:

P _HTN ＝(S ₀ ，TN ₀ ，D，G)

wherein ,P_HTN For HTN planning problem, S ₀ In an initial state, TN ₀ For the initial task network, D is the domain description and G is the target state. Planning problem P for any HTN _HTN If the solution exists, an action sequence is used as a solution of the problem, and the action sequence is the solution of the planning problem, so that the tasks in all task networks can be completed.

HTN task node: the HTN task node is composed of child nodes, a prize value R at the current node, a node entry number N, and an accumulated prize value Q. HTN nodes are defined as a five-tuple:

node＝(n_c，n_p，N，R，Q)

the node is a current task node, n_c is used for storing child nodes, n_p is used for storing parent nodes, and if the current node is a root node, the node is empty. N is the number of node entries, R is the immediate prize value at the node, and Q is the cumulative prize value at the node. The node entering times N are used for calculating UCT values, wherein the UCT values are the basis for node selection in the node selection stage to enter the simulation stage, and are defined as follows:

wherein Q_max (n_c) is the maximum value of the cumulative prize value of all the child nodes under the node n_c, N (node) is the access times of the parent node of the node n_c, N (n_c) is the access times of the node n_c, and c is the search factor, which is a constant.

2. The method for planning on-orbit service tasks of the multi-mechanical arm space robot according to claim 1, which mainly comprises the following parts:

multi-mechanical arm space robot: an atomic task sequence or an action sequence and the like which are responsible for executing the output of the HTN planner;

HTN task planner: the planner consists of various decomposition methods, operators and corresponding domain description files.

3. The on-orbit service mission planning method of a multi-arm space robot according to claim 1, comprising the steps of:

(3-1) target planning problem input, current State information S ₀ Input and task network TN input;

(3-2) extracting an initial task t from the task network TN, taking the task as a root task node, and recording the node times N (t).

(3-3) judging whether the task t is an atomic task;

and (3-4) if yes, selecting any operator O epsilon O meeting the current precondition to be executed by the mechanical arm, updating the current state, and deleting the atomic task t from the task network TN. If not, entering the next step;

(3-5) node extension: selecting any decomposition method meeting the current precondition of the task t to decompose, taking the obtained subtask t 'as a child node of the task t, and recording the node entering times N (t');

(3-6) judging whether t' is an atomic task, if not, entering a simulation stage: randomly selecting a decomposable method t 'to decompose the method t' and giving a prize value until the decomposition is stopped for an atomic task (3-7); if it is (3-7);

(3-7) backtracking all the passed task nodes from the atomic task node t' to the task node t where the task node t is currently located, and accumulating the obtained reward value to obtain a Q value;

(3-8) judging whether all the decomposition methods of the task node t are executed, if so, entering (3-9), otherwise, entering (3-5);

(3-9) the current node has no extensible task node, at which point the selection phase is entered: calculating UCT values of all the expansion task nodes t', arranging the UCT values according to descending order, selecting the task node with the maximum value as the current task node, replacing the task t in the task network TN, and returning to (3-2);

and (3-10) judging whether the task network TN is empty, if so, planning to complete and output an action sequence, otherwise, reselecting the task in the task network TN.