A Fast Forward Search Task Planning Method for Space Robot Assisted Operations

Abstract

This invention relates to a fast forward search task planning method for space robot assisted operations, belonging to the field of task planning. By modifying pruning and expansion strategies, the fast forward search task planning method is introduced into the field of space assisted operation tasks. For space assisted operation tasks where operation actions are the primary focus, this invention changes the state search to an action search. By introducing an operation cost method, actions in the action search process are sorted, and the action sequence with the minimum operation cost is selected. Compared with the initial version of the fast forward search planning method, this invention incorporates the research domain and introduces cost calculation from the current state to available actions, solving the problem of action selection in space assisted operation tasks.

Description

Technical Field

[0001] This invention belongs to the field of task planning and relates to a typical task planning method for space robot assisted operation tasks. Background Technology

[0002] Currently, commonly used task planning algorithms in the field of space robot assisted operation include graph algorithms and hierarchical network algorithms. Graph algorithms suffer from low computational efficiency due to the large number of mutually exclusive actions and propositions involved, often requiring significant computation time for a specific space assisted operation task. Hierarchical network algorithms require prior expert knowledge and a decomposition method for the planning process, resulting in insufficient autonomy.

[0003] Fast Forward Search (FAST) is a highly efficient heuristic search planning method. Its efficiency is primarily based on the following key techniques: 1) constructing a relaxed planning graph for calculating heuristic estimates of states; 2) employing an enhanced hill-climbing algorithm as the fast forward heuristic search algorithm; and 3) defining favorable actions as an effective pruning strategy. This planning method won the Best Performance Award in the International Intelligent Planning Competition. Based on the ideas of relaxed planning heuristics and forward local search, it has been widely used to solve various complex planning problems.

[0004] Space robots require numerous movements, including movement, grasping, and releasing, to assist in space operations. For example, in the US RRM3 project, a space robot was needed to assist in refueling. This required the robot to first open its protective shield, then unscrew the cap, and enter the delivery valve to refuel. This process involved multiple actions and the switching of corresponding tools.

[0005] To autonomously generate action sequences for space robot assisted operation tasks, this invention employs a highly efficient fast forward search planning method. After applying this planning method to the field of space robot refueling, it was found that space robot assisted operation tasks focus more on action selection within a given state. Previous depth-first search algorithms, which were not highly targeted to states, often resulted in overlapping preconditions between actions. Therefore, this invention considers changing the search method in the fast forward search planning method for space robot assisted operation tasks from state search to action search. The resulting action selection problem cannot be solved using previous methods that favor advantageous actions, because these methods did not consider the execution cost of the space robot performing advantageous actions. Summary of the Invention

[0006] The technical problem to be solved by this invention is:

[0007] To avoid the shortcomings of existing technologies, this invention provides a fast forward search task planning method for space robot assisted operation. By modifying the pruning and expansion strategies, the difficulty of the action operation is introduced as a cost into the solution of the action sequence for a certain task, so that the whole method can obtain a set of action execution sequences that meet the minimum constraint of operation difficulty.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] Because of the high efficiency of the Fast Forward Search (FFS) algorithm in intelligent planning, this invention aims to use FFS to solve the task planning problem in space-assisted operation tasks. However, space robot assisted operation tasks focus more on the transitions between actions and minimizing the pose disturbance of the space robot during these transitions. The FFS-based state search does not meet the requirements of space-assisted operation tasks. Therefore, this invention addresses the action sequence solution requirement of space-assisted operation tasks by changing the state search to an action search. It addresses the action selection problem arising from the action search, designs a corresponding cost function to evaluate the operational difficulty of each action, and selects the action with the lowest operational difficulty. Finally, it obtains an action sequence that satisfies the constraints of the space-assisted operation task.

[0010] A fast forward search task planning method for space robot assisted operation is characterized by the following steps:

[0011] Step 1: Establish a mathematical model for the space-assisted operation task, define the preconditions and effects of each action, and the desired state s of the space-assisted operation task. goal The action is represented as shown in equation (1), and the effect of the action is shown in equation (2):

[0012] a=[name(a) precondition(a) effect(a)] (1)

[0013] γ(sa)=(s-effect - (a))∪effect + (a) (2)

[0014] Where 'a' represents the action, 'name' represents the name of action 'a', 'precondition' represents the prerequisite for action 'a', and 'effect' represents the effect of action 'a'. The effect of action 'a' can be represented as adding an effect. + (a) and deletion effect - (a), γ(sa) represents the effect of the action a on state s, specifically manifested as state s, effect -(a) intersection and effect + The union of (a);

[0015] Step 2: Design the cost function for operational difficulty:

[0016] f(a)=g(a)+h(a) (3)

[0017] Where f(a) represents the total cost of determining action a; g(a) represents the cost of executing action a from the current state s, including the cost of the robot going from the current state to the toolbox to change tools and the cost of pose perturbation along the way. This formula indicates that the task always completes all tasks that the current tool can complete first; h(a) represents the cost from action a to the desired state. This cost is calculated by the relaxed planning graph algorithm, specifically the number of actions m obtained by the relaxed planning graph algorithm. This formula indicates that the action closest to the desired state is always obtained.

[0018] Step 3: Expand the planning map

[0019] Based on the action model established in step 1, the graph planning algorithm is used to expand from the initial state to the desired state, obtaining the available actions A(s) whose preconditions belong to the current state. i ), that is, precondition∈s i Available actions A(s) i );

[0020] Step 4: Select the action with the lowest cost, a * ;

[0021] For the set of available actions A(s) obtained in step 3 i Using formula (3), calculate the cost f(a) for each action a, and obtain the action a corresponding to min{f(a)}. * ;

[0022] Step 5: Use action a * Update world state s i ; in the selected action a * Then, action a * The effect is added to world states. i Above, a new world state s is obtained. i+1 ;

[0023] Step 6: Repeat steps 3-5 until a new world state s is reached. i+1 Includes expected state s goal Thus concludes the task planning method for space-assisted operations based on the fast forward search planning method, ultimately yielding the action sequence with the minimum cost.

[0024] A computer system is characterized by comprising: one or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method described above.

[0025] A computer-readable storage medium is characterized by storing computer-executable instructions, which, when executed, are used to implement the above-described method.

[0026] The beneficial effects of this invention are as follows:

[0027] This invention provides a fast forward search task planning method for space-assisted operations tasks. By modifying pruning and expansion strategies, the fast forward search task planning method is introduced into the space-assisted operations task domain. For space-assisted operations tasks primarily involving operations, this invention changes state search to action search. By introducing an operation cost method, actions in the action search process are sorted, and the action sequence with the minimum operation cost is selected. Compared with the initial version of the fast forward search planning method, this invention incorporates the domain under study and introduces a method from the current state to the available action A(s). i The cost calculation between A(s) and A(s) solved the problem. i This paper addresses the problem of action selection in space-assisted manipulation tasks. Simulation results demonstrate that this method can quickly solve for the action sequence of space-assisted manipulation tasks, demonstrating significant application value in future applications. Attached Figure Description

[0028] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0029] Figure 1 Flowchart of a fast forward search task planning method for space-assisted operations tasks;

[0030] Figure 2 Complex action sequences obtained based on a fast forward search task planning method;

[0031] Figure 3 The complete action sequence is obtained based on the fast forward search task planning method;

[0032] Figure 4 The complete action sequence is obtained based on the improved fast forward search task planning method;

[0033] Figure 5 A list of spatial auxiliary operation tasks based on PDDL description. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0035] This invention presents a fast forward search task planning method for space-assisted operation tasks. Based on the fast forward search planning method, improvements have been made specifically for space-assisted operation tasks. These improvements include: replacing state search with action search and introducing action selection with operational difficulty costs, resulting in a planning method applicable to the field of space robot assisted operation. Simulation verification demonstrates that this method can quickly and efficiently obtain the action sequence for space-assisted operation tasks. Specific implementation details are as follows:

[0036] A fast forward search task planning method for space-assisted operation tasks includes:

[0037] Step 1: Establish a mathematical model for space-assisted operation tasks

[0038] This invention uses PDDL (Planning Domain Definition Language) to model actions in space-assisted operation tasks, and sets the preconditions and effects of each action, as well as the expected state s of the space-assisted operation task. goal Using the above information, the domain file and problem file for the task planning problem are determined; the actions can be represented as shown in equation (1), and the effects of the actions are shown in equation (2):

[0039] a=[name(a) precondition(a) effect(a)] (1)

[0040] γ(sa)=(s-effect - (a))∪effect + (a) (2)

[0041] Where 'a' represents the action, 'name' represents the name of action 'a', 'precondition' represents the prerequisite for action 'a', and 'effect' represents the effect of action 'a'. The effect of action 'a' can be represented as adding an effect. + (a) and deletion effect - (a), γ(sa) represents the effect of the action a on state s, specifically manifested as state s, effect - (a) intersection and effect+ The union of (a);

[0042] Step 2: Design the cost function for operational difficulty. To facilitate the measurement of the operational difficulty of each action, this invention divides actions into two types: complex actions (e.g., disassembly, refueling, etc.) and ordinary actions (e.g., moving, grasping, and releasing). In the fast forward search planning method, the operational difficulty of an action is described by the cost function as shown in equation (3):

[0043] f(a)=g(a)+h(a) (3)

[0044] Where f(a) represents the total cost of determining action a; g(a) represents the cost of executing action a from the current state s, including the cost of the robot going from the current state to the toolbox to change tools and the cost of pose perturbation along the way. This formula indicates that the task always completes all tasks that the current tool can complete first; h(a) represents the cost from action a to the desired state. This cost is calculated by the relaxed planning graph algorithm, specifically the number of actions m obtained by the relaxed planning graph algorithm. This formula indicates that the action closest to the desired state is always obtained.

[0045] Step 3: Expand the planning graph. Based on the action model established in Step 1, use the planning graph algorithm to expand from the initial state to the desired state, obtaining the available actions A(s) whose preconditions belong to the current state. i ), that is, precondition∈s i Available actions A(s) i );

[0046] Step 4: Select the action with the lowest cost, a * For the set of available actions A(s) obtained in step 3 i Using formula (3), calculate the cost f(a) for each action a, and obtain the action a corresponding to min{f(a)}. * ;

[0047] Step 5: Use action a * Update world state s i ; in the selected action a * Then, action a * The effect is added to world states. i Above, a new world state s is obtained. i+1 ;

[0048] Step 6: Repeat steps 3-5 until a new world state s is reached. i+1 Includes expected state s goal Thus concludes the task planning method for space-assisted operations based on the fast forward search planning method, ultimately yielding the action sequence with the minimum cost.

[0049] Example 1:

[0050] Step 1: Establish a mathematical model for the space-assisted operation mission. In the specific implementation process, for the US RRM3 (Robotic Refueling Mission 3) project, the space-assisted operation mission background was designed as follows: Before the satellite leaves the ground, technicians fill the fuel tank through a valve, then triple-seal the valve and cover it with a protective device. Therefore, in the specific implementation, the following four complex actions were specifically set: screwing, uncovering, inserting, and refueling, and simple movement actions were used to connect the above complex actions. Subsequently, the specific movement, grasping, and release sequences were obtained by decomposing the movement actions. The action model is defined according to formula (1) and formula (2), and the definition method of the action model is shown in formula (4):

[0051]

[0052] The action model mainly consists of four parts: actionName represents the name of the action; parameters represent all parameters related to the action; and so on. i -tp represents the i-th parameter. i The type of the i-th parameter is indicated; precondition indicates the precondition of the action, which can only be executed when all preconditionStates are true; effect indicates the effect after the action is executed, which includes all effectStates.

[0053] The domain and problem files described above actions are represented using PDDL. The action models designed in the files are as follows: Figure 5 As shown.

[0054] Step 2: Construct a heuristic function. For the action models established in Step 1, construct a heuristic function to calculate the cost of each action model. For the above space-assisted operation task, multiple actions are involved and the tools for each action are different. Therefore, the movement cost and tool switching cost g(a) need to be considered in the task planning process. The calculation methods for this cost are: the satellite pose disturbance caused by the movement of the space robot and the additional cost c for each tool switch. In addition, the distance h(a) between each action and the desired state is also considered in the task planning process. This distance is calculated by the relaxed planning graph algorithm. Specifically, it is the number of actions required to get from the current action to the desired state, as shown in Equation (6). The purpose is to select the action with the minimum number of actions from the current action to the desired state. In summary, the final cost function is constructed as shown in Equation (5):

[0055] f(a)=g(a)+h(a) (5)

[0056] in,

[0057]

[0058]

[0059] Where α, β, and γ represent the pitch, yaw, and roll angles of the space robot's end effector, respectively. e (i) represents the i-th angle of the end effector of the space robot, x, y, and z represent the positions of the end effector of the space robot, and pos represents the position of the end effector of the space robot. e (i) represents the i-th position of the end effector of the space robot, c represents the cost of each tool switch (taken as 3 in this study), n represents the number of tool switches, and O represents the distance from action a to the desired state s. goal The total number of actions required, h0(a) is initialized to ∞.

[0060] Step 3: Expand the relaxed planning graph. Compared to the planning graph algorithm, the relaxed planning graph algorithm ignores the mutual exclusion relationships between actions. Starting from the initial state, it expands layer by layer through forward search, obtaining the state s of each layer. i Then we obtain the state s where the preconditions are met. i The set of available actions A(s) i As shown in equation (8):

[0061] A(s i )={a|precondition(a)∈s i} (8)

[0062] Step 4: Select the action with the lowest cost, a * The action set A(s) obtained by traversing equation (8) i Using formula (5), the cost value of each action is calculated, and the action a with the smallest cost value is selected. * And discard other actions, and put a * Record it into the expected action sequence π.

[0063] Step 5: Use action a * Update status s i Update the current state s using formula (2). i , and obtain the new state s i+1 .

[0064] Step 6: Repeat steps 3 through 5 until a new world state s is reached. i+1 Includes the desired state. If a certain state s occurs...i+1 The corresponding action set A(s) i+1 The cost of all actions in ) is greater than that of state s. i The corresponding action set A(s) i If the cost of all actions in state s is not met, the task planning is declared a failure; otherwise, it continues. This applies when the desired state exists in state s. i The plan was announced to be completed, and a solution to the plan had been found.

[0065] Finally, simulations verified the proposed fast forward search task planning method for space-assisted operations. Simulations were performed on a space refueling task within space-assisted operations. The simulations yielded the action sequence with the lowest operational difficulty, while more complex action sequences were obtained as follows: Figure 2 As shown. Furthermore, based on the hierarchical concept, the action sequence ultimately decomposed using the basic fast forward search algorithm is as follows: Figure 3 As shown, for actions without a specific order requirement, such as butt1 and butt2, the lack of a cost function evaluation leads to a random action sequence in the task planning, failing to meet the constraint of minimizing energy during space robot assisted operation. The action sequence finally decomposed using the improved fast forward search algorithm is shown below. Figure 4 As shown, the obtained action sequence satisfies the constraint of minimum energy, and the order of some actions is determined by the magnitude of the cost value.

[0066] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention.

Claims

1. A fast forward search task planning method for space robot assisted operation, characterized in that... The steps are as follows: Step 1: Establish a mathematical model for the space-assisted operation task, define the preconditions and effects of each action, and the desired state s of the space-assisted operation task. goal The action is represented as shown in equation (1), and the effect of the action is shown in equation (2): a=[name(a) precondition(a) effect(a)] (1) γ(s a)=(s-effect - (a))∪effect + (a) (2) Where 'a' represents the action, 'name' represents the name of action 'a', 'precondition' represents the prerequisite for action 'a', and 'effect' represents the effect of action 'a'. The effect of action 'a' can be represented as adding an effect. + (a) and deletion effect - (a), γ(sa) represents the effect of the action a on state s, specifically manifested as state s and effect. - (a) intersection and effect + The union of (a); The action model is defined according to formulas (1) and (2), and the definition method of the action model is shown in formula (3): The action model mainly consists of four parts: actionName represents the name of the action; parameters represent all parameters related to the action; and so on. i -tp represents the i-th parameter. i The i-th parameter represents the type of the i-th parameter; precondition represents the precondition of the action, which can only be executed if all preconditionState values ​​are true; effect represents the effect after the action is executed, which includes all effectState values. Step 2: Design the cost function for operational difficulty: f(a)=g(a)+h(a) (4) in, Where f(a) represents the total cost of formulating action a; g(a) represents the cost of executing action a from the current state s, including the cost of the robot moving from the current state to the toolbox to change tools and the cost of pose perturbations along the way, g(a) means that all tasks that the current tool can perform are always completed first; h(a) represents the cost from action a to the desired state, which is calculated by the relaxed planning graph algorithm, specifically the number of actions m obtained by the relaxed planning graph algorithm, h(a) means that the action closest to the desired state is always obtained; α, β, and γ represent the pitch, yaw, and roll angles of the space robot end effector, respectively. e (i) represents the i-th angle of the end effector of the space robot, x, y, and z represent the positions of the end effector of the space robot, and pos represents the position of the end effector of the space robot. e (i) represents the i-th position of the end effector of the space robot, c represents the cost of switching tools each time, n represents the number of times the tool is switched, and h0(a) is initialized to ∞. Step 3: Expand the relaxed planning graph Based on the action model established in step 1, the graph planning algorithm is used to expand from the initial state to the desired state, obtaining the available actions A(s) whose preconditions belong to the current state. i ), that is, precondition∈s i Available actions A(s) i ); Step 4: Select the action with the lowest cost, a * ; For the set of available actions A(s) obtained in step 3 i Using formula (4), calculate the cost f(a) for each action a, and obtain the action a corresponding to min{f(a)}. * ; Step 5: Use action a * Update world state s i ; in the selected action a * Then, action a * The effect is added to world states. i Above, a new world state s is obtained. i+1 ; Step 6: Repeat steps 3-5 until a new world state s is reached. i+1 Includes expected state s goal Thus concludes the task planning method for space-assisted operations based on the fast forward search planning method, ultimately yielding the action sequence with the minimum cost.

2. A computer system, characterized in that... include: One or more processors, a computer-readable storage medium for storing one or more programs, wherein, when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method of claim 1.

3. A computer-readable storage medium, characterized in that... The device stores computer-executable instructions, which, when executed, are used to implement the method of claim 1.

Images