Motion planning method and device of intelligent agent and intelligent agent

Abstract

The invention discloses an intelligent agent motion planning method and device and an intelligent agent, and belongs to the field of intelligent agents. The motion planning method of the intelligent agent comprises the following steps: predicting to obtain a pose set corresponding to each key point position included in an initial motion sequence based on environment information in a target scene; based on a pose set corresponding to a target adjacent point location in the plurality of key point locations, obtaining a plurality of candidate sub-motion paths between the target adjacent point locations; and processing the plurality of sub-candidate motion paths corresponding to the adjacent point locations based on the environment information, and determining a target motion plan. According to the motion planning method of the intelligent agent, the reasonability of path planning and the precision and accuracy of obstacle avoidance are improved, and the motion planning method is suitable for any scene and any task, and is particularly suitable for a long task sequence execution scene in a narrow scene.

Description

Technical Field

[0001] The present application belongs to the field of intelligent agents, and in particular, relates to a motion planning method, device and intelligent agent. Background Art

[0002] With the development of science and technology, intelligent agents are widely used in people's daily work and life. In related technologies, sensor data is mainly collected by sensors set on the intelligent agent, so as to control the intelligent agent to move or grasp based on the sensor data. However, this method can only control the intelligent agent to perform relatively simple operations, such as grasping or placing, and cannot be applied to complex environments and the execution of complex tasks. For example, when executing long task sequences, the control accuracy is low and the versatility is poor, which affects the user experience. Summary of the invention

[0003] The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application proposes a motion planning method, device and agent for an intelligent agent, which improves the rationality of path planning and the precision and accuracy of obstacle avoidance, and is applicable to any scene and any task, especially to the execution scenario of a long task sequence in a narrow scene.

[0004] In a first aspect, the present application provides a motion planning method for an intelligent agent, the method comprising:

[0005] Based on the environmental information of the target scene, a posture set corresponding to each key point included in the initial action sequence is predicted; the key point corresponds to a target action state, and the posture set includes at least one first posture information, and the first posture information is used to characterize the posture state of the intelligent agent after executing the target action state corresponding to the key point;

[0006] Based on a set of poses corresponding to adjacent target points in a plurality of key points, a plurality of candidate sub-motion paths between adjacent target points are obtained;

[0007] Based on the environmental information, multiple candidate sub-motion paths corresponding to each adjacent point are processed to determine the target motion plan.

[0008] According to the motion planning method of the intelligent agent of the present application, the posture set corresponding to each key point in the initial action sequence is obtained by prediction, and multiple candidate sub-motion paths between adjacent points are obtained based on the posture set. The feasibility of the multiple candidate sub-motion paths is judged to screen and obtain the target motion plan. The path prediction can be refined to ensure that it is collision-free and achievable at each position, thereby improving the rationality of path planning and the precision and accuracy of obstacle avoidance. It is suitable for any scenario and any task, and is especially suitable for the execution scenarios of long task sequences in narrow scenarios.

[0009] According to an embodiment of the present application, the step of processing a plurality of candidate sub-motion paths corresponding to each adjacent point based on the environmental information to determine the target motion plan includes:

[0010] Based on the environmental information, a target sub-motion path is obtained by screening a plurality of candidate sub-motion paths corresponding to the target adjacent point in the plurality of adjacent points;

[0011] The target sub-motion paths corresponding to each adjacent point are integrated to obtain the target motion path;

[0012] The target motion plan is determined based on the target motion path and the first pose information corresponding to each key point in the target motion path.

[0013] According to an embodiment of the present application, the step of selecting a target sub-motion path from a plurality of candidate sub-motion paths corresponding to a target adjacent point in a plurality of adjacent points based on the environmental information includes:

[0014] Perform feasibility detection on a plurality of candidate sub-motion paths corresponding to the target adjacent points based on the environmental information, and obtain a realization state corresponding to each of the candidate sub-motion paths, wherein the realization state includes a feasible state and an infeasible state;

[0015] The target sub-motion path is determined based on the candidate sub-motion paths corresponding to the feasible state.

[0016] According to an embodiment of the present application, determining the target sub-motion path based on the candidate sub-motion path corresponding to the feasible state includes:

[0017] Predicting the motion input cost corresponding to the execution of each candidate sub-motion path corresponding to each feasible state by the agent;

[0018] The candidate sub-motion path corresponding to the minimum motion input cost is determined as the target sub-motion path.

[0019] According to an embodiment of the present application, the feasibility detection of multiple sub-candidate motion paths corresponding to the target adjacent points based on the environmental information to obtain the realization status corresponding to each of the sub-candidate motion paths includes:

[0020] Based on the environmental information, generating an obstacle-free area between adjacent target points;

[0021] In a case where the boundary of the intelligent body does not exceed the boundary of the obstacle-free area, determining that the realization state corresponding to the sub-candidate motion path is a feasible state;

[0022] In the case where the boundary of the intelligent body exceeds the boundary of the obstacle-free area, it is determined that the realization state corresponding to the sub-candidate motion path is an infeasible state.

[0023] According to an embodiment of the present application, the step of acquiring multiple candidate sub-motion paths between adjacent target points based on a set of poses corresponding to adjacent target points in multiple key points includes:

[0024] Randomly combining the first pose information in the pose set corresponding to the adjacent points of the target to obtain multiple pose combinations;

[0025] The first pose information in each of the pose combinations is connected to generate a candidate motion path corresponding to the pose combination, so as to obtain the multiple candidate motion paths.

[0026] According to an embodiment of the present application, the step of predicting a set of postures corresponding to key points included in the initial action sequence based on the environmental information of the target scene includes:

[0027] Based on the target action state corresponding to the key point, predict multiple candidate posture information corresponding to the key point;

[0028] A collision analysis is performed on the multiple candidate pose information based on the environmental information, and the at least one first pose information is obtained by screening out from the multiple candidate pose information.

[0029] According to an embodiment of the present application, the predicting of multiple candidate pose information corresponding to the key points based on the target action state corresponding to the key points includes:

[0030] Based on the target action state corresponding to the key point, predict a plurality of second posture information corresponding to the key point;

[0031] Clustering is performed on the multiple second pose information to obtain multiple candidate pose information corresponding to the key points.

[0032] According to an embodiment of the present application, before predicting a set of postures corresponding to key points included in the initial action sequence based on the environmental information of the target scene, the method further includes:

[0033] Based on the environment information and the target task information, an initial action sequence corresponding to the agent is determined.

[0034] In a second aspect, the present application provides a motion planning device for an intelligent agent, the device comprising:

[0035] A first processing module is used to predict a posture set corresponding to each key point included in the initial action sequence based on environmental information in the target scene; the key point corresponds to a target action state, and the posture set includes at least one first posture information, and the first posture information is used to represent the posture state of the intelligent agent after executing the target action state corresponding to the key point;

[0036] A second processing module is used to obtain multiple candidate sub-motion paths between adjacent target points based on a set of poses corresponding to adjacent target points in multiple key points;

[0037] The third processing module is used to process multiple sub-candidate motion paths corresponding to each adjacent point based on the environmental information to determine the target motion plan.

[0038] According to the motion planning device of the intelligent body of the present application, a set of postures corresponding to each key point in the initial action sequence is obtained by prediction, and multiple candidate sub-motion paths between adjacent points are obtained based on the posture set. The feasibility of the multiple candidate sub-motion paths is judged to screen and obtain the target motion plan. The path prediction can be refined to ensure that it is collision-free and achievable at each position, thereby improving the rationality of path planning and the precision and accuracy of obstacle avoidance. It is suitable for any scenario and any task, and is particularly suitable for the execution of long task sequences in narrow scenarios.

[0039] In a third aspect, the present application provides an intelligent agent that moves based on the motion planning method for the intelligent agent as described in the first aspect.

[0040] In a fourth aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the motion planning method for an intelligent agent as described in the first aspect above is implemented.

[0041] In a fifth aspect, the present application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the motion planning method for an intelligent agent as described in the first aspect above.

[0042] In a sixth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the motion planning method for an intelligent agent as described in the first aspect above.

[0043] The above one or more technical solutions in the embodiments of the present application have at least one of the following technical effects:

[0044] By predicting the pose set corresponding to each key point in the initial action sequence, multiple candidate sub-motion paths between adjacent points are obtained based on the pose set, and the feasibility of multiple candidate sub-motion paths is judged to screen and obtain the target motion plan. The path prediction can be refined to ensure that it is collision-free and achievable at each position, thereby improving the rationality of path planning and the precision and accuracy of obstacle avoidance. It is suitable for any scenario and any task, especially for the execution of long task sequences in narrow scenarios.

[0045] Furthermore, by performing collision analysis on the candidate posture information corresponding to each key point to extract posture information that can ensure no collision with other objects in the environment, the feasibility of the acquired first posture information can be significantly improved, and the risk of collision between the intelligent agent and the surrounding environment can be reduced, which is particularly suitable for intelligent agent control in narrow environments.

[0046] Furthermore, by further clustering the multiple second pose information corresponding to each key point based on the key point prediction to obtain candidate pose information, the multiple second pose information can be subjected to dimensionality reduction processing to reduce the amount of calculation, increase the calculation rate, and thus improve the efficiency of path planning.

[0047] Furthermore, when there are multiple feasible sub-candidate motion paths, the sub-candidate motion path with the lowest motion input cost is further screened based on the motion input cost as the target sub-motion path. On the basis of ensuring the feasibility of the path between adjacent key points obtained, the path is further made the optimal path, thereby improving the accuracy and rationality of path planning and improving the efficiency of the intelligent agent operation.

[0048] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0050] Figure 1 It is one of the flow charts of the motion planning method of an intelligent agent provided in an embodiment of the present application;

[0051] Figure 2 It is one of the principle schematic diagrams of the motion planning method of the intelligent agent provided in the embodiment of the present application;

[0052] Figure 3 This is the second principle schematic diagram of the motion planning method of the intelligent agent provided in the embodiment of the present application;

[0053] Figure 4 This is the second flow chart of the motion planning method of the intelligent agent provided in the embodiment of the present application;

[0054] Figure 5 is a schematic diagram of the structure of a motion planning device for an intelligent agent provided in an embodiment of the present application;

[0055] Figure 6 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0057] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0058] In conjunction with the accompanying drawings, the motion planning method of an intelligent agent, the motion planning device of an intelligent agent, the electronic device and the readable storage medium provided in the embodiments of the present application are described in detail through specific embodiments and their application scenarios.

[0059] The motion planning method of the intelligent agent may be applied to a terminal, and may be specifically executed by hardware or software in the terminal.

[0060] The motion planning method for an intelligent agent provided in an embodiment of the present application may be executed by an intelligent agent, or an electronic device connected to the intelligent agent for communication, or a functional module or functional entity in the electronic device that can implement the motion planning method for the intelligent agent. The electronic devices mentioned in the embodiment of the present application include but are not limited to mobile phones, tablet computers, computers, cameras, and wearable devices. The motion planning method for an intelligent agent provided in an embodiment of the present application is described below using an intelligent agent as an example of an execution subject.

[0061] like Figure 1As shown, the motion planning method of the intelligent agent includes: step 110, step 120 and step 130.

[0062] The motion planning method of the agent can be applied to agent control scenarios, wherein the agent can be a robot or other structure that can achieve the same function. The robot can include but is not limited to household agents, production agents, service agents, mobile robotic arms, aerial work agents and other intelligent agents.

[0063] like Figure 2 As shown, the intelligent body may include an actuator and a mobile mechanism, wherein the actuator is arranged on the mobile mechanism and is driven to move by the mobile mechanism.

[0064] The actuator is used to perform operations such as grasping, pushing, pulling, and placing, and can be a robotic arm.

[0065] The moving mechanism can be rollers or tracks, etc.

[0066] The object of execution is the object of execution of the executing agency.

[0067] It can be understood that in the scenario where the agent performs tasks, the agent can perform simple tasks such as grasping, placing, pushing and pulling, or navigation.

[0068] In some cases, the agent can also perform long task sequences, where a long task sequence is a task composed of multiple simple tasks; for example, in a home agent environment, the agent places the food on the coffee table in the refrigerator; in this long task sequence, multiple simple tasks are included, such as: grabbing food, moving food from the coffee table to the refrigerator, opening the refrigerator door, placing food, and closing the refrigerator door.

[0069] Step 110: based on the environmental information of the target scene, predict a posture set corresponding to each key point included in the initial action sequence; the key point corresponds to the target action state, and the posture set includes at least one first posture information, and the first posture information is used to characterize the posture state of the intelligent agent after executing the target action state corresponding to the key point;

[0070] In this step, the target scene can be any scene, including but not limited to: home intelligent scene, factory scene, shopping mall scene, etc.

[0071] For example, it can be used for food delivery scenarios in restaurants, household cleaning scenarios, and material transportation scenarios in factories.

[0072] Environmental information is used to represent scene-related information in the target scene, including but not limited to the scene structure, the position coordinates of each object in the scene, and the relationship between objects.

[0073] In some embodiments, the environmental information may include actual environmental information and virtual environmental information; wherein the actual environmental information is environmental information under a real scene, and the virtual environmental information is simulated environmental information used to simulate a real scene.

[0074] The initial action sequence is a series of key actions required for the agent to perform the target task.

[0075] The target task may be any executable task in the target scenario.

[0076] The key action is the action state that needs to be achieved at the key point, that is, the target action state.

[0077] In the actual execution process, the initial action sequence can be input into the pre-built target kinematic model between the intelligent agent and the executed object for kinematic modeling to obtain the target action state corresponding to each key point.

[0078] The method for determining the initial action sequence will be described below and will not be elaborated here.

[0079] like Figure 2 As shown in the figure, taking the target task of "taking out the cup from the refrigerator and putting it on the dining table" as an example, the corresponding initial action sequence may include: opening the refrigerator door - picking up the cup from the refrigerator - putting the cup on the table; wherein the key points include points A, B and C, the target action state corresponding to point A is opening the refrigerator door, the target action state corresponding to point B is picking up the cup from the refrigerator, and the target action state corresponding to point C is putting the cup on the table.

[0080] It is understandable that the points in the initial action sequence are the points corresponding to the agent before executing the action state, and cannot reflect the final posture state of the agent after executing the action state.

[0081] Take the example of the agent opening the refrigerator door at point A. After the agent completes the action state, the final posture state of the agent may change, such as Figure 3 As shown, after opening the refrigerator door, the agent may be in front of the refrigerator, such as Figure 3 As shown in the first area 310 in FIG. 1 ; or the agent may also be located on the side of the refrigerator, such as Figure 3 As shown in the second area 320.

[0082] The first pose information is used to characterize the pose state of the intelligent agent after executing the target action state corresponding to the key point.

[0083] For each key point, one or more first pose information can be predicted.

[0084] By combining all the first pose information predicted for the same key point, we can obtain the pose set corresponding to the key point, thereby obtaining multiple pose sets corresponding to all key points.

[0085] like Figure 4 As shown, in some embodiments, step 110 may include:

[0086] Based on the target action state corresponding to the key points, multiple candidate pose information corresponding to the key points is predicted;

[0087] A collision analysis is performed on multiple candidate pose information based on environmental information, and at least one first pose information is obtained from the multiple candidate pose information.

[0088] In this embodiment, the candidate pose information includes position information and pose information.

[0089] During the actual execution process, for each key point, it is possible to predict multiple or all posture states that may appear after the intelligent agent completes the action state corresponding to the key point, and use the obtained multiple or all posture states as multiple candidate posture information corresponding to the key point.

[0090] For example, when the intelligent agent performs the action state of "putting the cup on the table", it is predicted that it may be located in the front, back, left, and right areas of the table, thereby obtaining multiple candidate posture information.

[0091] Then, based on the environmental information, a collision analysis is performed on each candidate pose information to determine whether the intelligent agent will collide with the surrounding environment when in this pose information. The candidate pose information that may collide is deleted, and the remaining candidate pose information is determined as the first pose information.

[0092] Continuing with the above-mentioned action state of "putting the cup on the table" as an example, based on the environmental information, it is determined that the north side of the table is against the wall. It can be predicted that a collision will occur when the agent is on the north side of the table, thereby deleting the candidate posture information corresponding to the north side.

[0093] In some embodiments, at least one first pose information may be determined based on the following formula:

[0094]

[0095]

[0096]

[0097]

[0098]

[0099] Among them, a 1:N is the initial action sequence; q ai is a candidate key posture of ai, which includes the postures of the agent and the object being executed; ai is the i-th action state; N is the number of key points included in the initial action sequence; chain is the actuator; h chain is the environmental constraint, which describes the kinematic structure, including its forward kinematics and the physical constraints between the executed object and the environment; task is the real-time state value of the agent and the executed object in the target space. It can be understood that the real-time state value f task (q T ) is task-specific and is constrained by mapping the joint space to the task goal space; g ai For action a i ξ is the target information; ξ is the corresponding threshold.

[0100] In this embodiment, a “key configuration selection” algorithm is used to generate an initial action sequence a 1:N The key configuration sequence: K* al:N ={q* al , q* a2 , ..., q* aN}∈K al ×K a2 ×……×K aN ,in, is the action ai∈a 1:N All feasible candidate key configurations of , each key configuration q* ai ∈K ai They are all used as the target states of the agent and the object being executed in the motion planning problem corresponding to the action ai. al:N A series of motion planning problems on a1:N can find a feasible motion plan for the entire initial action sequence a1:N.

[0101] By minimizing kinematic and task constraints, and using small tolerances to reduce unwanted results. At the same time, safety constraints can be used to ai Constrained within joint limits, including joints of the agent and the object being executed.

[0102] The last term in the above formula is used to ensure that the key pose does not collide with other objects in the environment.

[0103] According to the motion planning method of the intelligent agent provided in the embodiment of the present application, by performing collision analysis on the candidate posture information corresponding to each key point to extract posture information that can ensure that there is no collision with other objects in the environment, the feasibility of the first posture information obtained can be significantly improved, and the risk of collision between the intelligent agent and the surrounding environment can be reduced. It is particularly suitable for intelligent agent control in narrow environments.

[0104] In some embodiments, based on the target action state corresponding to the key points, predicting multiple candidate pose information corresponding to the key points may include:

[0105] Based on the target action state corresponding to the key points, multiple second posture information corresponding to the key points are predicted;

[0106] Clustering is performed on the multiple second pose information to obtain multiple candidate pose information corresponding to the key points.

[0107] In this embodiment, the multiple second posture information are all the predicted posture information.

[0108] It is understandable that the amount of second pose information corresponding to each key point may be relatively large.

[0109] During the actual execution process, multiple second pose information can be clustered to obtain multiple clusters, and the pose information corresponding to the cluster center of each cluster can be determined as the candidate pose information corresponding to the cluster, so as to effectively reduce the amount of pose information corresponding to each key point, thereby reducing the motion planning cost.

[0110] For example, in some embodiments, a clustering algorithm (such as k-means++) may be used to reduce the dimensionality of the plurality of second pose information and map them to a low-dimensional space to avoid repeated calculation of similar configurations.

[0111] Among them, the k-means++ algorithm can divide n observations (i.e., the second posture information) into k clusters by minimizing the intra-cluster variance; after clustering, a posture closest to the cluster center is selected from each cluster to perform motion feasibility check.

[0112] According to the motion planning method of the intelligent body provided in the embodiment of the present application, by further clustering the multiple second pose information corresponding to each key point based on the prediction of the key points to obtain candidate pose information, the multiple second pose information can be subjected to dimensionality reduction processing to reduce the amount of calculation, increase the calculation rate, thereby improving the efficiency of path planning and reducing the cost of motion planning.

[0113] The following describes how to determine the initial action sequence.

[0114] In some embodiments, before step 110, the method may further include: determining an initial action sequence corresponding to the agent based on the environment information and the target task information.

[0115] In this embodiment, the target task information is relevant information of the target task.

[0116] In some embodiments, determining an initial action sequence corresponding to the agent based on the environment information and the target task information may include:

[0117] Based on the target task information, determining at least one of the agent displacement action information, the agent grasping action information, and the executed object displacement information;

[0118] An initial action sequence is determined based on at least one of the intelligent agent displacement action information, the intelligent agent grasping action information, the executed object displacement information, and the environmental information.

[0119] In this embodiment, the initial action sequence may include target action states arranged in sequence based on a time sequence.

[0120] The agent displacement action information is GoTo(chain,q1,q2), which moves the actuator chain from the current position q1 to the position q2, and this action does not require interaction with the environment; GoTo can represent tasks in which the motion chain structure does not need to be changed, such as the agent navigation task, which is a typical task of this type.

[0121] The agent's grasping action information is Pick(object, state, chain), which moves the actuator chain to the target state state and grasps the executed object object, and expands the current chain structure by adding virtual joints after kinematic inversion; Pick can ensure the tasks of the agent's interaction with the environment, such as picking up objects, grabbing door handles, or picking up tools.

[0122] The displacement information of the executed object is Place(object, state, chain), which moves the executed object object connected to the actuator chain to the target posture state and disconnects the current chain at the virtual joint; Place indicates the task of the agent stopping interaction with the environment, such as putting down an object, opening the door to a certain angle and releasing the gripper, and using tools to operate other objects.

[0123] It can be understood that through the agent displacement action information, the agent grasping action information and the displacement information of the executed object, the changes in the motion chain structure can be regarded as different actions, thereby covering most mobile operation tasks and effectively describing most tasks that change the kinematic state of the environment, providing a basis for tasks and motion planning methods for diverse operation objects.

[0124] In the actual execution process, the motion chain structure of the intelligent agent can be decomposed into multiple different target action states (such as GoTo, Pick, and Place, etc.) based on the target task information, and then the different target action states can be connected based on the time sequence to obtain the initial action sequence.

[0125] According to the motion planning method of the intelligent agent provided in the embodiment of the present application, the motion chain structure is decomposed into multiple target action states based on the target task information, which can effectively cover most mobile operation tasks, thereby effectively describing tasks of various kinematic states; on this basis, the target action states are connected based on the time sequence to obtain the initial action sequence, which has high accuracy and precision, and is particularly suitable for complex motion chain structures, and thus suitable for complex execution actions.

[0126] Step 120: based on the pose sets corresponding to the target adjacent points in the plurality of key points, obtain a plurality of candidate sub-motion paths between the target adjacent points;

[0127] In this step, the target adjacent point may be any adjacent point among the multiple key points.

[0128] The candidate sub-motion path is the moving path from the first pose information corresponding to the previous point to the first pose information corresponding to the next point.

[0129] like Figure 2 and Figure 3 As shown, when the agent is located in front of the refrigerator door, the candidate sub-movement path is from the front of the refrigerator door to point B; when the agent is located on the side of the refrigerator door, the candidate sub-movement path is from the side of the refrigerator door to point B.

[0130] For each group of adjacent points, one or more candidate sub-motion paths corresponding to the group of adjacent points can be obtained.

[0131] In some embodiments, step 120 may include:

[0132] Randomly combine the first pose information in the pose set corresponding to the adjacent points of the target to obtain multiple pose combinations;

[0133] The first pose information in each pose combination is connected to generate a candidate motion path corresponding to the pose combination, and multiple candidate motion paths are obtained.

[0134] In this embodiment, continue to refer to Figure 2 , taking the target adjacent points A and B as an example, where the position set corresponding to point A is {a1, a2}, and the position set corresponding to point B is {b1, b2}, a first position information in the position set {a1, a2} and a first position information in the position set {b1, b2} corresponding to point B can be randomly extracted and combined to obtain a candidate motion path from point A to point B, such as multiple candidate motion paths from point A to point B: a1-b1, a1-b2, a2-b1 and a2-b2.

[0135] Step 130: Process multiple candidate sub-motion paths corresponding to each adjacent point based on the environmental information to determine the target motion plan.

[0136] In this step, the environmental information is used to characterize the position information of each object in the target scene and the obstacle information, etc. The obstacle information can be any obstacle-related information that blocks the action of the intelligent agent.

[0137] In the actual execution process, obstacle information between adjacent target points can be obtained, and candidate motion paths can be screened based on the obstacle information to obtain feasible motion paths between adjacent target points.

[0138] Target motion planning is used to represent the entire planning information of the intelligent agent when performing the entire target task. Target motion planning includes: target motion path and posture information at the target position in the target motion path.

[0139] In the same way, a feasible motion path between any adjacent points can be obtained, and then the feasible motion paths between the adjacent points of each mark are connected based on the connection order of the points to obtain the target motion plan.

[0140] Continuing with the example of the first pose information corresponding to point A being the side of the refrigerator door or the front of the refrigerator, when the agent is in front of the refrigerator door, the candidate sub-movement path is from the front of the refrigerator door to point B, which is Figure 2 It can be seen that there are no obstacles between the candidate sub-movement paths, so the path is preliminarily considered feasible; when the intelligent agent is located on the side of the refrigerator door, the candidate sub-movement path is to move from the side of the refrigerator door to point B. There is an open refrigerator door (i.e., an obstacle) between the candidate sub-movement paths, which blocks the moving path of the intelligent agent to the next point, so the path is considered infeasible.

[0141] During the research and development process, the inventors discovered that in the related technologies, sensor data is mainly collected by sensors installed on the intelligent body, so as to control the intelligent body to move or grasp and other actions based on the sensor data. For example, in the actual operation of the intelligent body, the sensor perceives the obstacle information nearby in real time, and controls the intelligent body to bypass when an obstacle is detected in front. However, this method is only applicable to simple short task control. When controlling a long task sequence, there are problems such as large control error and low control efficiency, resulting in poor stability of the intelligent body. In addition, there is also a method in the related technologies that controls the intelligent body to run according to a preset path based on a set program, so as to realize the control of a long task sequence. However, this method often requires setting the only fixed track and specific points for the operation of the intelligent body in advance, so that the intelligent body can travel along the running track to a specific point to perform specific operations. This method has poor flexibility and cannot cope with complex and changeable application scenarios. Moreover, the above methods are not applicable to the control of intelligent bodies in narrow spaces.

[0142] In the present application, the initial action sequence is subdivided into multiple key points and the posture information corresponding to each key point is predicted to obtain a posture set corresponding to each point, and multiple candidate sub-motion paths between adjacent points are obtained based on the posture set corresponding to adjacent points, thereby improving the comprehensiveness of the obtained sub-motion paths and being able to improve the precision and accuracy of the obtained paths; on this basis, the feasibility of multiple sub-candidate motion paths corresponding to each adjacent point is screened based on environmental information to determine the target motion plan, which can ensure the effectiveness of the running path of the intelligent agent when performing the entire target task, ensure that it is collision-free and achievable at each position, improve the rationality of path planning, and significantly improve the precision and accuracy of obstacle avoidance. It is suitable for any scenario and any task, especially for long task sequence execution scenarios in narrow scenarios.

[0143] According to the motion planning method of the intelligent agent provided in the embodiment of the present application, the posture set corresponding to each key point in the initial action sequence is obtained by prediction, and multiple candidate sub-motion paths between adjacent points are obtained based on the posture set, and the feasibility of the multiple candidate sub-motion paths is judged to screen and obtain the target motion plan. The path prediction can be refined to ensure that it is collision-free and achievable at each position, thereby improving the rationality of path planning and the precision and accuracy of obstacle avoidance. It is suitable for any scenario and any task, and is particularly suitable for the execution scenarios of long task sequences in narrow scenarios.

[0144] In some embodiments, step 130 may include:

[0145] Based on the environmental information, a target sub-motion path is obtained by screening a plurality of candidate sub-motion paths corresponding to target adjacent points in a plurality of adjacent points;

[0146] The target sub-motion paths corresponding to each adjacent point are integrated to obtain the target motion path;

[0147] The target motion plan is determined based on the target motion path and the first pose information corresponding to each key point in the target motion path.

[0148] In this embodiment, the target sub-motion path is a feasible candidate sub-motion path between adjacent target points.

[0149] There can be one or more target sub-motion paths.

[0150] The target motion path is the motion path of the agent when performing the entire target task.

[0151] For example, taking the target task "take out the cups from the refrigerator and put them on the dining table" as an example, continue to refer to Figure 2 The target motion path finally obtained is the path of the agent moving from point A to point B, and then from point B to point C.

[0152] In the case where there are multiple target sub-motion paths, each target sub-motion path may be fused separately to obtain multiple target motion paths.

[0153] It can be understood that the target motion path includes multiple key points, each key point corresponds to the first pose information under the target motion path. Based on the first pose information, the position coordinates and posture information of the intelligent body at the key point can be determined, and the position coordinates and posture information of the intelligent body at the target key point are added to the target key point in the target motion path to obtain the target motion plan.

[0154] Continue to refer Figure 4 In some embodiments, based on the environmental information, selecting a target sub-motion path from a plurality of candidate sub-motion paths corresponding to a target adjacent point in a plurality of adjacent points to obtain the target sub-motion path may include:

[0155] Perform feasibility detection on multiple sub-candidate motion paths corresponding to adjacent points of the target based on environmental information, and obtain the realization status corresponding to each sub-candidate motion path, where the realization status includes a feasible state and an infeasible state;

[0156] Based on the candidate sub-motion paths corresponding to the feasible states, a target sub-motion path is determined.

[0157] In this embodiment, the feasibility detection is used to detect whether the sub-candidate motion path is feasible.

[0158] In the actual execution process, the checkMotionFeasibility function can be used to verify the feasibility.

[0159] For example, let's continue to use the example of the first pose information corresponding to point A after opening the refrigerator door or in front of the refrigerator. When the agent is in front of the refrigerator door, the candidate sub-movement path is from the front of the refrigerator door to point B, which is Figure 2 It can be seen that there are no obstacles between the candidate sub-movement paths, so the path is preliminarily considered feasible; when the intelligent agent is located on the side of the refrigerator door, the candidate sub-movement path is to move from the side of the refrigerator door to point B. There is an open refrigerator door (i.e., an obstacle) between the candidate sub-movement paths, which blocks the moving path of the intelligent agent to the next point, so the path is considered infeasible.

[0160] After obtaining the realization status corresponding to each sub-candidate motion path, multiple sub-candidate motion paths between adjacent points can be screened based on the realization status, unfeasible sub-candidate motion paths are removed, feasible sub-candidate motion paths are retained, and the target sub-motion path between adjacent points is determined based on the feasible sub-candidate motion paths.

[0161] In some embodiments, determining the target sub-motion path based on the sub-candidate motion path corresponding to the feasible state may include: determining the sub-candidate motion path corresponding to the feasible state as the target sub-motion path.

[0162] In this embodiment, when the number of the sub-candidate motion paths corresponding to the feasible state is 1, the sub-candidate motion paths corresponding to the feasible state may be directly determined as the target sub-motion paths.

[0163] For another example, when there are multiple candidate sub-motion paths corresponding to feasible states, the candidate sub-motion paths corresponding to the multiple feasible states may be determined as target sub-motion paths, respectively, for the user to select based on actual needs.

[0164] Of course, in other embodiments, when there are multiple sub-candidate motion paths corresponding to feasible states, the best feasible sub-candidate motion path can be selected from the sub-candidate motion paths corresponding to multiple feasible states as the target sub-motion path based on algorithm settings, etc., and this application is not limited here.

[0165] According to the motion planning method of the intelligent agent provided in the embodiment of the present application, feasibility screening is performed on multiple sub-candidate motion paths between any adjacent key points based on environmental information, which can ensure that the target sub-motion paths between any adjacent points finally obtained are feasible, and the intelligent agent will not collide with the surrounding environment when running between the adjacent points. It has high control accuracy and stability and a wide range of applications.

[0166] In some embodiments, performing feasibility detection on multiple candidate sub-motion paths corresponding to target adjacent points based on environmental information and obtaining the realization status corresponding to each candidate sub-motion path may include:

[0167] Generate obstacle-free areas between adjacent target points based on environmental information;

[0168] When the boundary of the intelligent body does not exceed the boundary of the obstacle-free area, determining that the realization state corresponding to the sub-candidate motion path is a feasible state;

[0169] When the boundary of the intelligent body exceeds the boundary of the obstacle-free area, it is determined that the realization state corresponding to the sub-candidate motion path is an infeasible state.

[0170] In this embodiment, the obstacle-free area is an area where no obstacles exist.

[0171] The obstacle-free area is the area between adjacent target points and near the area where the sub-candidate motion path passes.

[0172] Continue to refer Figure 3 , taking the case where the intelligent agent is in the second area 320 at point A as an example, the obstacle-free area between point A and point B is the area between the table and one side of the refrigerator.

[0173] In the case where the boundary of the intelligent body does not exceed the boundary of the obstacle-free area, it can be approximately considered that the intelligent body can pass through the obstacle-free area, and the realization state corresponding to the sub-candidate motion path is determined to be a feasible state.

[0174] When the boundary of the agent exceeds the boundary of the obstacle-free area, it is considered that the agent cannot pass through the obstacle-free area (for example, the channel is too narrow and the base of the agent cannot pass through), and the realization state corresponding to the sub-candidate motion path is determined to be an infeasible state.

[0175] In some embodiments, whether the boundary of the agent exceeds the boundary of the obstacle-free area can be determined based on the following method:

[0176] Divide the target scene into multiple identical grids, and mark obstacle-free areas in the multiple identical grids at the same scale;

[0177] Transform the agent's mobile mechanism into a planar region that is consistent with the scale of the grid;

[0178] When the edge grid of the plane area does not exceed the edge grid of the obstacle-free area, determining that the realization state corresponding to the sub-candidate motion path is a feasible state;

[0179] When the edge grid of the plane area exceeds the edge grid of the obstacle-free area, it is determined that the realization state corresponding to the sub-candidate motion path is an infeasible state.

[0180] In the actual implementation process, the A* algorithm can be used to determine the connectivity of the mobile base between adjacent key points to determine the feasibility of the movement.

[0181] According to the motion planning method of the intelligent body provided in the embodiment of the present application, by comparing the boundary of the intelligent body with the boundary of the obstacle-free area to determine whether the sub-candidate motion path is a feasible path, the calculation is simple, the operation is simple and convenient, and the accuracy of the judgment result is high.

[0182] In some embodiments, determining a target sub-motion path based on the candidate sub-motion paths corresponding to the feasible states may include:

[0183] Predict the motion input cost corresponding to the sub-candidate motion path corresponding to each feasible state executed by the agent;

[0184] The candidate sub-motion path corresponding to the minimum motion input cost is determined as the target sub-motion path.

[0185] In this embodiment, the motion input cost is used to represent the cost that the agent needs to consume when performing a task based on each sub-candidate motion path.

[0186] It can be understood that the smaller the motion input cost is, the lower the cost required to execute a task based on the sub-candidate motion path is, and the more reasonable the sub-candidate motion path is.

[0187] When there are multiple candidate sub-motion paths between adjacent points, it means that the pose set corresponding to the key point includes multiple feasible first pose information. Then the selectBest function can be used to select one of them that satisfies the following formula:

[0188]

[0189] Among them, q ai is a candidate key pose of ai; ai is the i-th action state; N is the number of key points included in the initial action sequence; a1:N is the initial action sequence; c(.) is the cost function used to evaluate the potential motion investment of the agent between two configurations.

[0190] The target sub-motion path can be determined.

[0191] According to the motion planning method of the intelligent agent provided in the embodiment of the present application, when there are multiple feasible sub-candidate motion paths, the sub-candidate motion path with the lowest motion input cost is further screened based on the motion input cost as the target sub-motion path. On the basis of ensuring the feasibility of the path between adjacent key points obtained, the path is further made the optimal path, thereby improving the accuracy and rationality of the path planning and improving the efficiency of the operation of the intelligent agent.

[0192] The motion planning method of the intelligent agent provided in the embodiment of the present application can be executed by the motion planning device of the intelligent agent. In the embodiment of the present application, the motion planning device of the intelligent agent performing the motion planning method of the intelligent agent is taken as an example to illustrate the motion planning device of the intelligent agent provided in the embodiment of the present application.

[0193] An embodiment of the present application also provides a motion planning device for an intelligent agent.

[0194] like Figure 5 As shown, the motion planning device of the intelligent agent includes: a first processing module 510, a second processing module 520 and a third processing module 530.

[0195] The first processing module 510 is used to predict the posture set corresponding to each key point included in the initial action sequence based on the environmental information under the target scene; the key point corresponds to the target action state, and the posture set includes at least one first posture information, and the first posture information is used to represent the posture state of the intelligent agent when executing the target action state corresponding to the key point;

[0196] The second processing module 520 is used to obtain multiple candidate sub-motion paths between adjacent target points based on the pose sets corresponding to the target adjacent target points in the multiple key points;

[0197] The third processing module 530 is used to process multiple candidate sub-motion paths corresponding to each adjacent point based on the environmental information to determine the target motion plan.

[0198] According to the motion planning device of the intelligent body provided in the embodiment of the present application, the posture set corresponding to each key point in the initial action sequence is obtained by prediction, and multiple candidate sub-motion paths between adjacent points are obtained based on the posture set, and the feasibility of the multiple candidate sub-motion paths is judged to screen and obtain the target motion plan. The path prediction can be refined to ensure that it is collision-free and achievable at each position, thereby improving the rationality of path planning and the precision and accuracy of obstacle avoidance. It is suitable for any scene and any task, and is particularly suitable for the execution scenarios of long task sequences in narrow scenes.

[0199] In some embodiments, the third processing module 530 may also be used to:

[0200] Based on the environmental information, a target sub-motion path is obtained by screening a plurality of candidate sub-motion paths corresponding to target adjacent points in a plurality of adjacent points;

[0201] The target sub-motion paths corresponding to each adjacent point are integrated to obtain the target motion path;

[0202] The target motion plan is determined based on the target motion path and the first pose information corresponding to each key point in the target motion path.

[0203] In some embodiments, the third processing module 530 may also be used to:

[0204] Perform feasibility detection on multiple sub-candidate motion paths corresponding to adjacent points of the target based on environmental information, and obtain the realization status corresponding to each sub-candidate motion path, where the realization status includes a feasible state and an infeasible state;

[0205] Based on the candidate sub-motion paths corresponding to the feasible states, a target sub-motion path is determined.

[0206] In some embodiments, the third processing module 530 may also be used to:

[0207] Predict the motion input cost corresponding to the sub-candidate motion path corresponding to each feasible state executed by the agent;

[0208] The candidate sub-motion path corresponding to the minimum motion input cost is determined as the target sub-motion path.

[0209] In some embodiments, the third processing module 530 may also be used to:

[0210] Generate obstacle-free areas between adjacent target points based on environmental information;

[0211] When the boundary of the intelligent body does not exceed the boundary of the obstacle-free area, determining that the realization state corresponding to the sub-candidate motion path is a feasible state;

[0212] When the boundary of the intelligent body exceeds the boundary of the obstacle-free area, it is determined that the realization state corresponding to the sub-candidate motion path is an infeasible state.

[0213] In some embodiments, the second processing module 520 may also be used to:

[0214] Randomly combine the first pose information in the pose set corresponding to the adjacent points of the target to obtain multiple pose combinations;

[0215] The first pose information in each pose combination is connected to generate a candidate motion path corresponding to the pose combination, and multiple candidate motion paths are obtained.

[0216] In some embodiments, the first processing module 510 may also be used to:

[0217] Based on the target action state corresponding to the key points, multiple candidate pose information corresponding to the key points is predicted;

[0218] A collision analysis is performed on multiple candidate pose information based on environmental information, and at least one first pose information is obtained from the multiple candidate pose information.

[0219] In some embodiments, the first processing module 510 may also be used to:

[0220] Based on the target action state corresponding to the key points, multiple second posture information corresponding to the key points are predicted;

[0221] Clustering is performed on the multiple second pose information to obtain multiple candidate pose information corresponding to the key points.

[0222] In some embodiments, the device may also include a fourth processing module for determining an initial action sequence corresponding to the intelligent agent based on environmental information and target task information before predicting a set of postures corresponding to each key point included in the initial action sequence based on environmental information in the target scene.

[0223] The motion planning device of the intelligent body in the embodiment of the present application can be an electronic device, or a component in the electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal, or it can be other devices other than the terminal. Exemplarily, the electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, a car electronic device, a mobile Internet device (Mobile Internet Device, MID), an augmented reality (augmented reality, AR) / virtual reality (virtual reality, VR) device, an intelligent body, a wearable device, an ultra-mobile personal computer (ultra-mobile personal computer, UMPC), a netbook or a personal digital assistant (personal digital assistant, PDA), etc., and can also be a server, a network attached storage (Network Attached Storage, NAS), a personal computer (personal computer, PC), a television (television, TV), a teller machine or a self-service machine, etc., which is not specifically limited in the embodiment of the present application.

[0224] The motion planning device of the intelligent agent in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an IOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0225] The motion planning device of the intelligent body provided in the embodiment of the present application can achieve Figures 1 to 4 To avoid repetition, the various processes implemented by the method embodiment are not described here.

[0226] An embodiment of the present application also provides an intelligent agent, which moves based on the motion planning method of the intelligent agent as described in any of the above embodiments.

[0227] In some embodiments, Figure 6 As shown, an embodiment of the present application also provides an electronic device 600, including a processor 601, a memory 602, and a computer program stored in the memory 602 and executable on the processor 601. When the program is executed by the processor 601, each process of the above-mentioned intelligent body motion planning method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0228] It should be noted that the electronic devices in the embodiments of the present application include the mobile electronic devices and non-mobile electronic devices mentioned above.

[0229] An embodiment of the present application also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various processes of the above-mentioned intelligent agent motion planning method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0230] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0231] An embodiment of the present application also provides a computer program product, including a computer program, which implements the motion planning method of the above-mentioned intelligent agent when executed by a processor.

[0232] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0233] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned intelligent body motion planning method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0234] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0235] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0236] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0237] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

[0238] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0239] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present application, and that the scope of the present application is defined by the claims and their equivalents.

Claims

1. A motion planning method for an intelligent agent, characterized in that: include: Based on the environmental information of the target scene, a posture set corresponding to each key point included in the initial action sequence is predicted; the key point corresponds to a target action state, and the posture set includes at least one first posture information, and the first posture information is used to characterize the posture state of the intelligent agent after executing the target action state corresponding to the key point; Based on a set of poses corresponding to adjacent target points in a plurality of key points, a plurality of candidate sub-motion paths between adjacent target points are obtained; Based on the environmental information, multiple candidate sub-motion paths corresponding to each adjacent point are processed to determine the target motion plan.

2. The motion planning method of an intelligent agent according to claim 1, characterized in that: The step of processing a plurality of candidate sub-motion paths corresponding to each adjacent point based on the environmental information to determine the target motion plan includes: Based on the environmental information, a target sub-motion path is obtained by screening a plurality of candidate sub-motion paths corresponding to the target adjacent point in the plurality of adjacent points; The target sub-motion paths corresponding to each adjacent point are integrated to obtain the target motion path; The target motion plan is determined based on the target motion path and the first pose information corresponding to each key point in the target motion path.

3. The motion planning method of an intelligent agent according to claim 2, characterized in that: The step of selecting a target sub-motion path from a plurality of candidate sub-motion paths corresponding to a target adjacent point in a plurality of adjacent points based on the environmental information includes: Perform feasibility detection on a plurality of sub-candidate motion paths corresponding to the target adjacent points based on the environmental information, and obtain a realization state corresponding to each of the sub-candidate motion paths, wherein the realization state includes a feasible state and an infeasible state; The target sub-motion path is determined based on the candidate sub-motion paths corresponding to the feasible state.

4. The motion planning method of an intelligent agent according to claim 3, characterized in that: The determining the target sub-motion path based on the candidate sub-motion path corresponding to the feasible state includes: Predicting the motion input cost corresponding to the execution of each candidate sub-motion path corresponding to each feasible state by the agent; The candidate sub-motion path corresponding to the minimum motion input cost is determined as the target sub-motion path.

5. The motion planning method of an intelligent agent according to claim 3, characterized in that: The performing feasibility detection on a plurality of sub-candidate motion paths corresponding to the target adjacent points based on the environmental information to obtain the realization status corresponding to each of the sub-candidate motion paths includes: Based on the environmental information, generating an obstacle-free area between adjacent target points; In a case where the boundary of the intelligent body does not exceed the boundary of the obstacle-free area, determining that the realization state corresponding to the sub-candidate motion path is a feasible state; In the case where the boundary of the intelligent body exceeds the boundary of the obstacle-free area, it is determined that the realization state corresponding to the sub-candidate motion path is an infeasible state.

6. The motion planning method of an intelligent agent according to any one of claims 1 to 5, characterized in that: The step of obtaining a plurality of candidate sub-motion paths between adjacent target points based on a set of poses corresponding to adjacent target points in a plurality of key points includes: Randomly combining the first pose information in the pose set corresponding to the adjacent points of the target to obtain multiple pose combinations; The first pose information in each of the pose combinations is connected to generate a candidate motion path corresponding to the pose combination, so as to obtain the multiple candidate motion paths.

7. The motion planning method of an intelligent agent according to any one of claims 1 to 5, characterized in that: The method of predicting a set of postures corresponding to key points included in the initial action sequence based on the environmental information of the target scene includes: Based on the target action state corresponding to the key point, predict multiple candidate posture information corresponding to the key point; A collision analysis is performed on the multiple candidate pose information based on the environmental information, and the at least one first pose information is obtained by screening out from the multiple candidate pose information.

8. The motion planning method of an intelligent agent according to claim 7, characterized in that: The predicting of a plurality of candidate pose information corresponding to the key points based on the target action state corresponding to the key points includes: Based on the target action state corresponding to the key point, predict a plurality of second posture information corresponding to the key point; Clustering is performed on the multiple second pose information to obtain multiple candidate pose information corresponding to the key points.

9. The motion planning method of an intelligent agent according to any one of claims 1 to 5, characterized in that: Before predicting a set of postures corresponding to key points included in the initial action sequence based on the environmental information of the target scene, the method further includes: Based on the environment information and the target task information, an initial action sequence corresponding to the agent is determined.

10. A motion planning device for an intelligent agent, characterized in that: include: A first processing module is used to predict a posture set corresponding to each key point included in the initial action sequence based on environmental information in the target scene; the key point corresponds to a target action state, and the posture set includes at least one first posture information, and the first posture information is used to represent the posture state of the intelligent agent after executing the target action state corresponding to the key point; A second processing module is used to obtain multiple candidate sub-motion paths between adjacent target points based on a set of poses corresponding to adjacent target points in multiple key points; The third processing module is used to process multiple sub-candidate motion paths corresponding to each adjacent point based on the environmental information to determine the target motion plan.

11. An intelligent agent, characterized in that: The intelligent agent moves based on the motion planning method for the intelligent agent as described in any one of claims 1-9.

12. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements the motion planning method of the intelligent agent as described in any one of claims 1-9.

13. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the motion planning method of an intelligent agent as described in any one of claims 1 to 9 is implemented.

14. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the motion planning method of the intelligent agent as described in any one of claims 1-9 is implemented.

Images