Mechanical arm self-adaptive control method and spinal surgery planning and executing system

Abstract

The invention provides a mechanical arm self-adaptive control method and a spinal surgery planning and executing system. The method comprises the steps that control parameters, output based on a large model, of a mechanical arm are obtained; the control parameters at least comprise a rigidity parameter, a damping parameter and a pose parameter; constructing a kinetic equation of the mechanical arm; based on the kinetic equation, determining a control instruction corresponding to the control parameter; and executing the control instruction in real time. According to the method, the control parameters generated by the large model are effectively combined with the mechanical arm kinetic equation, and the control instruction is adaptively determined according to the kinetic model and executed in real time, so that the intelligent parameter generation advantage of the large model can be utilized, and the stability, safety and real-time performance of control can be guaranteed under the dynamic constraint.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mechanical arm control, in particular to a mechanical arm adaptive control method and a spine surgery planning and execution system. BACKGROUND

[0002] With the development of intelligent manufacturing, service robots and surgical robots, the demand for mechanical arms to perform tasks with high compliance, high precision and high safety in complex environments is increasing. Traditional mechanical arm control methods are mainly based on fixed-parameter force-position hybrid control, impedance control or adaptive control, and their control performance is heavily dependent on manually preset stiffness parameters, damping parameters and target pose parameters. In complex task environments, manually set control parameters often fail to balance stability, compliance and task performance. SUMMARY

[0003] To solve the above problems, the first aspect of the present application provides a mechanical arm adaptive control method, which comprises: obtaining control parameters of a mechanical arm based on large model output; the control parameters at least include stiffness parameters, damping parameters and pose parameters; constructing a dynamics equation of the mechanical arm; determining a control instruction corresponding to the control parameters based on the dynamics equation; real-time execution of the control instruction.

[0004] The second aspect of the present application provides a spine surgery planning and execution system for executing the above-mentioned mechanical arm adaptive control method, which comprises: a data acquisition module for acquiring static input data and dynamic input data of a patient and encoding; a VLA control center for planning surgical parameters of spine surgery and constructing a visual language action model based on a Transformer architecture to generate control parameters of a mechanical arm; a mechanical arm system for executing the mechanical arm adaptive control method of claim 1 or 2 according to the control parameters of the mechanical arm.

[0005] In the present application, the control parameters generated by the large model are effectively combined with the dynamics equation of the mechanical arm, and the control instruction is adaptively determined and executed in real time according to the dynamics model, so as to not only take advantage of the intelligent parameter generation of the large model, but also ensure the stability, safety and real-time performance of the control under the dynamics constraint. BRIEF DESCRIPTION OF DRAWINGS

[0006] Figure 1 Architectural diagram of the mechanical arm adaptive control method according to the embodiments of the present application; Figure 2Fig. 1 is a schematic diagram of a spinal surgery planning and execution system according to an embodiment of the present application; Figure 3 Fig. 2 is a schematic diagram of internal processing logic of a VLA of a spinal surgery planning and execution system according to an embodiment of the present application; Figure 4 Fig. 3 is a schematic diagram of a semantic-driven variable impedance control of a spinal surgery planning and execution system according to an embodiment of the present application; Figure 5 Fig. 4 is a schematic diagram of a full-process closed-loop workflow of a spinal surgery planning and execution system according to an embodiment of the present application; Figure 6 Fig. 5 is a schematic diagram of a planning layer of a spinal surgery planning and execution system according to an embodiment of the present application; Figure 7 Fig. 6 is a schematic diagram of cross-modal alignment of a spinal surgery planning and execution system according to an embodiment of the present application; Figure 8 Fig. 7 is a schematic diagram of an architecture of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0007] In order to make the above objectives, features and advantages of the present application more apparent, specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present application can be more thoroughly understood and the scope of the present application can be accurately conveyed to those skilled in the art.

[0008] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in the present application should be understood as their common meanings by those skilled in the art to which the present application belongs.

[0009] In order to facilitate understanding, the following terms that can be used are explained here: The embodiments of the present application provide a mechanical arm adaptive control method, and a specific scheme of the system is shown in Figure 1 .

[0010] In combination with Figure 1 , which is an architecture diagram of a mechanical arm adaptive control method according to an embodiment of the present application; wherein the mechanical arm adaptive control method comprises: S101, obtaining a control parameter of a mechanical arm based on a large model output; the control parameter at least includes a stiffness parameter, a damping parameter and a pose parameter; wherein the control parameter is a control parameter output by the large model and corresponding to a surgical stage of the spine nail placement. The stiffness parameter is used for impedance or compliance control of the robot arm; the damping parameter is used for stabilizing the robot arm motion; and the pose parameter corresponds to a target position and attitude respectively.

[0011] S102, constructing a dynamics equation of the robot arm; In an embodiment, the dynamics equation is:

[0012] wherein, F des is an output power, K des is a variable stiffness parameter, D des is a variable damping parameter, des is a target pose, curr is an actual pose, des is a target velocity, ’ curr is an actual velocity.

[0013] S103, determining a control instruction corresponding to the control parameter based on the dynamics equation; The control parameter is input into the dynamics equation to obtain an output power of the end of the robot arm. The end force is mapped to the joint space through a Jacobian matrix transformation to obtain a real-time control instruction. The real-time control instruction is the product of the end force and the transposed Jacobian matrix.

[0014] In this way, the output power is mapped to the joint torque, and then the corresponding control instruction is generated based on the joint torque, so that the control process has small calculation amount and strong real-time performance.

[0015] S104, executing the control instruction in real time.

[0016] wherein the control instruction is output to each driver of the robot arm for execution.

[0017] It should be noted that the control instruction is executed in a loop, that is, steps S101-S103 continuously generate control instructions, and the control instructions are executed one by one in this step.

[0018] In this application, the output parameter of the large model has a low update frequency (limited by calculation time, etc.), and the parameter of the robot arm control has a high update frequency; therefore, in the case where the large model is not updated, the controller uses the latest effective parameter and performs a smooth transition.

[0019] Preferably, safety parameters (stiffness parameters, damping parameters) are set. When the output of the large model is unavailable / abnormal, the robotic arm automatically reverts to the default safety parameters for control (at this time, the preset trajectory parameters / initial trajectory parameters can be used as pose parameters).

[0020] In this application, the control parameters generated by the large model are effectively combined with the dynamic equations of the robotic arm, and the control commands are adaptively determined and executed in real time based on the dynamic model. This not only allows us to take advantage of the intelligent parameter generation of the large model, but also ensures the stability, safety and real-time performance of the control under dynamic constraints.

[0021] Among them, the large model automatically predicts control parameters suitable for the current task, working condition and environment through training data, enabling the robotic arm to perform tasks in a compliant and stable manner.

[0022] In this application, the stiffness, damping, and pose control parameters output by the large model are used to enable the robotic arm to automatically adjust the control strategy according to different tasks, avoiding a large amount of manual parameter tuning work in traditional methods and improving the intelligence level of the control system.

[0023] In this application, by constructing the dynamic equations of the robotic arm and generating control commands based on the dynamic solution, the control process meets the requirements of dynamic consistency and system stability, effectively enhancing the robustness of the robotic arm in complex environments.

[0024] In this application, control commands are executed based on real-time solving of dynamic equations, enabling the robotic arm to maintain a compliant response when the external environment changes or when interacting with humans, thereby improving the safety and adaptability of task execution.

[0025] In one implementation, after obtaining the control parameters of the robotic arm based on the large model output in step S101, the method further includes: Perform parameter shaping and constraint projection on the control parameters output by the large model.

[0026] In this way, the large model provides suggested parameters, which are then processed for safety and executability before being distributed, thus avoiding adverse consequences caused by unstable output from the large model.

[0027] Specifically, the stiffness and damping parameters are subject to amplitude and rate of change limits. That is, the magnitudes of the stiffness and damping parameters are limited within preset ranges, and the differences between the stiffness and damping parameters and the output at the previous moment are also limited within preset ranges.

[0028] In this process, stiffness parameters are projected onto a set of symmetric positive definite matrices to perform constraint projection.

[0029] In this process, a low-pass / spline transition is applied to the pose parameters to ensure the continuity of the first and second derivatives of the pose parameters, thereby achieving a smooth pose effect.

[0030] This application provides a spinal surgery planning and execution system for performing the above-described adaptive control method for a robotic arm. The specific scheme of this system is as follows: Figures 2-7 As shown below, the spinal surgery planning and execution system will be described in detail.

[0031] Combination Figure 2 , Figure 3 , Figure 5 As shown, the spinal surgery planning and execution system includes: The data acquisition module 100 is used to acquire and encode the patient's static and dynamic input data. The VLA control center 200 is used to plan surgical parameters for spinal surgery and build a visual language motion model based on the Transformer architecture to generate control parameters for the robotic arm. The robotic arm system 300 is used to execute the robotic arm adaptive control method according to the control parameters of the robotic arm as described in claim 1 or 2.

[0032] Regarding spinal surgery planning and execution systems, it's important to note that existing systems, while improving screw placement accuracy, still suffer from the following core limitations that restrict their intelligence: a lack of "feel" and semantic understanding: existing robot control algorithms (such as traditional PID position control) are "blind" and "rigid." They cannot understand the patient's pathophysiological state. For example, for patients with severe osteoporosis, the axial thrust and gripping force during screw placement should be gentler than in normal bone to prevent screw pullout or fracture. Existing robots cannot read T-values ​​from medical record texts to automatically adjust the robotic arm's force control parameters (impedance / admittance parameters).

[0033] In this application, a visual language action model is used to generate control parameters for the robotic arm, enabling adaptive control of the robotic arm. This allows the model to adaptively adjust the operational force based on the patient's pathological characteristics, solving the problem that existing robots, using fixed control parameters, cannot automatically adjust the stiffness and damping of the robotic arm according to the degree of osteoporosis or anatomical variations.

[0034] In this application, a large model is used to output not only surgical planning schemes, but also to directly output the motion commands (Action Tokens) and impedance control parameters of the robotic arm, achieving a deep integration of "medical theory, mathematics and physics".

[0035] In one implementation, combined with Figure 2 ,Figure 5 As shown, the spinal surgery planning and execution system also includes: The control correction module is used to detect real-time deviations in the control of the robotic arm and generate correction data to correct the trajectory of the robotic arm.

[0036] During the execution of the robotic arm, the LVLA model continuously receives visual feedback from the intraoperative camera. If the visual system detects a deviation in the fusion device implantation angle or slippage (Visual Event), the model generates a new corrective execution semantic unit to adjust the robotic arm trajectory in real time, achieving hand-eye coordination.

[0037] In one implementation, the static input data includes the patient's preoperative medical images and pathological text data; the dynamic input data includes robotic arm end-effector force sensor data, intraoperative registration point cloud, and intraoperative real-time video stream.

[0038] Among them, preoperative medical imaging can be preoperative CT / MRI images to construct a 3D anatomical environment.

[0039] Among them, pathological text data can be electronic medical record texts, etc., to extract bone mineral density T-values, slippage grade, past surgical history, etc.

[0040] In one implementation, in the data acquisition module, encoding is performed by using 3D-ViT to encode image features and Clinical-BERT to encode text features, thereby constructing a high-dimensional multimodal embedding.

[0041] In this application, the data is divided into two categories: image features and text features, and the two categories are encoded in different ways to construct high-dimensional multimodal data as subsequent input.

[0042] In one implementation, combined with Figure 3 The planning and generation module shown includes: The planning layer is used to plan the optimal staple path and fusion device model, as well as the staple placement path, based on the patient's medical images and medical records. The strategy layer is used to build a visual language action model based on the Transformer architecture. It generates control parameters for the robotic arm based on the pin placement path and pathological text data from the planning layer.

[0043] The strategy layer is used to build a VLA model based on the Transformer architecture, the core of which is to regard "robot actions" as a kind of "language".

[0044] The VLA model receives the aforementioned multimodal data and current surgical stage prompts (such as "L4 left screed placement in progress"). These current surgical stage prompts are planned through a planning layer.

[0045] The VLA model outputs a discretized sequence of motion tokens, which are decoded into robotic arm control commands. Each command consists of two parts: trajectory commands (the next pose increment or pose coordinate parameters) and force control commands (the stiffness and damping parameters of the robotic arm).

[0046] The strategy layer generates control parameters for the robotic arm based on the pin placement path from the planning layer and the pathological text data.

[0047] For example, such as Figure 4 As shown, if the planning layer identifies the current bone as "osteoporosis (T-score < -2.5)," the VLA model automatically reduces the axial feed force threshold during the drilling stage and outputs a lower end-effector stiffness parameter, enabling the robotic arm to smoothly retract when encountering sudden changes in resistance and avoid penetration. If the planning layer identifies the current bone as "normal bone," the VLA model outputs a high-stiffness, high-precision mode.

[0048] The control parameters of the robotic arm are processed by the VLA model in the strategy layer based on the input image slice sequence, text prompt, and robotic arm state, and then output.

[0049] The commands output by VLA are input into the Cartesian space impedance controller at the bottom layer of the robotic arm, as described above, to control the robotic arm.

[0050] In this way, the robotic arm exhibits high stiffness in the cortical bone (hard) to ensure accuracy; when entering the cancellous bone or approaching the nerve area (soft / dangerous), the robotic arm automatically switches to a low stiffness, high damping mode to ensure safety.

[0051] Thus, the system described in this application possesses compliant control based on "medical common sense": for the first time, pathological information (such as osteoporosis) in medical records is directly converted into the robot's mechanical control parameters (impedance), giving the robotic arm a "feel" similar to that of a senior doctor and improving surgical safety.

[0052] In this application, the system has end-to-end intelligence: it eliminates cumbersome intermediate rule coding, and the VLA model directly maps from perception data to action instructions, improving the system's adaptability to unstructured environments and unexpected situations.

[0053] In this application, the same system is responsible for both preoperative planning and intraoperative execution, which solves the problems of information loss and coordinate transformation errors that may occur when planning data is imported into the execution device.

[0054] In one embodiment, the planning layer includes a multimodal processing unit, an encoding unit, a cross-modal alignment unit, a coupled planning unit, and a mechanical constraint unit. The multimodal processing unit is used to preprocess the patient's image data and text data; The encoding unit is used to encode and extract preprocessed image and text data to obtain visual feature vectors and text feature vectors. Cross-modal alignment unit, used to align visual feature vectors with text feature vectors; Coupled planning unit, used to couple large model generation with biomechanical constraints, to generate the optimal planning scheme for fusion device and screw; The mechanical verification unit is used to verify the optimal planning scheme and output a planning report.

[0055] It should be noted that existing technologies usually treat pedicle screw planning and cage selection as two separate tasks, ignoring the fact that after the cage is implanted, it will open the intervertebral space, change the physiological curvature and relative position of the vertebral body, thus making the screw path planned based on preoperative static images no longer applicable during surgery (i.e., the "screw-cage decoupling" problem).

[0056] In this way, the patient's medical imaging data and clinical text data are semantically aligned, and the reasoning ability of the multimodal large model is used to collaboratively plan the optimal fusion device model and location, and the pedicle screw path is planned based on the fused vertebral body morphology prediction results.

[0057] In the mechanical verification unit, geometric verification is performed to check whether the planned screws have penetrated the pedicle wall and whether the fusion device has exceeded the endplate range.

[0058] In the mechanical verification unit, the report output includes a 3D planning model containing screw parameters (diameter, length, outward angle, tail tilt angle) and fusion model, and generates a natural language planning report.

[0059] The report includes explanations of the decisions (e.g., "CBT screw trajectory is recommended because the patient is diagnosed with severe osteoporosis and needs to increase the pull-out strength").

[0060] In one implementation, combined with Figure 6 As shown, the encoding unit is configured as follows: The preprocessed image data is input into a 3D visual encoder to extract visual feature vectors containing anatomical structure information. Text data is input into a medical pre-trained language model to extract text feature vectors containing information on pathological risks and surgical constraints.

[0061] The preprocessing process includes: acquiring thin-slice CT scan data and MRI data of the target segment of the patient's spine, constructing a three-dimensional voxel model, and performing image standardization (resampling and normalization) processing.

[0062] The text data refers to: acquiring the patient's electronic medical record (EMR) text and extracting unstructured text information including chief complaint, diagnostic report (including bone mineral density T-score and slippage grade), surgical indications, and past surgical history.

[0063] In the coding unit, the 3D visual encoder can be a variant based on 3D-ViT or 3D-ResNet; anatomical structural information can include pedicle boundaries, cortical bone thickness, intervertebral disc height, etc.

[0064] Among them, the medical pre-trained language model can be ClinicalBERT or a finely tuned LLaMA.

[0065] In one implementation, combined with Figure 7 As shown, the cross-modal alignment unit is configured as follows: Flatten the visual feature vectors into a visual sequence; Calculate the attention weights of the visual features based on the visual sequence and the text feature vector; Visual features are weighted and aggregated based on attention weights to obtain aligned fused features.

[0066] Let F be the extracted visual feature map. v ∈R HxWxDxC (Where H, W, and D are spatial dimensions, and C is the number of channels), flatten it into sequence X. v ∈R NxC； Let the extracted text feature sequence be X. T ∈R MxC () where M is the number of text tokens.

[0067] The attention weight calculation involves using text features as the query and visual features as the key and value to calculate the text-guided visual attention map A. The calculation formula is as follows: Among them, W Q， W K， W V Let be a learnable projection matrix, and dk be a scaling factor. The attention graph A∈R MxN It characterizes the association strength between each text word (e.g., "ossification") and each image voxel.

[0068] Specifically, the calculated attention weights are used to weight and aggregate visual features, and residual connections are introduced to obtain the fused feature M. fused

[0069] This formula ensures that the output features retain both the original anatomical structure information and incorporate the semantic constraints of the medical record text. In one implementation, the coupled planning unit is configured to: acquire morphological adaptation energy terms, safety boundary energy terms, and semantic consistency energy terms; construct the total energy loss based on the morphological adaptation energy terms, safety boundary energy terms, and semantic consistency energy terms; construct the optimal parameter vector based on minimizing the total energy loss; generate planning schemes for the fusion device and screws through a large model; and iteratively optimize the large model generation process based on the optimal parameter vector to obtain the best planning scheme for the fusion device and screws.

[0070] In this application, in addition to relying on the direct generation of large models, an energy loss function based on biomechanical constraints is introduced to find the optimal solution during the generation process.

[0071] The surgical planning problem is modeled as finding the optimal parameter vector Θ*={θ cage θ screw}, making the total energy loss function E total Minimize: The total energy function is defined as: The specific definitions of each item are as follows: The morphological adaptation energy item is used to assess the degree of matching between the predicted intervertebral height and the target physiological curvature after fusion cage implantation. Where is the predicted intervertebral height, is the predicted lordosis angle, and are the ideal anatomical target values. The safety boundary energy term is used to penalize the screw's penetration of the cortical bone. Let P be the set of points on the screw's central axis. For any point p∈P, the distance from p to the nearest inner wall of the cortical bone is defined as d(p): in, Let ϵ be the screw radius and ϵ be the safety margin. This formula indicates that once the distance between the screw surface and the cortical bone is less than the safety margin, the energy value surges.

[0072] Semantic consistency energy term This is used to constrain the planning results to meet specific requirements in the medical record text (e.g., osteoporosis patients need to maximize contact area).

[0073] in, This is the bone mineral density normalization factor extracted from the text (the worse the bone quality, the greater the weight). This represents the bone mineral density integral (Hounsfield Unit sum) traversed along the screw path Ω. This term enables the model to automatically locate high-density bone regions (such as those near the endplate or cortical bone) in osteoporosis cases.

[0074] In this way, by integrating medical record text information, the system can consider the patient's overall condition (such as the degree of osteoporosis) like an expert doctor, thereby developing a personalized plan that is more in line with clinical needs than simple geometric measurements.

[0075] This application innovatively introduces a coupling mechanism of "first Cage fitting - then morphological prediction - then pin path planning", which effectively solves the problem of intraoperative pin path deviation caused by neglecting the fusion device's expansion effect and improves the predictability and accuracy of surgical planning.

[0076] In this application, relying on the generation capabilities of large language models, not only data is provided, but also the rationale for planning, which helps junior doctors understand surgical strategies and has extremely high teaching and clinical guidance value.

[0077] In this application, the technical solution is illustrated using fusion surgery for patients with L4-L5 segment osteoporosis as an example: Preoperative stage: The system reads the patient's CT scan and medical records containing the description of "severe osteoporosis". The VLA model plans the path of the thicker diameter pedicle screw and sets the basic impedance parameters of the entire screw placement process to "low stiffness mode".

[0078] Intraoperative registration: A camera on the robotic arm captures images of the spinal surface, and the VLA model, combined with CT data, completes a vision-based, label-free registration.

[0079] Opening and drilling (action execution): The VLA model outputs commands to control the robotic arm to hold the bone drill bit.

[0080] Semantic intervention: Because the model “remembers” the patient’s osteoporosis, the axial force limiting command output by the model at the moment the drill contacts the bone cortex is 30% lower than that of conventional surgery.

[0081] Anomaly Handling: If the drill bit encounters a bone island (high-density point) causing a sudden increase in force feedback, the VLA model will not force the drill bit forward (to avoid slippage). Instead, it will generate a fine-tuning command in real time, slightly change the drill bit angle to bypass the hard point, and then correct back to the main path.

[0082] Fusioner Implantation (Nail-Cage Coupled Execution): The robotic arm grasps the fusion device. The VLA model controls the robotic arm to perform the dislocation operation based on the planned intervertebral disc height. The dislocation tension is monitored using an end effector force sensor. When the tension reaches the preset ligament limit (inferred by the model based on the patient's age), the dislocation automatically stops and the fusion device is implanted.

[0083] In this application, perception and execution are separated to address the problems that static planning cannot cope with changes in the dynamic environment during surgery and lacks intelligent correction capabilities based on real-time visual feedback.

[0084] In this application, a mapping model of "clinical text semantic features → robotic arm impedance control parameters" was established and applied. Through the VLA large model, abstract medical descriptions (such as "osteoporosis" and "old fracture") were transformed into specific physical control quantities (stiffness matrix and damping matrix), realizing truly personalized compliant control.

[0085] This application proposes a network structure that unifies visual perception, text understanding, and action generation under a single Transformer architecture. This network can not only plan static paths but also output discretized action tokens based on real-time feedback, directly driving a robotic arm to complete multi-task operations such as drilling, nail placement, and spreading.

[0086] In this application, the common sense reasoning ability of the VLA model is used to handle unexpected physical feedback during the operation (such as encountering abnormal bone islands or instrument slippage), and to generate compensation trajectories in real time without interrupting the surgical procedure, thus overcoming the limitation of traditional pre-programmed robots that can only move along fixed paths.

[0087] The spinal surgery planning and execution system provided in the above embodiments of this application corresponds to the robotic arm adaptive control method provided in the embodiments of this application. Therefore, the specific content in the system corresponds to the robotic arm adaptive control method. The specific content can be referred to the records in the robotic arm adaptive control method, which will not be repeated in this application.

[0088] The spinal surgery planning and execution system provided in the above embodiments of this application and the robotic arm adaptive control method provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by their stored applications.

[0089] Based on the same inventive concept, another embodiment of the present invention provides an electronic device for implementing the adaptive control method for a robotic arm described in the above embodiments. Figure 8 As shown, the electronic device includes a memory 301 and a processor 303.

[0090] Memory 301 can be configured to store a program.

[0091] Additionally, memory 301 can also be configured to store various other data to support operation on the electronic device. Examples of this data include instructions for any application or method used to operate on the electronic device, contact data, phonebook data, messages, pictures, videos, etc.

[0092] Memory 301 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. Processor 303, coupled to memory 301, is used to execute programs in memory 301 for: Obtain control parameters of the robotic arm based on the output of a large model; the control parameters include at least stiffness parameters, damping parameters, and pose parameters; Construct the dynamic equations of the robotic arm; Based on the dynamic equations, determine the control commands corresponding to the control parameters; The control commands are executed in real time.

[0093] In this application, Figure 8 The diagram only shows some components and does not mean that the electronic device includes only these components. Figure 8 The components shown.

[0094] The electronic device provided in this embodiment is based on the same inventive concept as the robotic arm adaptive control method provided in this application embodiment, and has the same beneficial effects as the methods adopted, run or implemented by the application stored therein.

[0095] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0096] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0097] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0098] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory. Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0099] This application also provides a computer-readable storage medium corresponding to the adaptive control method for a robotic arm provided in the foregoing embodiments, wherein a computer program (i.e., a program product) is stored thereon, and the computer program, when run by a processor, executes the adaptive control method for a robotic arm provided in any of the foregoing embodiments.

[0100] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, optical disc read-only memory (CDROM), digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0101] The computer-readable storage medium provided in the above embodiments of this application and the robotic arm adaptive control method provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0102] It should be noted that numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of this application may be practiced without these specific details. In some instances, well-known structures and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0103] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, system, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, system, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, system, article, or apparatus that includes said element.

[0104] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An adaptive control method for a robotic arm, characterized in that, include: Obtain the control parameters of the robotic arm based on the output of a large model; The control parameters include at least stiffness parameters, damping parameters, and pose parameters; Construct the dynamic equations of the robotic arm; Based on the dynamic equations, determine the control commands corresponding to the control parameters; The control commands are executed in real time.

2. The adaptive control method for a robotic arm according to claim 1, characterized in that, The dynamic equation is: in, F des To output power, K des For variable stiffness parameters, D des For variable damping parameters, x des For the target pose, x curr For the actual pose, x' des For the target velocity, x ’ curr This refers to the actual speed.

3. A spinal surgery planning and execution system for executing the robotic arm adaptive control method as described in claim 1 or 2, characterized in that, include: The data acquisition module is used to acquire and encode the patient's static and dynamic input data. The VLA control center is used to plan surgical parameters for spinal surgery and build a visual language motion model based on the Transformer architecture to generate control parameters for the robotic arm. A robotic arm system for executing the adaptive control method of the robotic arm as described in claim 1 or 2, based on the control parameters of the robotic arm.

4. The spinal surgery planning and execution system according to claim 3, characterized in that, Also includes: The control correction module is used to detect real-time deviations in the control of the robotic arm and generate correction data to correct the trajectory of the robotic arm.

5. The spinal surgery planning and execution system according to claim 3, characterized in that, The static input data includes the patient's preoperative medical images and pathological text data; the dynamic input data includes the force sensor data at the end of the robotic arm, the intraoperative registration point cloud, and the intraoperative real-time video stream.

6. The spinal surgery planning and execution system according to claim 3, characterized in that, The planning and generation module includes: The planning layer is used to plan the optimal staple path and fusion device model, as well as the staple placement path, based on the patient's medical images and medical records. The strategy layer is used to build a visual language action model based on the Transformer architecture. It generates control parameters for the robotic arm based on the pin placement path and pathological text data from the planning layer.

7. The spinal surgery planning and execution system according to claim 6, characterized in that, The planning layer includes a multimodal processing unit, an encoding unit, a cross-modal alignment unit, a coupled planning unit, and a mechanical constraint unit. The multimodal processing unit is used to preprocess the patient's image data and text data; The encoding unit is used to encode and extract preprocessed image and text data to obtain visual feature vectors and text feature vectors. Cross-modal alignment unit, used to align visual feature vectors with text feature vectors; Coupled planning unit, used to couple large model generation with biomechanical constraints, to generate the optimal planning scheme for fusion device and screw; The mechanical verification unit is used to verify the optimal planning scheme and output a planning report.

8. The spinal surgery planning and execution system according to claim 7, characterized in that, The encoding unit is configured as follows: The preprocessed image data is input into a 3D visual encoder to extract visual feature vectors containing anatomical structure information. Text data is input into a medical pre-trained language model to extract text feature vectors containing information on pathological risks and surgical constraints.

9. The spinal surgery planning and execution system according to claim 7, characterized in that, The cross-modal alignment unit is configured as follows: Flatten the visual feature vectors into a visual sequence; Calculate the attention weights of the visual features based on the visual sequence and the text feature vector; Visual features are weighted and aggregated based on attention weights to obtain aligned fused features.

10. The spinal surgery planning and execution system according to claim 7, characterized in that, The coupled planning unit is configured as follows: Obtain the energy items for morphological adaptation, safety boundary, and semantic consistency; The total energy loss is constructed based on the morphological adaptation energy term, the safety boundary energy term, and the semantic consistency energy term. Construct the optimal parameter vector based on minimizing total energy loss; The planning scheme for the fusion device and screws is generated through a large model; The large model generation process is iteratively optimized based on the optimal parameter vector to obtain the best planning scheme for the fusion device and the screw.

Images