Fracture reduction trajectory planning method, device and equipment of orthopedic robot and medium

Abstract

The invention provides a fracture reduction trajectory planning method, device and equipment of an orthopedic robot and a medium, and relates to the technical field of artificial intelligence and intelligent medical treatment. According to a fracture reduction task, multiple skills are split, and reduction environment state information of the fracture reduction task is obtained; adaptively deciding a skill to be executed from a plurality of skills according to the reset environment state information; calling a dynamic motion primitive model associated with the skill, and generating a sub-track corresponding to the current skill; after the termination condition of the current skill is met, the restoration environment state information is updated according to the final state information of the current skill sub-trajectory, decision and execution of the next skill are carried out until the fracture restoration task is completed, and a complete restoration trajectory generated by joining all the skill sub-trajectories is output; the complete reset track generated through connection meets the motion continuity, and self-adaption, smooth and continuous planning of the reset track of the parallel orthopedic robot in the complex operation environment is achieved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and smart healthcare, and in particular to a method, device, equipment, and medium for fracture reduction trajectory planning in orthopedic robots. Background Technology

[0002] Fractures are a common clinical condition, and the surgical treatment outcome significantly impacts the recovery of limb function. Traditional surgery relies on the surgeon's experience and physical strength, posing a risk of iatrogenic injury due to fatigue or lack of experience. Introducing orthopedic robot-assisted surgery can improve the repeatability of fracture reduction procedures and reduce reliance on the surgeon's physical strength and experience. The motion trajectory planning of orthopedic robots directly affects their motion smoothness and operational safety, and is crucial for achieving safe fracture reduction in complex anatomical environments. Traditional planning methods based on random sampling (such as RRT) often generate unsmooth trajectories. Related technologies, such as the Dynamic Movement Primitives (DMP) method, which learns the surgeon's techniques through instruction, can generate smooth and natural motion trajectories. However, traditional DMP methods lack the ability to dynamically respond to environmental obstacles (such as the proximal fracture fragment) when learning and reproducing trajectories. When there are differences between the actual anatomical structure and the instruction environment, or when path adjustments are needed based on real-time conditions, the trajectories generated by traditional DMP cannot automatically avoid obstacles, resulting in poor safety adaptability in complex surgical environments. Summary of the Invention

[0003] This invention provides a method, device, equipment, and medium for fracture reduction trajectory planning in orthopedic robots, which addresses the shortcomings of traditional orthopedic robot trajectory planning methods in generating trajectories that simultaneously satisfy motion smoothness and achieve safe adaptive trajectories in complex and dynamic fracture reduction environments.

[0004] This invention provides a method for fracture reduction trajectory planning using an orthopedic robot, comprising: Based on the fracture reduction task, multiple skills are broken down, and each skill corresponds to a dynamic motion primitive model. The dynamic motion primitive model is used to generate the sub-trajectory of the corresponding skill. Obtain the reduction environment state information of the fracture reduction task, and adaptively decide the skill to be executed from the multiple skills based on the reduction environment state information; Invoke the dynamic motion primitive model associated with the selected skill to generate the sub-trajectory corresponding to the current skill; After the termination condition of the current skill is met, the reset environment state information is updated according to the final state information of the sub-trajectory corresponding to the current skill, and the decision and execution of the next skill are carried out until the fracture reset task is completed, and the complete reset trajectory generated by the connection of each skill sub-trajectory is output; wherein, the final state of the previous skill sub-trajectory is used as the initial state of the dynamic motion primitive model corresponding to the next skill, so that the complete reset trajectory generated by the connection satisfies motion continuity.

[0005] According to the fracture reduction trajectory planning method for orthopedic robots provided by the present invention, the plurality of skills include at least: Translational obstacle avoidance skills used to move the distal end of a fracture to a middle position that maintains a safe distance from the proximal end of the fracture. Rotational alignment skill for rotating the distal end of the fracture to align with the proximal end of the fracture in the intermediate position; Translational fitting technique for translating the distal end of the fracture along the alignment axis to fit the proximal fracture surface.

[0006] According to the fracture reduction trajectory planning method for orthopedic robots provided by the present invention, the reduction environment state information includes at least one of the following: relative pose and distance between fracture ends, posture alignment error, fracture surface fitting error, contact force information, executed skill sequence, number of skill switching, task completion indicator, obstacle information, and sampled motion speed.

[0007] According to the fracture reduction trajectory planning method for orthopedic robots provided by the present invention, after generating the sub-trajectory corresponding to the current skill, the method further includes real-time collision detection of the sub-trajectory corresponding to the current skill, including: During the generation of the sub-trajectory corresponding to the current skill, the feasible motion speed of the parallel orthopedic robot end effector is sampled in real time; Based on the obstacle information and sampled motion speed in the reset environment state information, the motion trajectory in the future time period is simulated and predicted. Based on the trajectory evaluation function, the motion trajectory within the future time period is evaluated. If the evaluation result indicates a collision risk, the sub-trajectory corresponding to the current skill is adjusted.

[0008] According to the fracture reduction trajectory planning method for orthopedic robots provided by the present invention, the adjustment of the sub-trajectory corresponding to the current skill includes: Using the trajectory evaluation function, a safety assessment is performed on each candidate trajectory, and the alignment degree between each candidate trajectory and the original target of the sub-trajectory corresponding to the current skill is calculated. Based on the results generated by the safety assessment and the alignment, the optimal trajectory is selected from the candidate trajectories, and the corresponding part of the current skill sub-trajectory identified as having a collision risk is replaced by the optimal trajectory.

[0009] According to the fracture reduction trajectory planning method for orthopedic robots provided by the present invention, the step of adaptively deciding on the skill to be executed from the plurality of skills based on the reduction environment state information includes: The reset environment state information is input into a pre-trained high-level policy network in the Option-Critic framework; The high-level policy network outputs the probability that each of the multiple skills is selected as the current skill to be executed; Based on the probability, the skill to be executed is determined.

[0010] According to the fracture reduction trajectory planning method for orthopedic robots provided by the present invention, after outputting the complete reduction trajectory generated by connecting the various skill sub-trajectories, the method further includes: A multi-objective optimization model is constructed with the goal of minimizing the sum of squared velocities, sum of squared accelerations, and sum of squared jerkes of each kinematic branch of the parallel orthopedic robot on the complete reset trajectory. A multi-objective evolutionary algorithm is used to solve the multi-objective optimization model under preset time, velocity, acceleration and displacement constraints to obtain the optimized smooth reset trajectory.

[0011] The present invention also provides a fracture reduction trajectory planning device for a parallel orthopedic robot, comprising: The splitting module is used to split the fracture reduction task into multiple skills, each skill corresponding to a dynamic motion primitive model, and the dynamic motion primitive model is used to generate the sub-trajectory of the corresponding skill. The decision module is used to obtain the reduction environment status information of the fracture reduction task, and adaptively decide the skill to be executed from the multiple skills based on the reduction environment status information. The generation module is used to call the dynamic motion primitive model associated with the selected skill to generate the sub-trajectory corresponding to the current skill; The output module is used to update the reset environment state information according to the final state information of the sub-trajectory corresponding to the current skill after the termination condition of the current skill is met, and to make decisions and execute the next skill until the fracture reset task is completed, and output the complete reset trajectory generated by connecting the sub-trajectories of each skill; wherein, the final state of the previous skill sub-trajectory is used as the initial state of the dynamic motion primitive model corresponding to the next skill, so that the complete reset trajectory generated by the connection satisfies motion continuity.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the fracture reduction trajectory planning method for an orthopedic robot as described in any of the preceding claims.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the fracture reduction trajectory planning method for the orthopedic robot described in any of the preceding claims.

[0014] The present invention provides a method, apparatus, device, and medium for fracture reduction trajectory planning in orthopedic robots. This method involves breaking down the fracture reduction task into multiple skills, each corresponding to a dynamic motion primitive model (MPM), which generates sub-trajectories for the corresponding skill. The method acquires the reduction environment state information for the fracture reduction task and adaptively decides which skill to execute from among the multiple skills based on this information. It then calls the MPM associated with the selected skill to generate the sub-trajectory corresponding to the current skill. Finally, after the termination condition of the current skill is met, the reduction environment is updated based on the final state information of the sub-trajectory corresponding to the current skill. The system processes the state information and makes decisions and executes the next skill until the fracture reduction task is completed, outputting the complete reduction trajectory generated by connecting the sub-trajectories of each skill. The final state of the previous skill sub-trajectory is used as the initial state of the dynamic motion primitive model corresponding to the next skill, so that the complete reduction trajectory generated by the connection satisfies motion continuity. This achieves adaptive and smooth continuous planning of the reduction trajectory of the parallel orthopedic robot in complex surgical environments, decomposes the complex reduction task into basic skill units that can be flexibly combined and scheduled, and effectively solves the problem in the existing technology that trajectory planning is difficult to simultaneously take into account real-time motion smoothness and scene adaptability through end-to-end intelligent decision-making. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is one of the flowcharts of the fracture reduction trajectory planning method for orthopedic robots provided in the embodiments of the present invention; Figure 2 This is the second flowchart of the fracture reduction trajectory planning method for orthopedic robots provided in the embodiments of the present invention; Figure 3This is the third flowchart of the fracture reduction trajectory planning method for orthopedic robots provided in this embodiment of the invention; Figure 4 This is the fourth flowchart of the fracture reduction trajectory planning method for orthopedic robots provided in the embodiments of the present invention; Figure 5 This is a functional structure diagram of the fracture reduction trajectory planning device for the parallel orthopedic robot provided in an embodiment of the present invention; Figure 6 This is a functional structure diagram of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0018] Figure 1 A flowchart of the fracture reduction trajectory planning method for orthopedic robots provided in this embodiment of the invention is shown below. Figure 1 As shown, the fracture reduction trajectory planning method for orthopedic robots provided in this embodiment of the invention includes: Step 101: Based on the fracture reduction task, break down the task into multiple skills, each skill corresponding to a dynamic motion primitive model, which is used to generate the sub-trajectory of the corresponding skill. In this embodiment of the invention, the Dynamic Movement Primitives (DMP) model can be expressed as the following differential equation: ; in, As a time scale factor, Current position At the current speed, Let z be the current acceleration and z be the target position. Characterizes the stiffness of the system. Characterizing damping properties, This is a nonlinear forcing term.

[0019] Nonlinear forcing term The complex dynamics used to capture the teaching trajectory can be defined as: ; in, basis functions In the phase variable The non-negative Gaussian radial basis functions defined above are used to ensure the smoothness and continuity of the generated trajectory. Indicates the first basis functions The weighting coefficients can be obtained through regression learning on the teaching trajectory. Describing basis functions The quantity of phase variables. The evolution process can be given by the following first-order linear differential equation: ; in, It is a parameter that controls the convergence speed of the phase variable. For phase variables The first-order linear differential.

[0020] Step 102: Obtain the reduction environment status information of the fracture reduction task, and adaptively decide the skill to be executed from the multiple skills based on the reduction environment status information; In this embodiment of the invention, the repositioning environment status information includes at least one of the following: relative pose and distance between fracture ends, posture alignment error, fracture surface fit error, contact force information, executed skill sequence, number of skill switching, task completion indicator, obstacle information, and sampled motion speed.

[0021] Step 103: Invoke the dynamic motion primitive model associated with the selected skill to generate the sub-trajectory corresponding to the current skill; Step 104: After the termination condition of the current skill is met, update the reset environment state information according to the final state information of the sub-trajectory corresponding to the current skill, and make decisions and execute the next skill until the fracture reset task is completed, and output the complete reset trajectory generated by connecting the sub-trajectories of each skill; wherein, the final state of the previous skill sub-trajectory is used as the initial state of the dynamic motion primitive model corresponding to the next skill, so that the complete reset trajectory generated by connecting satisfies motion continuity.

[0022] Traditional planning methods based on random sampling often generate uneven trajectories; imitation learning-based planning methods, such as dynamic motion primitives, lack the ability to dynamically respond to environmental obstacles (such as proximal fracture fragments) when learning and reproducing trajectories. When there are differences between the actual anatomical structure and the teaching environment, or when the path needs to be adjusted according to real-time conditions, the generated trajectory has poor safety adaptability in complex surgical environments.

[0023] The fracture reduction trajectory planning method for orthopedic robots provided in this invention involves: dividing the fracture reduction task into multiple skills, each corresponding to a dynamic motion primitive model (MPMA), which generates sub-trajectories for the corresponding skill; acquiring the reduction environment state information of the fracture reduction task; adaptively deciding the skill to be executed from the multiple skills based on the reduction environment state information; calling the MPMA associated with the selected skill to generate the sub-trajectory corresponding to the current skill; and updating the reduction environment state information based on the final state information of the sub-trajectory corresponding to the current skill after the termination condition of the current skill is met. The system receives information and makes decisions and executes the next skill until the fracture reduction task is completed, outputting the complete reduction trajectory generated by connecting the sub-trajectories of each skill. The final state of the previous skill sub-trajectory is used as the initial state of the dynamic motion primitive model corresponding to the next skill, so that the complete reduction trajectory generated by the connection satisfies motion continuity. This realizes adaptive and smooth continuous planning of the reduction trajectory of the parallel orthopedic robot in complex surgical environments, decomposes the complex reduction task into basic skill units that can be flexibly combined and scheduled, and effectively solves the problem in the existing technology that trajectory planning is difficult to simultaneously take into account real-time motion smoothness and scene adaptability through end-to-end intelligent decision-making.

[0024] Based on any of the above embodiments, the plurality of skills include at least: Translational obstacle avoidance skills used to move the distal end of a fracture to a middle position that maintains a safe distance from the proximal end of the fracture. Rotational alignment skill for rotating the distal end of the fracture to align with the proximal end of the fracture in the intermediate position; Translational fitting technique for translating the distal end of the fracture along the alignment axis to fit the proximal fracture surface.

[0025] In this embodiment of the invention, the fracture reduction trajectory includes three core skills that can be flexibly selected and switched according to the actual situation during surgery.

[0026] Translational obstacle avoidance technique: The distal fracture fragment is translated to a mid-position that maintains a safe distance from the proximal fracture fragment and surrounding anatomical structures, creating a collision-free space for subsequent rotational manipulation. This process must avoid collisions between the distal and proximal fracture fragments and surrounding soft tissues.

[0027] Rotational alignment technique: At the intermediate position reached, rotate only the distal end of the fracture to align its posture with that of the proximal end, thereby aligning the axis of the distal end with the axis of the proximal end.

[0028] Translational fitting technique: Based on the alignment of the axes, translate the distal end of the fracture along the aligned axes to make the fracture surface of the distal end fit precisely with that of the proximal end of the fracture.

[0029] The above three skills can be flexibly combined and used according to the real-time situation during the operation. For example, in the case of a simple fracture, reduction can be achieved directly by translation and fitting. In the case of a complex fracture, it is necessary to repeatedly adjust the distal end of the fracture and perform multiple skills until the ideal reduction effect is achieved.

[0030] Based on any of the above embodiments, after generating the sub-trajectory corresponding to the current skill, the method further includes real-time collision detection of the sub-trajectory corresponding to the current skill, including: Step 201: During the generation of the sub-trajectory corresponding to the current skill, the feasible motion speed of the parallel orthopedic robot end effector is sampled in real time; Step 202: Based on the obstacle information and sampled motion speed in the reset environment state information, simulate and predict the motion trajectory in the future time period; Step 203: Based on the trajectory evaluation function, evaluate the motion trajectory within the future time period. If the evaluation result indicates a collision risk, then trigger the adjustment of the sub-trajectory corresponding to the current skill.

[0031] In this embodiment of the invention, adjusting the sub-trajectory corresponding to the current skill includes: Step 301: Using the trajectory evaluation function, perform a safety assessment on each candidate trajectory and calculate the alignment degree between each candidate trajectory and the original target of the sub-trajectory corresponding to the current skill. In this embodiment of the invention, the risk of collision between the distal and proximal ends of the fracture and soft tissues is calculated in real time during trajectory generation. If a high collision risk is detected, a trajectory adjustment mechanism is triggered. The trajectory adjustment mechanism includes two methods: local trajectory correction and global replanning. When the trajectory is locally in a high collision risk state, the trajectory is locally corrected to avoid obstacles using an improved Dynamic Window Approach (DWA). If the locally corrected trajectory is still in a high collision risk state, global replanning is triggered, and high-level and low-level strategies are re-run according to the current state to generate a new skill trajectory. A safety detection module runs throughout the entire trajectory generation process to ensure the safety of the final reset trajectory.

[0032] The improved DWA algorithm independently samples linear velocity and angular velocity for translational and rotational skill trajectories, and selects the optimal velocity based on their respective trajectory evaluation functions to achieve obstacle avoidance.

[0033] When processing translational skill trajectories, the algorithm generates a set of feasible linear velocity ranges based on the current linear velocity, combined with acceleration limits and braking distances. This range includes all candidate linear velocities from the minimum linear velocity to the maximum linear velocity.

[0034] Step 302: Based on the results generated by the safety assessment and the alignment, select the optimal trajectory from the candidate trajectories, and replace the corresponding part of the current skill sub-trajectory that is identified as having a collision risk with the optimal trajectory.

[0035] In this embodiment of the invention, for each candidate linear velocity, starting from the center position of the distal fracture surface at the current fracture site, a unit vector is used to simulate the translational trajectory of that center point over a future period. A comprehensive score is calculated for each candidate trajectory using a translational trajectory evaluation function. The translational trajectory evaluation function consists of three weighted parts: the first part reflects the alignment between the trajectory endpoint and the desired target position; the second part reflects the minimum distance between the trajectory point sequence and the distal fracture site; and the third part is directly related to the magnitude of the linear velocity.

[0036] When processing the trajectory of a spinning skill, the algorithm generates a set of feasible angular velocities based on the current angular velocity, combined with angular acceleration limits and braking distance. This range includes all candidate angular velocities from the minimum to the maximum angular velocity, with each candidate angular velocity being a three-dimensional vector.

[0037] For each candidate angular velocity, starting from the current orientation of the center of the distal fracture surface, a sequence of orientations for the rotational trajectory of that center over a future period is simulated. A comprehensive score is calculated for each candidate trajectory using a rotational trajectory evaluation function. This function consists of two weighted parts: the first part reflects the alignment between the final orientation of the rotational trajectory and the desired orientation; the second part assesses the risk of collision between the distal and proximal fracture fragments or surrounding soft tissues during rotation.

[0038] When performing local trajectory correction, the algorithm selects the linear velocity and angular velocity that maximize the scores of their respective evaluation functions from candidate velocities for translation and rotation. This strategy aims to ensure the safety of the local motion trajectory while making the trajectory's endpoint as close as possible to the desired position and attitude.

[0039] Based on any of the above embodiments, the step of adaptively deciding on the skill to be executed from the plurality of skills according to the reset environment state information includes: Step 401: Input the reset environment state information into the pre-trained high-level policy network in the Option-Critic framework; Step 402: The high-level policy network outputs the probability that each of the multiple skills is selected as the current skill to be executed; Step 403: Based on the probability, decide on the skill to be executed.

[0040] In this embodiment of the invention, the complex fracture reduction trajectory planning task is decomposed into three flexibly combinable skills, each of which generates a trajectory using a Dynamic Motion Primitive Model (DMP). The Option-Critic framework can autonomously decide which skill to execute and when to switch skills based on real-time intraoperative conditions, adapting to individual differences and intraoperative changes, thereby generating a safe and continuous reduction trajectory. The complete mechanism of the Option-Critic hierarchical reinforcement learning framework working collaboratively with the Dynamic Motion Primitive Model (DMP) is as follows: Figure 2 As shown, the system adaptively decides the skill to be executed, such as translational obstacle avoidance, rotational alignment, or translational fitting, based on the real-time perceived state of the reduction environment (e.g., relative pose and distance of the fracture ends). This is achieved through a high-level policy network within the Option-Critic framework. After the decision, the system calls the DMP model bound to the skill. This model uses the previous skill's termination state as the starting point and the desired termination state of the current skill as the target, generating a sub-trajectory with inherent smoothness. During trajectory generation, the system simultaneously performs real-time collision detection and local adjustments (e.g., using an improved dynamic window method) to ensure the safety of the trajectory in the complex anatomical environment. Once the trajectory is verified for safety, it is saved, and the system state is updated accordingly, triggering the next round of skill decision-making and trajectory generation, forming a closed-loop iteration of "perception-decision-generation-detection". Finally, all sequentially generated and safety-verified skill sub-trajectories are seamlessly spliced ​​together using the previous final state as the next initial state, forming a globally continuous, safe, collision-free, and smoothly moving complete fracture reduction trajectory.

[0041] In the Option-Critic framework, the state space contains states. This includes the position and posture information of the distal fracture relative to the proximal fracture, the distance between the center of the distal fracture surface and the center of the proximal fracture surface, posture alignment error, contact force, target pose, current skill execution identifier, historical skill sequence, number of skill switching, and task completion rate.

[0042] The Option-Critic framework uses three reduction skill options corresponding to three fracture reduction operations. These three reduction skills can be executed in any order and can be repeated, as shown below: ,in, This refers to the skill of lateral movement and obstacle avoidance, which is responsible for moving the distal end of a fracture to a safe intermediate position. This refers to the rotational alignment skill, which is responsible for rotating the distal end of the fracture at the midpoint to align the distal and proximal ends of the fracture along their axes. This refers to the translational fitting skill, responsible for translating the distal fracture fragments along the alignment axis to achieve surface fitting between the distal and proximal fracture fragments. This is a high-level strategy within the Option-Critic framework during fracture reduction. Learn how to select the most appropriate skill based on the current reset state. Each reset skill... All include internal strategies and termination function Internal strategy Learn how to choose specific actions when performing the current skill. Termination function. Learn how to design a skill that terminates the current skill, allowing a switch to the next skill.

[0043] The core of the Option-Critic framework consists of three main functions: the option value function, the option-action value function, and the arrival value function.

[0044] The option value function represents the current reset state. Select and execute the reset skill option below. The expected total value is used to guide high-level strategy selection of the reset skill options most beneficial to the current reset phase. The option value function can be expressed as the following mathematical equation.

[0045]

[0046] in, Indicates the current reset state Next, reset skill options Internal strategy for selecting specific reset actions The probability of. This is the option-action value function. The value is determined by all possible reset actions. The weighted average is obtained.

[0047] Option-Action Value Function indicates the current reset state. Next, execute the reset skill option. Specific reset actions are taken at that time. The value of an option-action value function can be expressed as the following mathematical equation.

[0048]

[0049] in, This indicates the immediate feedback of the reset action. This represents the discount factor, with values ​​ranging from 1 to 2. . Let be the state transition probability, representing the state transition probability in the current reset state. Perform a reset action Then, it transitions to the new reset state. The probability of. To reach the value function. Indicates the execution of a reset action. After that, all possible next reset states The expected value of the discount.

[0050] Reaching the value function indicates entering a new reset state. Then, continue with the reset skill option. The value of [the value]. The value function can be expressed as the following mathematical equation:

[0051] in, Indicates the new reset state Next, terminate the reset skill option. The probability of. Indicates the new reset state Below, do not terminate the reset skill option. The probability of. Indicates the new reset state Continue executing the reset skill option The total value. Indicates the new reset state If the reset skill option is not terminated... The expected rewards of continuing to use this reset skill option. This indicates the new reset state. The highest value among all reset skill options. Indicates the new reset state If the reset skill option is terminated... The function calculates the expected reward obtained by reselecting the optimal reset skill option after termination. It implements adaptive switching between reset skills during the reset process, ensuring that the switch to the next reset skill option is performed at the appropriate time.

[0052] In each decision-making step, the Option-Critic framework selects the skill option to be executed through its high-level policy network based on real-time reset state information, including position, attitude, distance, alignment error, etc. The selection of skill options is not restricted by a fixed order, but is dynamically determined based on the current state and task completion status.

[0053] (1) Translational obstacle avoidance skill: The distal fracture fragment is translated to an intermediate position, maintaining a safe distance from the proximal fracture fragment and surrounding anatomical structures. An expert teaching trajectory for this skill is learned using a DMP (Data Management Platform), generating a trajectory from the initial position to the intermediate position. The generated trajectory must meet obstacle avoidance requirements, preventing collisions between the distal fracture fragment and the proximal fracture fragment and surrounding soft tissues. When the distal fracture fragment reaches near the intermediate position and the distance from the proximal fracture fragment and soft tissues exceeds a safe threshold, the termination condition function outputs a high probability, ending the skill. The internal strategy for training this skill is implemented using a translational obstacle avoidance skill reward function. Encourage the distal and proximal ends of the fracture to maintain a safe distance from the soft tissues to avoid collisions and allow for smooth movement.

[0054] Translational obstacle avoidance skill reward function The mathematical representation of is as follows:

[0055] in, , These are the weighting coefficients for each reward item, and all of them have positive values. It indicates the minimum collision distance between the distal and proximal ends of the fracture and soft tissues. , indicating based on acceleration The trajectory smoothness score is given, where N represents the total number of trajectory points. , All represent scaling factors, and all values ​​are positive.

[0056] DMP learns this reset skill and ultimately generates an obstacle avoidance trajectory from the initial position to a safe intermediate position, creating collision-free spatial conditions for subsequent rotation operations.

[0057] (2) Rotation Alignment Skill: At the intermediate position, the distal fracture fragment is rotated to align with the proximal fracture fragment, achieving axis alignment. A DMP is used to learn the rotation trajectory taught by an expert for this skill, generating a rotation trajectory from the initial posture to the target posture. The generated trajectory achieves the rotation of the distal fracture fragment to the proximal fracture fragment alignment posture while preventing collisions between the distal and proximal fracture fragments and soft tissues. When the posture alignment error between the distal and proximal fracture fragments is less than a threshold, the termination condition function outputs a high probability, terminating this skill. When training the internal strategy of this skill, the rotation alignment skill reward function is used. Encourage minimizing orientation alignment errors, avoiding collisions during rotation, and maintaining a smooth rotation process.

[0058] Rotation Alignment Skill Reward Function The mathematical representation of is as follows:

[0059] in, , , These are the weighting coefficients for each reward item, and all of them have positive values. This indicates the orientation alignment error. This indicates a penalty for collision risk during the rotation process. This indicates the angular acceleration penalty term. , , All represent scaling factors, and all values ​​are positive.

[0060] DMP learns this reduction skill to achieve precise alignment of the distal and proximal fracture positions, creating conditions for subsequent translational fitting.

[0061] (3) Translational Fitting Skill: Along the aligned axis, the distal fracture fragment is translated to fit against the proximal fracture surface. A DMP is used to learn the translational trajectory taught by an expert for this skill, generating a straight-line trajectory along the axis from the intermediate position to the reduction position. When the positional error between the distal and proximal fracture fragments is less than a threshold and the contact force is within a safe range, the termination condition function outputs a high probability, terminating this skill. The internal strategy for training this skill uses the translational fitting skill reward function. Encourage minimizing positional errors in the fit, maintaining moderate contact force, and avoiding excessive force.

[0062] Translation and Fitting Skill Reward Function The mathematical representation of is as follows:

[0063] in, , , These are the weighting coefficients for each reward item, and all of them have positive values. This indicates the positional fit error. This indicates the quality score of the contact force, encouraging that the force is within a safe range. It indicates the impact and punishment of force. , This represents the scaling factor, and all of its values ​​are positive.

[0064] DMP learns this reduction skill to ultimately achieve precise alignment of the fracture surface.

[0065] (4) Global reward function When training high-level policies in the Option-Critic framework, a global reward function is used. To optimize overall task completion efficiency, operation time, and skill switching frequency, the system is guided to intelligently combine and schedule multiple skill options to safely and efficiently complete fracture reduction tasks. A global reward function is designed. To coordinate the learning of the above three repositioning skills, the specific mathematical representation is as follows:

[0066] in, This indicates a function that indicates task completion. This indicates the total duration of the reset trajectory. This indicates the number of times the skill switch has been reset. , , These are the weight parameters. , This represents the scaling factor, and all of its values ​​are positive.

[0067] The high-level policy dynamically selects the currently executed skill. When the termination function indicates that the current skill should be terminated, the high-level policy selects the next skill based on the new state, continuing until the task is completed, i.e., the position and attitude errors are both less than the threshold, and the contact force is within a safe range. During the generation of each skill trajectory, the safety detection module checks and adjusts the skill trajectory in real time. To ensure the smoothness of the trajectory during skill switching, the initial state when the next skill's DMP generates its trajectory is set to the termination state of the previous skill trajectory, ensuring the continuity of position and velocity. Finally, the skill trajectory segments generated by a series of dynamically selected skills are smoothly connected through the continuity constraint of the initial and termination states, forming a time-parameterized complete reset trajectory that meets the requirements of safety, continuity, and accuracy.

[0068] Based on any of the above embodiments, after outputting the complete reset trajectory generated by connecting the various skill sub-trajectories, the method further includes: A multi-objective optimization model is constructed with the goal of minimizing the sum of squared velocities, sum of squared accelerations, and sum of squared jerkes of each kinematic branch of the parallel orthopedic robot on the complete reset trajectory. A multi-objective evolutionary algorithm is used to solve the multi-objective optimization model under preset time, velocity, acceleration and displacement constraints to obtain the optimized smooth reset trajectory.

[0069] In this embodiment of the invention, under multiple constraints of time, velocity, acceleration, and displacement, a second-generation non-dominated sorting genetic algorithm (NSGA-II) is used to optimize the smoothness of the reset trajectory. This is achieved by simultaneously minimizing the sum of squared velocities, sum of squared accelerations, and sum of squared jerk accelerometers of each branch, thus smoothing the overall trajectory. Figure 3 As shown, the input to the multi-objective evolutionary algorithm is a pre-planned complete reset trajectory (specifically, the displacement sequence of each branch of the parallel robot), from which the kinematic quantities such as velocity, acceleration, and jerk of each branch are derived. The optimization objective is to simultaneously minimize the sum of squares of acceleration and jerk of all branches on the entire trajectory (actually, the sum of squares of velocity, acceleration, and jerk) to minimize motion shocks and jitter and improve smoothness. The entire optimization process must be carried out under strict time, velocity, acceleration, and displacement constraints to ensure that the optimized trajectory is not only smoother but also still meets the physical limits of the robot and the spatial requirements of the surgery. To this end, the system uses the NSGA-II multi-objective evolutionary algorithm to solve this complex optimization problem with multiple constraints, ultimately outputting an optimized trajectory that achieves the best balance between safety and motion stability.

[0070] Set the time step vector as the optimization variable: ,in, For the number of trajectory points, This represents the m-th time interval.

[0071] The orthopedic robot used has a total of 6 branches. The displacement of each branch of the parallel robot can be calculated based on the reset trajectory generated by the Option-Critic framework and DMP. ; in, Indicates the first ( ) branch in the ( The displacement of ) trajectory points.

[0072] Branch speeds acceleration accelerometer It can be represented as:

[0073]

[0074]

[0075] Larger branch velocities, accelerations, and jerks all increase the risk of loss of control in orthopedic robots, leading to severe mechanical impacts and negatively impacting the smoothness of the repositioning motion. Therefore, the optimization objectives are set as the sum of squares of branch velocities, accelerations, and jerks, which can be expressed as:

[0076]

[0077]

[0078] The constraints include time constraints, velocity constraints, acceleration constraints, and displacement constraints.

[0079] Time constraints include time step constraints and total trajectory duration constraints. The time step constraint is between a given minimum and maximum time step to ensure the optimized time step remains within acceptable limits. The total trajectory duration constraint is between a given minimum and maximum total duration to ensure the optimized total trajectory duration remains within acceptable limits. The branch velocity constraint is between a given minimum and maximum branch velocity to ensure the optimized branch velocity remains within acceptable limits. The branch acceleration constraint is between a given minimum and maximum branch acceleration to ensure the optimized branch acceleration remains within acceptable limits.

[0080] The branch displacement constraint ensures that the difference between the optimized branch displacement and the original branch displacement is minimized, guaranteeing that the path points of the optimized reset trajectory are the same as those of the unoptimized trajectory, thus preserving the obstacle avoidance performance of the original trajectory. Using the NSGA-II multi-objective optimization algorithm, while satisfying the above multiple constraints, the sum of squared velocities, sums of squared accelerations, and sums of squared jerk accelerometers of the branches are minimized simultaneously, thereby optimizing the smoothness of the reset trajectory while maintaining obstacle avoidance performance.

[0081] like Figure 4 As shown, the fracture reduction trajectory planning method for orthopedic robots provided in this embodiment of the invention includes: Phase 1: Skill Decision: The Option-Critic high-level strategy selects the skill to be executed from the skill library based on the current state.

[0082] Initialization: The system loads the teaching trajectory of the current skill and sets the starting point (the termination state of the previous skill) and the target (the expected termination state of the current skill) for trajectory generation.

[0083] Phase Two: Trajectory generation: The DMP model learns the teaching trajectory of the skill and generates a sub-trajectory of the current skill based on the given start and end points.

[0084] Safety Check: Immediately perform a safety check on the generated trajectory. If successful: Save the trajectory, interact with the environment to receive a reward, and then update the current state. If unsuccessful: Proceed to a local correction subprocess. Perform the safety check again after correction. If it still fails, it may be necessary to replan or adjust parameters.

[0085] Skill Termination Check: Check if the termination function of the current skill has been triggered (e.g., whether the target pose has been reached or the termination condition has been met). If not triggered: Return to the trajectory generation step and continue executing the current skill. If triggered: Save the current skill trajectory and proceed to the next stage.

[0086] Phase Three: Task Judgment: Determine if the current skill trajectory is the last segment (i.e., whether the reset task is complete). If not: After updating the current state, return to the first stage, and the Option-Critic strategy selects the next skill to begin a new outer loop iteration. If complete: Piece together all the sequentially saved historical skill trajectories to obtain a complete reset trajectory.

[0087] Phase Four: Kinematic calculation: Based on the complete reset trajectory, the displacement, velocity, acceleration and jerk of each branch are calculated using the robot's kinematic model.

[0088] Multi-objective optimization: Under multiple constraints including time, velocity, acceleration, and displacement, the NSGA-II algorithm is used to optimize the trajectory. The optimization objective is to simultaneously minimize the sum of squared velocities, sums of squared accelerations, and sums of squared jerkes for all branches, in order to achieve overall motion smoothness. The optimized reset trajectory is then obtained, and the process ends.

[0089] Parallel Process: Throughout the process, whenever a skill completes its execution (i.e., after the termination function is triggered), the Option-Critic framework uses the obtained reward and state transition information to evaluate the policy gradient of that skill and updates the parameters of its higher-level policy along the gradient direction. This means that while using the policy, the system is also continuously learning and optimizing the policy itself, making its decisions increasingly intelligent.

[0090] The fracture reduction trajectory planning method for orthopedic robots provided in this invention utilizes an Option-Critic framework and a Displacement Manipulation (DMP)-based approach. Based on the fracture condition, the DMP learns translational obstacle avoidance, rotational alignment, or translational fitting skills and generates skill trajectories. When a skill trajectory poses a collision risk, an improved DWA algorithm is used to locally correct it. By sequentially combining skill trajectories, a collision-free and safe reduction path is planned between the distal and proximal fracture ends and soft tissues in complex anatomical environments. Furthermore, this method transforms the smoothness requirement into a kinematic constraint and constructs a multi-objective optimization problem. The NSGA-II algorithm is used to solve this multi-objective optimization problem, obtaining a reduction trajectory that smooths the motion of each branch of the parallel robot, thus improving the stability of the parallel robot's reduction motion. The proposed trajectory planning method can simultaneously meet the requirements of safety and smoothness in reduction motion.

[0091] It should be noted that the method provided by this invention has good adaptability and can plan the reduction trajectory of the distal end of the fracture in different fracture situations. The method provided by this invention can also provide a technical reference for the motion planning of other surgical robots.

[0092] Figure 5 This is a schematic diagram of the fracture reduction trajectory planning device for a parallel orthopedic robot provided in an embodiment of the present invention, as shown below. Figure 5 As shown, the fracture reduction trajectory planning device for the parallel orthopedic robot provided in this embodiment of the invention includes: The splitting module 501 is used to split multiple skills according to the fracture reduction task, each skill corresponding to a dynamic motion primitive model, and the dynamic motion primitive model is used to generate the sub-trajectory of the corresponding skill. The decision module 502 is used to obtain the reduction environment status information of the fracture reduction task, and adaptively decide the skill to be executed from the multiple skills based on the reduction environment status information. The generation module 503 is used to call the dynamic motion primitive model associated with the selected skill to generate the sub-trajectory corresponding to the current skill; The output module 504 is used to update the reset environment state information according to the final state information of the sub-trajectory corresponding to the current skill after the termination condition of the current skill is met, and to make decisions and execute the next skill until the fracture reset task is completed, and output the complete reset trajectory generated by connecting the sub-trajectories of each skill; wherein, the final state of the previous skill sub-trajectory is used as the initial state of the dynamic motion primitive model corresponding to the next skill, so that the complete reset trajectory generated by connecting satisfies motion continuity.

[0093] The fracture reduction trajectory planning device for a parallel orthopedic robot provided in this invention splits the fracture reduction task into multiple skills, each corresponding to a dynamic motion primitive model. The dynamic motion primitive model is used to generate sub-trajectories for the corresponding skill. The device acquires the reduction environment state information of the fracture reduction task and adaptively decides which skill to execute from the multiple skills based on this information. It then calls the dynamic motion primitive model associated with the selected skill to generate the sub-trajectory corresponding to the current skill. After the termination condition of the current skill is met, the device updates the reduction environment state based on the final state information of the sub-trajectory corresponding to the current skill. The system processes information and makes decisions and executes the next skill until the fracture reduction task is completed, outputting the complete reduction trajectory generated by connecting the sub-trajectories of each skill. The final state of the previous skill sub-trajectory is used as the initial state of the dynamic motion primitive model corresponding to the next skill, so that the complete reduction trajectory generated by the connection satisfies motion continuity. This realizes adaptive and smooth continuous planning of the reduction trajectory of the parallel orthopedic robot in complex surgical environments, decomposes the complex reduction task into basic skill units that can be flexibly combined and scheduled, and effectively solves the problem in the existing technology that trajectory planning is difficult to simultaneously take into account real-time motion smoothness and scene adaptability through end-to-end intelligent decision-making.

[0094] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The memory 630 includes a computer program, an operating system, and acquired data. The processor 610 can call the logical instructions in the memory 630 to execute a fracture reduction trajectory planning method for a parallel orthopedic robot. This method includes: splitting the fracture reduction task into multiple skills, each skill corresponding to a dynamic motion primitive model, the dynamic motion primitive model being used to generate a sub-trajectory for the corresponding skill; acquiring the reduction environment state information of the fracture reduction task, and adaptively deciding on the skill to be executed from the multiple skills based on the reduction environment state information; calling the dynamic motion primitive model associated with the selected skill to generate the sub-trajectory corresponding to the current skill; after the termination condition of the current skill is met, updating the reduction environment state information based on the final state information of the sub-trajectory corresponding to the current skill, and making a decision and executing the next skill until the fracture reduction task is completed, and outputting a complete reduction trajectory generated by connecting the sub-trajectories of each skill; wherein, the final state of the previous skill sub-trajectory is used as the initial state of the dynamic motion primitive model corresponding to the next skill, so that the complete reduction trajectory generated by connecting the sub-trajectories satisfies motion continuity.

[0095] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to related technologies, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0096] On the other hand, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a method for planning fracture reduction trajectory of a parallel orthopedic robot provided by the above methods. The method includes: splitting multiple skills according to the fracture reduction task, each skill corresponding to a dynamic motion primitive model, the dynamic motion primitive model being used to generate sub-trajectories of the corresponding skill; acquiring reduction environment state information of the fracture reduction task, adaptively deciding the skill to be executed from the multiple skills according to the reduction environment state information; calling the dynamic motion primitive model associated with the selected skill to generate the sub-trajectory corresponding to the current skill; after satisfying the termination condition of the current skill, updating the reduction environment state information according to the final state information of the sub-trajectory corresponding to the current skill, and making a decision and executing the next skill until the fracture reduction task is completed, and outputting a complete reduction trajectory generated by connecting the sub-trajectories of each skill; wherein, the final state of the previous skill sub-trajectory is used as the initial state of the dynamic motion primitive model corresponding to the next skill, so that the complete reduction trajectory generated by connecting the sub-trajectories satisfies motion continuity.

[0097] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0098] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fracture reduction trajectory planning method for a parallel orthopedic robot, characterized in that, include: Based on the fracture reduction task, multiple skills are broken down, and each skill corresponds to a dynamic motion primitive model. The dynamic motion primitive model is used to generate the sub-trajectory of the corresponding skill. Obtain the reduction environment state information of the fracture reduction task, and adaptively decide the skill to be executed from the multiple skills based on the reduction environment state information; Invoke the dynamic motion primitive model associated with the selected skill to generate the sub-trajectory corresponding to the current skill; After the termination condition of the current skill is met, the reset environment state information is updated according to the final state information of the sub-trajectory corresponding to the current skill, and the decision and execution of the next skill are carried out until the fracture reset task is completed, and the complete reset trajectory generated by the connection of each skill sub-trajectory is output; wherein, the final state of the previous skill sub-trajectory is used as the initial state of the dynamic motion primitive model corresponding to the next skill, so that the complete reset trajectory generated by the connection satisfies motion continuity.

2. The fracture reduction trajectory planning method for a parallel orthopedic robot according to claim 1, wherein, The multiple skills include at least: Translational obstacle avoidance skills used to move the distal end of a fracture to a middle position that maintains a safe distance from the proximal end of the fracture. Rotational alignment skill for rotating the distal end of the fracture to align with the proximal end of the fracture in the intermediate position; Translational fitting technique for translating the distal end of the fracture along the alignment axis to fit the proximal fracture surface.

3. The method of claim 1, wherein, The repositioning environment status information includes at least one of the following: relative position and distance between fracture ends, posture alignment error, fracture surface fit error, contact force information, executed skill sequence, number of skill switching, task completion indicator, obstacle information, and sampled motion speed.

4. The fracture reduction trajectory planning method for a parallel orthopedic robot according to claim 3, wherein, After generating the sub-trajectory corresponding to the current skill, the method further includes real-time collision detection of the sub-trajectory corresponding to the current skill, including: During the generation of the sub-trajectory corresponding to the current skill, the feasible motion speed of the parallel orthopedic robot end effector is sampled in real time; Based on the obstacle information and sampled motion speed in the reset environment state information, the motion trajectory in the future time period is simulated and predicted. Based on the trajectory evaluation function, the motion trajectory within the future time period is evaluated. If the evaluation result indicates a collision risk, the sub-trajectory corresponding to the current skill is adjusted.

5. The method of claim 4, wherein, The adjustment of the sub-trajectory corresponding to the current skill includes: Using the trajectory evaluation function, a safety assessment is performed on each candidate trajectory, and the alignment degree between each candidate trajectory and the original target of the sub-trajectory corresponding to the current skill is calculated. Based on the results generated by the safety assessment and the alignment, the optimal trajectory is selected from the candidate trajectories, and the corresponding part of the current skill sub-trajectory identified as having a collision risk is replaced by the optimal trajectory.

6. The method of fracture reduction trajectory planning for a parallel orthopedic robot according to claim 1, wherein, The step of adaptively deciding on the skill to be executed from the plurality of skills based on the reset environment state information includes: The reset environment state information is input into a pre-trained high-level policy network in the Option-Critic framework; The high-level policy network outputs the probability that each of the multiple skills is selected as the current skill to be executed; Based on the probability, the skill to be executed is determined.

7. The fracture reduction trajectory planning method for a parallel orthopedic robot according to claim 1, characterized in that, After the complete reset trajectory generated by connecting the output skill sub-trajectories is completed, it also includes: A multi-objective optimization model is constructed with the goal of minimizing the sum of squared velocities, sum of squared accelerations, and sum of squared jerkes of each kinematic branch of the parallel orthopedic robot on the complete reset trajectory. A multi-objective evolutionary algorithm is used to solve the multi-objective optimization model under preset time, velocity, acceleration and displacement constraints to obtain the optimized smooth reset trajectory.

8. A fracture reduction trajectory planning device for a parallel orthopedic robot, characterized in that, include: The splitting module is used to split the fracture reduction task into multiple skills, each skill corresponding to a dynamic motion primitive model, and the dynamic motion primitive model is used to generate the sub-trajectory of the corresponding skill. The decision module is used to obtain the reduction environment status information of the fracture reduction task, and adaptively decide the skill to be executed from the multiple skills based on the reduction environment status information. The generation module is used to call the dynamic motion primitive model associated with the selected skill to generate the sub-trajectory corresponding to the current skill; The output module is used to update the reset environment state information according to the final state information of the sub-trajectory corresponding to the current skill after the termination condition of the current skill is met, and to make decisions and execute the next skill until the fracture reset task is completed, and output the complete reset trajectory generated by connecting the sub-trajectories of each skill; wherein, the final state of the previous skill sub-trajectory is used as the initial state of the dynamic motion primitive model corresponding to the next skill, so that the complete reset trajectory generated by the connection satisfies motion continuity.

9. 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 fracture reduction trajectory planning method for the parallel orthopedic robot as described in any one of claims 1 to 7.

10. A non-transitory readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the fracture reduction trajectory planning method for a parallel orthopedic robot as described in any one of claims 1 to 7.

Images