Surgical robot navigation obstacle avoidance control method

Abstract

The invention relates to a control method for navigation obstacle avoidance of a surgical robot. Performing dynamic construction by adopting a continuous state space of intra-operative process modeling; tissue, tool and robot pose data are collected in real time, and a continuous change tensor field is calculated; establishing a process field atlas, and inputting a state space into a subsequent navigation control model to realize prospective modeling of an environment trend; carrying out adaptive evolution by adopting an obstacle avoidance strategy of a risk threshold drift model; environment change is monitored in real time, and a risk trend is evaluated through a time window sliding mechanism; path fine tuning constraint of navigation task integrity protection is adopted; collaborative navigation optimization based on man-machine strategy intention fusion is adopted; recording micro-control behaviors, including small-range fine tuning and path locking, of the doctor on the operation terminal; establishing an intention preference model by using the operation behavior, and taking the intention preference model as a dynamic navigation strategy adjustment factor; when the obstacle avoidance strategy of the robot conflicts with the intention of the doctor, triggering the collaborative optimizer to realize soft fusion of the navigation strategy; and meanwhile, safety and control consistency are guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of surgical robots, and particularly relates to a control method for navigation and obstacle avoidance of surgical robots. Background Art

[0002] At present, although the control methods for navigation and obstacle avoidance of surgical robots have achieved certain application results in many medical fields, there are still many key deficiencies and technical bottlenecks in actual complex surgical scenarios, which limit the further improvement of their performance and wide clinical promotion. First of all, most existing methods still mainly rely on static environment modeling and rule-driven path planning, lacking the effective perception and modeling ability for dynamic, non-linear, and unpredictable changes in the surgical scenario. During the actual surgical process, soft tissue deformation, temporary intervention of doctor's instruments, physiological movements (such as pulsation caused by breathing and heartbeat), etc. will all cause continuous changes in the surgical environment. However, traditional path planning algorithms often construct a navigation map based on preoperative images or single-frame intraoperative data, and correct the path based on the presence or absence of obstacles. This "post-response" obstacle avoidance strategy lacks the ability to predict the evolution trend of the environment, and it is extremely easy to have inconsistencies between the robot's operation path and the actual passable path, resulting in problems such as frequent path reconstruction, operation jamming, or mistakes. Secondly, existing methods generally lack the modeling of complex coupling relationships between soft tissues. Spatial changes in the surgical scenario often do not occur in isolation. For example, after a scalpel touches the tissue in a certain area, it may affect multiple surrounding areas through tissue tension chains or liquid pressure transmission. However, traditional methods usually only monitor the collision distance between the robot and the obstacle, and cannot identify the dynamic causal relationships between tissues. Therefore, when performing obstacle avoidance control, it is impossible to reasonably evaluate the potential chain effects of path fine-tuning on the overall environment, and it is easy to have the problem of "successful local obstacle avoidance but increased overall interference", which in turn affects the stability of the operation and the coherence of the completion of the surgical task. Thirdly, the input data structure of the current mainstream navigation control system is single, usually only based on low-dimensional quantization indexes such as distance fields, cost maps, or current risk values, and it is impossible to integrate multi-source information to uniformly express the complex relationships between the risk propagation trend, the stability of the task objective, and the execution accuracy. As a result, it is difficult for the path planning system to perform global optimization under multi-objective constraints. Especially when the task objective area has high spatial constraints (such as clearance beside arteries, suture beside nerves), the system often causes problems such as path concession, operation redundancy, and time extension because it is unable to make a balance between "safe obstacle avoidance" and "path accuracy". Fourthly, the traditional path fine-tuning mechanism is too dependent on obstacle detection and lacks task semantic constraints. Existing methods usually equate obstacle avoidance behavior with avoiding from the obstacle area, lacking the consideration of "whether the path still closely adheres to the surgical target area", and lacking a feedback evaluation model for "path-task" consistency. As a result, once an obstacle avoidance adjustment occurs, the path is prone to gradually drift to non-critical areas. Especially in minimally invasive surgery, such a deviation of even 1-2 millimeters may affect the postoperative effect, the judgment of the resection margin, or even cause tissue injury.

[0003] Finally, the current path recovery mechanism mainly relies on manual callback by doctors or preset route correction, lacking the intelligent judgment of whether the path has "semantic derailment" due to obstacle avoidance, and there is no mechanism to dynamically calculate whether the path should return to the original task area. As a result, after multiple obstacle avoidances, the overall path deviates from the target, but the system still determines it as a valid path, and the robot cannot actively restore the original trajectory, causing clinical risks such as cutting deviation or incomplete suturing. Summary of the Invention

[0004] The purpose of the present invention is to provide a control method for surgical robot navigation obstacle avoidance, so as to solve some of the drawbacks and deficiencies pointed out in the background technology.

[0005] The technical solution adopted by the present invention to solve its above technical problems is as follows: A control method for surgical robot navigation obstacle avoidance, including: S1. Continuously constructing a dynamic state space by intraoperative process modeling:

[0006] S1.1. Real-time collect tissue, tool, and robot pose data, and calculate a continuously changing tensor field;

[0007] S1.2. Establish a process field atlas, convert the change trend in the surgical environment into a dynamic state space available for navigation; input the state space into the subsequent navigation control model for forward-looking modeling of the environmental trend;

[0008] S2. Adaptive evolution of the obstacle avoidance strategy using a risk threshold drift model:

[0009] S2.1. In path planning, set an initial risk value for each potential conflict area; real-time monitor environmental changes, and evaluate the risk trend through a time window sliding mechanism;

[0010] S2.2. Introduce a dynamic obstacle avoidance weight function to control the trigger threshold of the obstacle avoidance strategy, and adaptively avoid or restore the original path according to the risk trend;

[0011] S3. Path fine-tuning constraint using navigation task integrity protection:

[0012] S3.1. Set the spatial probability distribution of the path target area including the task hot spot area;

[0013] S3.2. Introduce this distribution as a constraint factor in the generation of the obstacle avoidance path to limit the deviation range of path search;

[0014] S4. Collaborative navigation optimization using human-machine strategy intention fusion:

[0015] S4.1. Record the micro-manipulation behaviors of the doctor on the operation terminal, including small-range fine-tuning and path locking; establish an intention preference model using the operation behaviors and use it as a dynamic navigation strategy adjustment factor;

[0016] S4.2. When there is a conflict between the robot's obstacle avoidance strategy and the doctor's intention, trigger the collaborative optimizer to softly fuse the navigation strategy: make appropriate concessions while ensuring safety and manipulation consistency.

[0017] Furthermore, the method for dynamically constructing the continuous state space of intraoperative process modeling includes:

[0018] Adopt the dynamic tensor field modeling method to perceive the state of the surgical area from two dimensions of time and space, and organize the multi-modal perception data into a four-dimensional tensor field; the tensor field is used to express the trend of each spatial position in the operation over time and to perceive in advance the areas that will change in the future; define the following tensor-oriented function:

[0019]

[0020] Where:

[0021] is the dynamic trend function, representing the change direction and rate of the spatial point (x, y, z) at time t; is the perception tensor field, integrating visual images, tissue structures, and tool pose information; is the spatial gradient, reflecting the direction of tissue deformation or structural change in the local area; is the time derivative, used to capture the change rate of the above terms over time; α 1 、α 2 are adjustment parameters, respectively controlling the influence weights of the tensor body and the spatial gradient on the overall trend judgment.

[0022] Furthermore, the method for dynamically constructing the continuous state space of intraoperative process modeling includes:

[0023] Based on the tensor trend, construct a surgical process field atlas, form a dynamic causal network between regions, decompose the surgical area into multiple dynamic units with causal relationships, and depict the interaction patterns between them; the trend tensor function Define the following coupling degree between regions:

[0024]

[0025] Where:

[0026] φ(Ω i ,Ω j ) represents the dynamic causal coupling degree between region Ω i and region Ω j within the time interval [t 0 ,t 1 ; is the time derivative of the regional tensor trend function, representing the degree of drastic change in the regional state; · is the dot product operation, used to evaluate whether the trend directions of two regions are consistent; |·| is to take the absolute value to eliminate the direction difference; β is the coupling sensitivity coefficient, used to adjust the sensitivity of different tissues or surgical types to the chain reaction.

[0027] Furthermore, the method for continuously constructing the state space of the intraoperative process modeling includes:

[0028] Compress the atlas, extract the core trend information, and generate the evolving state space input to the navigation controller, that is, the function Ψ(t) that comprehensively expresses the risk change trend, the stability of the task area, and the path feasibility factors; define the following aggregation expression:

[0029]

[0030] Where:

[0031] Ψ(t) is the evolving state space function and is the dynamic input of the navigation control module; n represents the number of dynamic regions currently participating in the modeling; γ i (t) is the importance weight of the i-th region, determined according to whether it is a critical operation path; is the time change rate representing the coupling relationship between this region and all other regions, that is, the risk propagation speed; μ i (t) represents the regional structural stability, that is, the rigid requirement for the execution of the task objective within this region; is a non-linear fusion operation, used to merge the interaction relationship between the trend change rate and the structural stability; limΔt→0 is the limit operation, making the state space time continuous and reflecting the potential risk pattern at each moment within the surgical area.

[0032] Furthermore, the path fine-tuning constraint method for protecting the integrity of the navigation task includes:

[0033] According to the preoperative image data and the intraoperative real-time perception information, the entire surgical operation area is divided into spatial units of navigation semantics. Each unit is dynamically assigned a task association probability according to its association degree with the surgical target, anatomical importance, and operability, forming a task hot spot area atlas in the sense of space, used to express the distribution trend of the priority crossing areas in the robot navigation path; and to describe the trend of this atlas evolving over time during the surgery, construct the following dynamic hot spot activity function:

[0034]

[0035] Where, represents the comprehensive activity of the task hot spot area at time t; P r(x, y, z, t) represents the task-related probability density of the point (x, y, z) in the navigation space at time t; is the hotspot change rate, which is used to measure the degree of fluctuation of the regional task value over time; is the spatial gradient of the hotspot, indicating the sharpness or diffusion trend of the boundary of the hotspot area; α 1 、α 2 are the time trend weight and the spatial form weight respectively; Ω represents the integration range of the entire navigation area, covering all spatial units passed by the robot.

[0036] Furthermore, the path fine-tuning constraint method for protecting the integrity of the navigation task includes:

[0037] During the path planning process, use this probability density map to restrict and control the obstacle avoidance search space, construct an offset range control function with task semantic constraints, so that the path search only allows fine-tuning near the high task weight area, and construct the following constraint formula:

[0038] Δ lim (x, y, z) = ρ · (1 - P r (x, y, z, t)) η

[0039] where, Δ lim (x, y, z) represents the maximum allowable path offset at the position (x, y, z); ρ is the overall path offset scale coefficient, which determines the basic size of the offset limit; P r (x, y, z, t) is the current task priority probability value of the position point; η is the non-linear regulation index, which is used to define the boundary transition rate of the task hotspot area.

[0040] Furthermore, the path fine-tuning constraint method for protecting the integrity of the navigation task includes:

[0041] Introduce a path recovery potential energy function to dynamically evaluate the task consistency of the current path; when the potential energy exceeds the set threshold, trigger the path callback mechanism to restore to a better task-related path trajectory; the recovery potential energy function is defined as follows:

[0042]

[0043] where, represents the recovery potential energy value of the overall path at time t, which is used to measure the severity of the semantic deviation of the path; ξ i (t) represents the spatial position of the i-th path point in the current path over time; is the square term of the movement rate of this path point, which is used to reflect the fluctuation of the path; γ i is the importance weight of the path point; λ iis the offset sensitivity coefficient of the path point, used to strengthen the traction effect of the task hot spot on the path; P r (ξ i (t)) is the current task probability value of the path point, indicating whether the path point is still in the semantic hot spot area.

[0044] The control method for navigation and obstacle avoidance of the surgical robot of the present invention has the following beneficial effects:

[0045] By constructing an intraoperative continuous state space and a dynamic tensor field model, the surgical robot is enabled to have the ability to sense the trends of soft tissue, instrument intervention, and changes in the surrounding environment, realizing the transformation of obstacle avoidance control from "static reaction" to "forward prediction", effectively improving the navigation sensitivity and decision-making initiative of the robot in complex dynamic surgical scenarios; introducing a process field atlas and a regional causal coupling mechanism, enabling the robot to identify the dynamic chain influence relationships between tissues, avoiding global interference caused by local path adjustments, thereby enhancing the system's understanding ability and coordinated control ability of the overall behavior of the surgical field;

[0046] Integrating the evolutionary state space function, comprehensively expressing the risk propagation trend, task objective stability, and path feasibility, enabling the control module to have unified, dynamic, and continuous navigation decision-making inputs, and achieving the optimal balance between path avoidance and task maintenance; by constructing a task hot spot area atlas and an offset limit function, forming a path fine-tuning mechanism with navigation semantic constraints, enabling the path to always fit the surgical target area during obstacle avoidance, improving the execution accuracy of the navigation path, and avoiding path drift caused by excessive avoidance. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is the flowchart of the control method for navigation and obstacle avoidance of the surgical robot of the present invention.

[0048] Figure 2 is the flowchart of the method for dynamically constructing the continuous state space of intraoperative process modeling of the present invention.

[0049] Figure 3 is the flowchart of the path fine-tuning constraint method for protecting the integrity of the navigation task of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0050] The following provides a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings.

[0051] Combined with the attached Figure 1, S1 adopts the continuous state - space dynamic construction of intraoperative process modeling. S1.1 Collect tissue, tool, and robot pose data in real - time and calculate the continuously changing tensor field. Specifically, during the surgical process, by integrating a multi - modal perception system, including endoscopic images, intraoperative ultrasound, instrument tracking devices, robot body sensors, etc., high - frequency sampling is performed on data such as soft - tissue deformation, doctor's instrument movement trajectories, three - dimensional positions, postures, and velocities of each joint and end - effector of the robot body in the surgical area. These perception data are mapped into a unified four - dimensional data structure to construct a tensor field model that changes continuously over time. This tensor field not only describes the geometric states of various key points in the current space but also includes their change directions, change rates, and influence propagation trends in the time dimension, reflecting the coupled evolution characteristics of the surgical environment in space and time. S1.2 Establish a process field atlas, transform the change trends in the surgical environment into a dynamic state space available for navigation, and input the state space into the subsequent navigation control model for forward - looking modeling of environmental trends. Specifically, perform regional decomposition and semantic aggregation processing on the aforementioned tensor field, divide the surgical space into several dynamic sub - regions, establish process causal association relationships between these sub - regions to form a set of process field atlas structures. Each region node records information such as its dynamic trend, interference probability, and tissue response behavior. At the same time, connect the regions into an atlas network through the degree of trend coupling. This atlas can effectively express how the change in one region affects other regions, reflecting the behavior propagation chain in the entire surgical environment. Further, compress this atlas into a state - space model with a unified form, extract the most critical state parameters for navigation control, form a navigation input set of dynamic risk perception and predictive trend expression, and input this set into the subsequent path - planning and obstacle - avoidance control model, enabling the surgical robot to not only rely on the current state judgment when performing obstacle - avoidance behavior but also predict in advance trend events such as upcoming deformation, tool approach, and tissue expansion in certain regions, thus realizing the forward - looking and initiative of obstacle - avoidance control.

[0052] S2.1 In path planning, set an initial risk value for each potential conflict area; monitor environmental changes in real time, and evaluate the risk trend through a time window sliding mechanism. Specifically, when the system executes path planning, it identifies spatial points or areas with dynamic uncertainties in the surgical area as potential conflict sources, such as the periodic expansion and contraction of soft tissues due to physiological movements, the temporary entry of the doctor's instruments into the working space, etc., and sets an initial risk value for these areas. This risk value is jointly estimated based on historical surgical data, preoperative modeling results, and current perception signals, initially reflecting the necessity of avoidance and the possibility of change in the area. Subsequently, the system continuously monitors the real-time status of these areas during the operation, and dynamically evaluates the evolution trend of their risk values through the time sliding window mechanism. The continuous status data of the area is collected within the sliding window, including multi-dimensional indicators such as deformation amount, proximity, vibration frequency, tool trajectory overlap rate, etc., and the change direction, rate, and fluctuation intensity of the risk trend are calculated to determine whether the risk of the area is increasing, decreasing, or stable. S2.2 Introduce a dynamic obstacle avoidance weight function to control the trigger threshold of the obstacle avoidance strategy, and adaptively avoid or resume the original path according to the risk trend. Specifically, based on the above risk trend evaluation, the system couples the risk trend change with the navigation decision-making process by constructing a dynamic obstacle avoidance weight function. The weight function is comprehensively calculated according to the risk change rate of each area, the coverage of the surgical path, and the rigid requirements for task completion, dynamically adjusting the influence of the area in the path optimization model, that is, increasing its obstacle avoidance weight during the continuous increase stage of the risk to strengthen the path avoidance behavior; while when the risk value of the area shows a downward trend and tends to be stable, its obstacle avoidance priority is appropriately reduced, allowing the robot to re-approach the original planned path or return to the preoperative task trajectory under the condition of meeting the safety redundancy conditions, so as to achieve the adaptive adjustment of path decision-making and the flexibility of the strategy, ensuring that the surgical robot can actively avoid potential risks in a dynamic environment and will not deviate from the task target path due to excessive obstacle avoidance, realizing the intelligent balance of obstacle avoidance control.

[0053] Ensure that while the surgical robot performs obstacle avoidance behavior, it always maintains the spatial consistency between the path and the surgical target, and avoids the decline of surgical accuracy or the failure of the task caused by the path deviation triggered by obstacle avoidance. S3.1 Set the spatial probability distribution of the path target area including the task hot spot area. The method is to construct a task hot spot map for the entire surgical area according to the preoperative surgical planning data, the intraoperative doctor's operation intention, and the spatial distribution of the target anatomical structure. This map expresses the possibility that each position point is set as the priority passing area of the navigation path in the form of spatial probability. The higher the probability value, the more critical the area is during the operation and the more it should be covered by the path. For example, the intraoperative resection line, suture trajectory, key operation window, etc. will be assigned high probability labels. This map can be dynamically updated according to the real-time image feedback and the doctor's fine-tuning behavior during the operation, reflecting the process of surgical rhythm and target change. S3.2 Introduce this distribution as a constraint factor in the generation of the obstacle avoidance path to limit the deviation range of the path search. Specifically, use the above spatial probability distribution as a soft constraint input in the navigation path planning process. When searching for the path, guide the algorithm to preferentially select candidate points close to the high-task probability area. At the same time, set the maximum allowable deviation range threshold to constrain the adjustment amplitude of the path during obstacle avoidance, so that the path can only be fine-tuned near the hot spot area. If a path candidate point deviates too far from the high-probability area, its cost function will be significantly increased, reducing its probability of being selected, thereby realizing the adhesion protection of the task target during the path optimization process.

[0054] In solving the deviation or conflict problem between the robot's autonomous navigation obstacle avoidance strategy and the doctor's manual control intention during the operation, by constructing a dynamically adjustable human-machine collaborative control mechanism, the robot's decision-making process can not only maintain the intelligent perception and obstacle avoidance ability of the risk environment, but also adapt to the local preferences, individual experiences and real-time judgments reflected by the doctor during the operation, ensuring the flexibility and controllability of the navigation path. S4.1 Record the micro-operation behaviors of the doctor on the operation terminal, including small-range fine-tuning and path locking; establish an intention preference model using the operation behaviors and use it as a dynamic navigation strategy adjustment factor. During this process, the system will record the micro-operation instructions issued by the doctor through the main control device (such as the teleoperation handle, image interaction interface, voice input, etc.) in real time, including the manual fine-tuning of the path nodes, the priority suggestions for the navigation direction, the avoidance and locking instructions for certain areas, etc. Through time series analysis and spatial intention trajectory reconstruction, extract the long-term behavior characteristics of the doctor's surgical rhythm, operation accuracy, and path passing preferences, and use these characteristics to construct a dynamic intention preference model. This model is continuously updated with the doctor's behavior and participates in the robot's decision-making process as an important weight factor in the path optimization module, so that the robot gives priority to considering the doctor's habits or intentions when generating the path and avoiding obstacles. Figure 1Consistent action plan. S4.2 When there is a conflict between the robot's obstacle avoidance strategy and the doctor's intention, the collaborative optimizer is triggered to softly fuse the navigation strategy: make appropriate concessions while ensuring safety and control consistency. Specifically, after the robot obtains an obstacle avoidance path suggestion based on the perception data and prediction model, the system will perform real-time matching analysis of this suggestion with the current doctor's intention preference model. If it is determined that there is an obvious directional conflict, such as the robot intending to avoid a certain area while the doctor hopes to approach, or the doctor has locked the path but the robot intends to deviate, the collaborative optimizer will be triggered to start the soft fusion strategy. Within the allowable range of safety redundancy conditions, the obstacle avoidance path is moderately adjusted, making appropriate concessions in the direction of the doctor's intention. At the same time, path mutations are eliminated through means such as motion smoothing and variable speed interpolation to ensure the continuity and controllability of the robot's operation, ultimately achieving the coordination and unity of the human-machine strategy, and enhancing the humanization of the system response and the doctor's trust in the operation.

[0055] Example 1:

[0056] Combined with the attached Figure 2 , in this embodiment, a complex hepatectomy was performed in a hepatobiliary minimally invasive center. Patient Mr. Li had a tumor deep in the liver segment and needed to be assisted by the "intelligent surgical robot system" to perform precise resection. The surgical team adopted the navigation and obstacle avoidance control method of the present invention to improve the flexibility of intraoperative tissue deformation prediction and obstacle avoidance paths, ensuring that while avoiding important vascular structures, the surgical path was steadily advanced.

[0057] At the beginning stage of the operation, the robot first executes: continuously constructing the dynamic state space of intraoperative process modeling. The system, through the integrated intraoperative vision system (such as a 3D structured light camera), ultrasonic imaging module, and instrument motion tracking module, real-time collects the surface deformation of the patient's liver, the change of tumor boundary tension, the beating frequency of surrounding blood vessels, and the attitude behavior of the doctor's operating instrument. All these data are aggregated into multi-modal data points at a frequency of 100 Hz per second and are converted into a perception tensor field after synchronous fusion. This tensor field covers each key point in the three-dimensional space of the liver surgery area, and at the same time records information such as tissue density, surface movement trend, and internal stretching direction that change over time.

[0058] Set the current observation point at the spatial coordinates (x = 12 mm, y = 28 mm, z = 14 mm), and the time is the 180th second during the operation. Obtain the tensor value at this point position as The tensor change rate observed in the adjacent frame time difference of 0.01 s is At the same time, its spatial gradient (generated by stretching / contraction) is It indicates that the tissue at this point position is slowly stretching in the direction of the tumor edge.

[0059] The system calls the core tensor-guided function of the present invention at this time for trend calculation, and the function form is as follows:

[0060]

[0061] Among them, the adjustment parameter α 1 and α 2 The values of are generally set to α 1 = 0.7 - 1.2 in clinical experience, which is used to emphasize the dominance of the original tensor information in trend judgment, and α 2 = 0.3 - 0.8, which is used to adjust the participation degree of tissue space deformation in trend judgment. Since the liver tissue is in a soft and easily deformable area this time, the system automatically adjusts the parameters to α 1 = 1.0, α 2 = 0.5.

[0062] Substitute the current perception data for calculation as follows:

[0063]

[0064] Output result Indicates that this point is undergoing dynamic changes during the operation with a medium intensity trend, and the direction is biased towards contraction but accompanied by a slight displacement. This value is accumulated over time to predict the tissue deformation direction and amplitude 5 seconds later, and is marked by the system as a risk area of "potentially evolving into an obstacle".

[0065] This trend is input into the navigation control model in real time. The controller judges that the planned path of the robot will intersect with this area in 3 seconds. If not adjusted, it will cause interference with the tissue deformation area when the instrument is advanced. Therefore, the control module combines the path reconstruction mechanism, based on the dynamic tensor field result, performs a 2-mm micro-offset on the current path and decelerates by 20%, and at the same time adds deformation correction filtering, finally achieving conflict-free advancement while ensuring that the path still fits the hepatic segment anatomical boundary.

[0066] When the operation enters the middle and late stages, the surgical robot is approaching the tumor margin for fine resection. At this time, it is necessary to predict the local tissue dynamic response caused by instrument intervention, liver micro-contraction, and doctor's operation adjustment, and avoid excessive path deviation or surgical interference when the robot avoids obstacles. To further improve the understanding of environmental changes and navigation stability, the system executes the second part of the intraoperative process modeling, that is, constructing a surgical process field atlas based on tensor trends, forming a dynamic causal network between regions, decomposing the surgical area into multiple dynamic units with causal relationships, and depicting the interaction patterns between them. Specifically, the robot navigation system divides the entire surgical space into 15 local regional units, each with a size of 10 cubic millimeters, and names them Ω 1 、Ω 2 ……Ω 15, for tracking the dynamic trends of tissues. Taking Ω located at the upper edge of the tumor 5 and the adjacent Ω 6 as an example, the system needs to determine whether the tissue deformation caused by the instrument compression of Ω 5 will be transmitted and affect Ω 6 , which in turn triggers the need for temporary obstacle avoidance correction of local segments of the navigation path. Therefore, it is necessary to evaluate the dynamic causal coupling degree between these two regions. The system first calls the previously established trend tensor function to perform trend time derivative analysis on the representative tensor points in Ω 5 and Ω 6 respectively. Continuous tensor evolution data for 4 seconds are collected within the time window t 0 = 180s, t 1 = 184s. After filtering and interpolation, shows that both regions are in a high dynamic response state during the operation. The system introduces the following coupling degree calculation formula for trend causal evaluation:

[0067]

[0068] In the current application, β is the coupling sensitivity coefficient, with a value range of 0.5 to 2.0. The specific value is adjusted according to the tissue type. For soft tissues (such as liver tissue), it is set to 1.3, considering that their deformations are easily implicated with each other. Substituting the data into the calculation, the dot product term is 0.094·0.081 = 0.007614, and taking the absolute value is still 0.007614. The integration interval is 4 seconds, and we get:

[0069]

[0070] The coupling degree value is 0.0396. The system determines that it is higher than the currently set coupling trigger threshold of 0.03, and believes that there is an obvious dynamic chain relationship between Ω 5 and Ω 6 . If the robot continues to apply force in Ω 5 , it will cause uncontrollable deformation of Ω 6 , affecting the safety distance between the end effector of the robot and the hepatic segment vein. The system immediately marks the "pre-strain sensitive area" label on Ω 6 in the navigation path map, and at the same time calls the micro-path avoidance mechanism to fine-tune the seventh segment of the planned path by 0.8 mm, shifting the robot's avoidance direction from the edge of Ω 6 to Ω 7 , and reducing the speed by 15% to maintain the deformation tracking accuracy.

[0071] The operation has entered a critical final stage. The robot needs to complete the cutting of the terminal lesion along the edge of the hepatic vein. The space in this area is narrow, the tissue is highly active, and the tolerance for cutting error is extremely low. At this time, the surgical navigation system officially activates the "evolving state space" as the input of the navigation control module, entering the final stage of the continuous state space dynamic construction method for intraoperative process modeling, compressing the previously constructed process field atlas, extracting the core trend information, and generating the high-density state expression required by the navigation controller. To ensure that the obstacle avoidance path can not only avoid dynamic interference but also not deviate from the key task area, the system extracts the key variables of the 12 dynamic region units Ω 1 to Ω 12 marked in the atlas one by one, constructs the evolving state space function Ψ(t), which is defined as follows:

[0072]

[0073] In this scenario, n = 12, indicating that 12 dynamic regions are currently included in the real-time state assessment; γ i (t) represents the navigation weight of each region. The system assigns values according to whether the region is on the main path. Among them, the weight of the region on the main path of the path is set to 0.9 - 1.0, the secondary region is set to 0.4 - 0.7, and the non-critical region is set to 0.1 - 0.3; represents the time change rate of the coupling relationship between this region and other regions, whether the regional risk spreads rapidly within a unit time to identify dynamic risk regions; μ i (t) represents the structural stability coefficient, and its value range is 0.2 - 1.0. The higher the value, the more rigid the task target is in this region and the path cannot be deviated; represents the non-linear fusion operation. In this scenario, the power fusion rule is used to enhance the weight proportion of stability in the high-dynamic state.

[0074] Between 180 and 184 seconds, for the Ω 5 region, it is measured that: γ 5 (t) = 0.95, which is the main region of the path; represents that the risk is spreading at a medium intensity; μ 5 (t) = 0.88, which is a high-rigidity task execution area; After substituting into the non-linear fusion operation, we get: Finally, the contribution of this item to the overall state space is 0.95·0.217 ≈ 0.206.

[0075] After the system performs similar calculations on all 12 regions, sums them up, and maintains time continuity in the limit of Δt→0, Ψ(t)≈1.632 at this moment is obtained. This value exceeds the current path adjustment threshold (set to 1.4). The system determines that the navigation environment is in a high-risk dynamic state. At the same time, the rigid constraint of the main path area is strong and cannot be freely avoided. Therefore, the "high-adhesion obstacle avoidance mode" is activated to avoid obstacles along the current task direction in a micro-offset manner. At the same time, the rotational speed of the robot end is locked not to exceed 2 mm / s, and "high-stability regression linear predictive control" is added to the path within the next 3 seconds to complete the necessary callback.

[0076] Example 2:

[0077] Combined with the attached Figure 3 , in this embodiment, Mr. Li's hepatectomy has entered the stage of deep tumor dissection. At this stage, the tolerance for path deviation is extremely low, and a slight deviation will cause the robot to enter a high-risk area near the main hepatic vein by mistake. Therefore, the system fully activates the "path fine-tuning constraint method for navigation task integrity protection" to ensure the high consistency between the surgical path and the task objective. First, the system divides the entire liver operation area into semantic space units based on the preoperative CT three-dimensional reconstruction image and the intraoperative real-time ultrasound fusion image, forming 150 spatial units. Each unit sets a dynamically assigned task association probability P r (x, y, z, t), with a value range of 0 to 1. The higher the value, the more the point needs to be covered by the navigation path. At the current time point t = 195 seconds, the system detects that the 73rd area unit (Ω 73 ) is located at the tumor edge and close to the operation priority channel marked by the doctor, and assigns P r = 0.91, indicating a strong hot spot area. The robot should maintain path stability at this point and avoid deviation.

[0078] With the local traction of the surgical instrument and the change of tissue fluid tension, the system continuously observes the fluctuation of the hot spot probability value within 4 seconds, and the time derivative is estimated as indicating an upward trend in the activity of this area. At the same time, according to the detection of the spatial gradient change of the hot spot map in the adjacent area, is calculated, indicating that the regional boundary is rapidly narrowing and the concentration of hot spots is increasing. To evaluate the current activity of the entire hot spot map, the system calls the dynamic hot spot activity function of the present invention for calculation:

[0079]

[0080] where α 1 , α 2 are adjustment weight coefficients. In this scenario, in order to emphasize the sharp change of hot spot fluctuation in time, they are correspondingly set as α1 = 1.2, α 2 = 0.8. This weight range can be adjusted according to the task sensitivity during actual deployment. The generally recommended range is α 1 ∈ [0.8, 1.5], α 2 ∈ [0.5, 1.2]. Let the integration region Ω cover all spatial units related to the current navigation. After the integration result is calculated using Gaussian partitioning, the overall hotspot activity is obtained The task hotspot threshold set by the system is 4.8. Once this value is exceeded, it means that the path is currently in a high adhesion constraint state and cannot deviate

[0081] Based on this, the system enables the path fine-tuning protection mechanism and issues a restriction instruction to the obstacle avoidance module during path generation. The path planning algorithm only allows searching for the optimal path within a region centered on Ω 73 with a radius not exceeding 2.5 mm. At the same time, the hotspot weight factor in the cost function is increased to 1.6, so that the path near the hotspot area has a significant scoring advantage, forcing the path to be fine-tuned along the hotspot central axis without sudden direction changes. Under this mechanism, the robot successfully avoids the expanding Ω 75 vascular dilation area and at the same time maintains an offset within 0.4 mm on the trajectory of the Ω 73 hotspot area. By verifying the path continuity point by point and ensuring that the task probability of all path points is greater than 0.85, the operation accuracy and target consistency are ensured to meet the surgical requirements

[0082] For patient Mr. Li, the hepatectomy has entered the final stage of lesion removal. The surgical robot is about to complete the cutting operation at the root of the deep tumor through an extremely narrow anatomical gap. This path requires avoiding the adjacent hepatic vein while advancing precisely along the cutting boundary set by the doctor before the operation. At this time, the system officially enables the key part of the path fine-tuning constraint method for navigation task integrity protection in the present invention: constructing an offset range control function with task semantic constraints to achieve precise limitation of the offset range of the obstacle avoidance path in the dynamic changes of complex tissues. The system uses the task hotspot probability density map updated during the operation, which integrates the preoperative CT reconstruction results, intraoperative ultrasound images, and navigation path feedback data, and updates P r (x, y, z, t) The value range is 0, 1. The higher the value, the more precise the position where the task path must pass. The current operating end of the robot is performing a key operation near the spatial coordinate point x = 18 mm, y = 36 mm, z = 12 mm. The system calculates the task probability value of this point as P r = 0.94, indicating a very high weight task area. To ensure that the path search does not deviate from this high-priority area, the system calls the path offset limit function for constraint:

[0083] Δ lim (x, y, z) = ρ · (1 - Pr (x, y, z, t)) η

[0084] In this function, Δ lim (x, y, z) represents the maximum allowable path offset range at this position, in millimeters; ρ is the overall path offset scale coefficient, which determines the basic allowable range of offset in non-hotspot areas. Its setting range is generally ρ ∈ [2.0, 6.0]. In this operation, to avoid excessive offset affecting accuracy, ρ = 3.0 is taken; η is the non-linear regulation index, which is used to control the steepness of the transition at the hotspot boundary. The larger the index value, the "harder" the boundary of the hotspot area, and the more difficult it is for the robot to approach the non-task area. The recommended range is η ∈ [1.5, 4.0]. In this example, considering the highly compact edge of the lesion, η = 3.5 is set. Substituting the values into the formula, we get:

[0085] Δ lim = 3.0·(1 - 0.94) 3.5 = 3.0·(0.06) 3.5 ≈ 3.0·2.58×10 -5 ≈ 7.74×10 -5 mm

[0086] The calculation results show that the maximum allowable range of path offset at the current position point is approximately 0.0000774 millimeters, and almost no deviation is allowed. The system then sets strong restrictions on the path planning algorithm, enabling it to only make minor-scale path point adjustments near the original trajectory. At the same time, the "rigid path lock" mechanism is enabled for this path segment, and route changes brought about by the obstacle avoidance strategy are not accepted, unless there are high-risk sudden interferences in the surrounding area. Conversely, for the Ω 102 auxiliary area 8 mm outside the operation area, its task probability value is only P r = 0.25. The system substitutes the same formula to calculate its allowable offset range:

[0087] Δ lim = 3.0·(1 - 0.25) 3.5 = 3.0·(0.75) 3.5 ≈ 3.0·0.334 ≈ 1.002 mm

[0088] It can be seen that in non-critical task areas, the system allows the path to deviate by up to 1 mm at most, improving the flexibility of navigation obstacle avoidance. Through this path control strategy of dynamic adjustment of regional perception and constraint intensity, the robot can safely avoid obstacles while closely following the task target, enabling the navigation system to have the dynamic adjustment ability of "combining rigidity and flexibility" and avoiding sacrificing path execution accuracy while pursuing safety.

[0089] When the hepatectomy of patient Mr. Li was approaching the final stage, the surgical robot was completing the trimming of the lesion edge along the critical resection path set by the doctor in the preoperative plan. Due to intraoperative tissue deformation and the dynamic changes of ultrasonic images, the path of the robot drifted slightly during the continuous obstacle avoidance process, and a part of the path segment deviated from the originally set task hot spot area. In particular, the path points from No. 31 to No. 38 deviated from the central axis of the high-task probability area Ω 73 and Ω 74 To evaluate in real time whether the current path still has task consistency and to determine whether to trigger the path recovery callback mechanism, the system starts the "path recovery potential energy function" in the present invention as an index for dynamic evaluation of path consistency, and its function expression is as follows:

[0090]

[0091] where, is the recovery potential energy of the path at time t, which is used to judge whether the path has seriously deviated from the task hot spot area; ξ × (t) is the spatial position trajectory of the i-th path point on the path, and the system tracks its displacement change in real time; represents the instantaneous velocity of this path point, reflecting whether there is unstable jitter in the path; γ i is the importance weight of the path point, and the value range is set to [0.6, 1.0]. For the path segment close to the edge of the resection target, the system automatically sets it to γ i = 0.95; λ i is the task traction weight coefficient of the path point, which is used to strengthen the pulling-back ability of the hot spot area on the path, and the setting range is [0.3, 1.2]. In the current stage, to strengthen the task adhesion, it is set to λ i = 1.0; and P r (ξ i (t)) is the task probability value of the current path point in the hot spot map, and this value is calculated from the real-time image.

[0092] Taking the path point ξ 33 (t) as an example, at the current position (x = 20.2, y = 38.1, z = 13.4), its speed is indicating that there is a relatively high jitter in the path point, and the task probability value has dropped from 0.92 previously to 0.61 currently, indicating that it has partially deviated from the high-task area. Substituting the data of this path point, the potential energy term is calculated as:

[0093]

[0094] The system calculates the current 12 key points (No. 31 to No. 42) on the path respectively and then sums them up to obtain the complete recovery potential energy value Higher than the system preset path deviation threshold of 18.0, which is automatically adjusted according to historical training data and task requirements, usually set in the range of [14.0, 20.0]. Currently, it is in the "high-rigidity path" state. Therefore, the system immediately determines that there is an obvious semantic deviation in the current path and triggers the path recovery mechanism. This mechanism includes three operations: one is to re-perform local replanning on the deviated path points to make them return to the original task hot area Ω 73 -Ω 74 Regression; the second is to dynamically limit the speed of the robot end, reducing the maximum propulsion speed from 3.2 mm / s to 1.0 mm / s to improve control accuracy; the third is to continuously monitor the change trend of the task probability values of all path points within the next 3 seconds and lock the maximum allowable deviation of each point not exceeding 0.3 mm.

[0095] After executing the recovery strategy, the system completes the path correction within 0.9 seconds. All the 33rd to 37th path points are restored within ±0.2 mm of the axis of the task hot center area. The task probability value is increased to an average of 0.89, and the recovery potential energy value is decreased to 11.72, far lower than the threshold. The robot smoothly enters the terminal resection action. This example illustrates the engineering practicability and mathematical logic of the path recovery potential energy function in the present invention. By integrating path jitter, task deviation degree, and semantic region adhesion force, a quantifiable, real-time, and triggerable recovery evaluation system is constructed, providing a "soft limit + active regression" navigation guarantee mechanism for high-precision surgical robots, effectively preventing path drift caused by obstacle avoidance from affecting the consistency of task execution.

[0096] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A control method for surgical robot navigation and obstacle avoidance, characterized in that The following steps are involved: S1. Dynamic construction of continuous state space using intraoperative process modeling: S1.1, collect tissue, tool, and robot posture data in real time, and calculate the continuously changing tensor field; S1.

2. Establish a process field map to convert the changing trend in the surgical environment into a dynamic state space that can be used for navigation; input the state space into the subsequent navigation control model to enable forward-looking modeling of environmental trends; S2. Adaptive evolution of obstacle avoidance strategy using risk threshold drift model: S2.

1. In path planning, set an initial risk value for each potential conflict area; Monitor environmental changes in real time and evaluate risk trends through a time window sliding mechanism; S2.2, introduce a dynamic obstacle avoidance weight function to control the triggering threshold of the obstacle avoidance strategy, and adaptively avoid or restore the original path according to the risk trend; S3. Path fine-tuning constraints using navigation task integrity protection: S3.

1. Set the spatial probability distribution of the path target area including the mission hotspot area; S3.2, introducing the distribution as a constraint factor in obstacle avoidance path generation to limit the offset range of path search; S4. Collaborative navigation optimization using human-machine strategy intention fusion: S4.

1. Record the doctor's micro-manipulation behaviors on the operation terminal, including small-range fine-tuning and path locking; Use operational behaviors to establish intention preference models and use them as dynamic navigation strategy adjustment factors; S4.

2. When the robot's obstacle avoidance strategy conflicts with the doctor's intention, the collaborative optimizer is triggered to achieve soft fusion of the navigation strategy: moderate concessions are made while ensuring safety and control consistency.

2. The control method for surgical robot navigation and obstacle avoidance according to claim 1, characterized in that The continuous state space dynamic construction method for intraoperative process modeling includes: A dynamic tensor field modeling method is adopted to perceive the state of the surgical area from two dimensions of time and space, and organize the multimodal perception data into a four-dimensional tensor field; the tensor field is used to express the trend of each spatial position changing over time during the operation, and to perceive in advance the areas that will change in the future.

3. The control method for surgical robot navigation and obstacle avoidance according to claim 2, characterized in that The continuous state space dynamic construction method for intraoperative process modeling includes: Based on the tensor trend, the surgical process field map is constructed to form a dynamic causal network between regions, decompose the surgical area into multiple dynamic units with causal relationships, and characterize the interaction pattern between them; trend tensor function The following inter-region coupling is defined: in: φ(Ω i ,Ω j ) represents the area Ω i With area Ω j The degree of dynamic causal coupling in the time interval [t0, t1]; is the time derivative of the regional tensor trend function, indicating the degree of drastic change in the regional state; · is the dot product operation, used to evaluate whether the trend directions of the two regions are consistent; |·| is the absolute value, used to eliminate directional differences; β is the coupling sensitivity coefficient, used to adjust the sensitivity of different tissues or surgical types to chain reactions.

4. The control method for surgical robot navigation and obstacle avoidance according to claim 3, characterized in that The continuous state space dynamic construction method for intraoperative process modeling includes: Compress the graph, extract the core trend information, and generate the evolution state space of the navigation controller input, that is, the function Ψ(t) that comprehensively expresses the risk change trend, mission area stability, and path feasibility factors; define the following aggregation expression: in: Ψ(t) is the evolution state space function, which is the dynamic input of the navigation control module; n is the number of dynamic regions currently involved in modeling; γ i (t) is the importance weight of the ith region, which is determined by whether it is a critical operation path; μ represents the time change rate of the coupling relationship between this area and all other areas, that is, the risk transmission speed; i (t) represents the stability of the regional structure, that is, the rigidity requirement for the execution of the task objectives in the region; It is a nonlinear fusion operation used to merge the interactive relationship between the trend change rate and the structural stability; limΔt→0 is a limit operation that makes the state space time-continuous and reflects the potential risk pattern in the surgical area at every moment.

5. The control method for surgical robot navigation and obstacle avoidance according to claim 1, characterized in that The path fine-tuning constraint method for protecting the integrity of the navigation task includes: dividing the entire surgical operation area into spatial units of navigation semantics based on preoperative imaging data and real-time intraoperative perception information, and dynamically assigning a task association probability to each unit based on the degree of association with the surgical target, anatomical importance, and operability, to form a task hotspot area map in a spatial sense, which is used to express the distribution trend of the priority traversal area in the robot's navigation path.

6. The control method for surgical robot navigation and obstacle avoidance according to claim 5, characterized in that The path fine-tuning constraint method for navigation task integrity protection includes: In the path planning process, the probability density map is used to restrict the obstacle avoidance search space and construct an offset range control function with task semantic constraints, so that the path search is only allowed to be fine-tuned near the high task weight area, and the following restriction formula is constructed: Δ lim (x,y,z)=ρ·(1-P r (x,y,z,t)) η Among them, Δ lim (x, y, z) represents the maximum allowable offset of the path at position (x, y, z); ρ is the overall path offset scale factor, which determines the basic size of the offset limit; P r (x, y, z, t) is the current task priority probability value of the location point; η is the nonlinear control index, which is used to define the boundary transition rate of the task hotspot area.

7. The control method for surgical robot navigation and obstacle avoidance according to claim 6, characterized in that The path fine-tuning constraint method for navigation task integrity protection includes: The path recovery potential energy function is introduced to dynamically evaluate the task consistency of the current path. When the potential energy exceeds the set threshold, the path callback mechanism is triggered to restore to a better task-associated path trajectory. The recovery potential energy function is defined as follows: in, represents the recovery potential value of the entire path at time t, which is used to measure the severity of the semantic deviation of the path; ξ i (t) represents the spatial position of the i-th path point in the current path over time; is the square term of the moving speed of the path point, which is used to reflect the fluctuation of the path; γ i is the importance weight of the path point; i is the offset sensitivity coefficient of the path point, which is used to strengthen the traction effect of the task hotspot on the path; P r (ξ i (t)) is the current task probability value of the waypoint, indicating whether the waypoint is still in the semantic hotspot area.

Images