A method for self-orthogonal scan control of an abdominal ultrasound robot

Abstract

This invention relates to the field of ultrasound examination auxiliary control, specifically disclosing a self-orthogonal scanning control method for an abdominal ultrasound robot. The method includes the following steps: system initialization, defining the end-effector coordinate system and configuring the hardware; then, by sampling the point cloud through the camera, fitting the surface normal to key points and neighborhoods along the scanning path, completing coarse alignment, and switching to high-frequency force feedback fine-tuning. The force-position hybrid control main loop is initiated, and the S variable is set according to the contact force state: if the contact force meets the standard, scanning proceeds along the set direction; otherwise, compensation is performed in a specific direction. This method, with "force feedback self-orthogonal" as its core, can adapt to complex curved surfaces and respiratory motion scenarios, optimize imaging quality and acoustic coupling effects, suppress measurement errors, and has strong anti-interference capabilities and high hardware compatibility, reducing the difficulty of implementation.

Description

Technical Field

[0001] This invention relates to the field of ultrasound examination auxiliary control, specifically to a self-orthogonal scanning control method for an abdominal ultrasound robot. Background Technology

[0002] Ultrasound examination is a widely used non-invasive imaging method in clinical practice, especially valuable in the examination of abdominal organs. Traditional ultrasound examinations rely on manual operation, requiring the ultrasound physician to hold the probe and move and compress it on the patient's body surface to obtain stable and clear ultrasound images. However, manual operation has shortcomings in repeatability, stability, and fatigue from prolonged operation, easily leading to unstable image quality. Therefore, in recent years, ultrasound robots based on robotic arms have gradually become a research hotspot. Utilizing the robot's precise posture control and force control capabilities to replace manual operation can effectively improve the standardization and objectivity of ultrasound examinations.

[0003] Existing ultrasound robot control technologies can be broadly categorized into three types: position control, force control, and force-position hybrid control. Position control ensures the accuracy of the robot's end effector trajectory, but it struggles to maintain a stable force when in contact with the human body, potentially leading to poor contact or excessive pressure between the probe and skin. Pure force control maintains constant contact pressure better, but sacrifices trajectory accuracy, resulting in scanning path deviation. Therefore, some studies have proposed a force-position hybrid control strategy. During scanning, this strategy utilizes force sensors to maintain a constant contact force between the probe and the human body, while simultaneously using position control to achieve scanning along a predetermined path, thus balancing image quality and trajectory accuracy.

[0004] However, existing force-position hybrid control technologies generally suffer from a deficiency: a lack of real-time control over the perpendicularity of the probe to the contact surface. Maintaining the probe surface perpendicular to the normal of the body surface is a critical requirement during ultrasound examinations. If the probe angle deviates, the incident direction of the ultrasound waves will form an angle with the tissue surface, leading to decreased image resolution, increased artifacts, or even failure to obtain the target cross-section. Furthermore, in abdominal ultrasound examinations, due to the complex curvature of the human abdomen and its dynamic deformation during respiration, relying solely on robotic arm trajectory planning or simple force control often fails to ensure that the probe remains in a self-orthogonal state throughout the scanning process.

[0005] Some existing studies attempt to use 3D vision to reconstruct surface normal vectors to provide a preliminary orthogonal reference for the probe. However, these methods are limited by the resolution of the vision sensor, lighting conditions, and reconstruction frequency, resulting in shortcomings in both real-time performance and accuracy, making them unreliable in continuous dynamic scanning. On the other hand, relying solely on force sensor feedback, while reflecting the contact state between the probe and the body surface to some extent, is affected by friction during moving scans, leading to distortion of the lateral component readings of the force sensor and thus affecting the accuracy of orthogonal control.

[0006] In summary, while existing ultrasound robot control methods have made some progress in force and position control, they still lack effective techniques to ensure real-time self-orthogonality between the probe and the contact surface. This is particularly true in abdominal organ ultrasound scanning, where the probe needs to move smoothly along the path while maintaining stable pressure at fixed points, placing higher demands on the real-time performance and robustness of orthogonal control. Therefore, there is an urgent need to propose a novel self-orthogonal control method that combines force sensor feedback and friction modeling to improve the imaging stability and accuracy of ultrasound robots in abdominal scanning tasks. Summary of the Invention

[0007] The purpose of this invention is to overcome the deficiencies of the prior art and provide a self-orthogonal scanning control method for abdominal ultrasound robots to solve the problems mentioned in the background art.

[0008] This invention provides a self-orthogonal scanning control method for an abdominal ultrasound robot, comprising:

[0009] System initialization and parameter configuration: Establish the coordinate system of the robotic arm end effector, define the Z-axis as the direction from the probe normal to the body surface, the Y-axis as the main movement direction of the probe, and the X-axis as the direction within the sector;

[0010] Visual prior acquisition and coarse alignment are performed. The camera samples local point clouds and sets key points at fixed intervals in the scanning path. The surface normal is fitted at the key points and their neighborhoods as a reference for the target normal.

[0011] The force-position hybrid control main cycle starts, reads the end pose and six-axis force data, and fine-tunes the end Z-direction displacement based on Z-axis force feedback to keep the contact force within the target range. When the contact force meets the requirements, a small step displacement is performed along the Y-axis and the state variable S=1 is set; if the contact force does not meet the requirements, the Y-direction displacement is paused and only Z-direction compensation is performed, and the state variable S=0 is set.

[0012] When the state variable S=0, monitor the lateral force F in the in-plane direction. x out-of-plane lateral force F y When any axis meets the condition, attitude fine-tuning is triggered;

[0013] Out-of-plane lateral force Fy When the plane crosses the boundary, a pitch correction around the X-axis is applied to eliminate out-of-plane deviation; the in-plane lateral force F x When the boundary is exceeded, a yaw correction around the Y-axis is applied to eliminate in-plane deviation, and each correction is subject to integral limiting.

[0014] When the state variable S=1, during the calibration phase, a short-range scan is performed along the Y direction in a flat area, and the probe's normal contact force F is recorded. z out-of-plane lateral force F y The coefficient of kinetic friction is calculated, and the equivalent lateral force F of the force feedback from the orthogonal stage is constructed. y ';With equivalent lateral force F y 'Replace the original out-of-plane lateral force F y With F x They serve as orthogonal criteria.

[0015] Furthermore, key points are set at fixed intervals, specifically every 2cm.

[0016] Furthermore, the parameter configuration includes: configuring a six-degree-of-freedom robotic arm, a six-dimensional force / torque sensor, an RGB-D camera, and a real-time controller, with the force sensor coordinates consistent with the end-effector coordinates; setting the target contact force range, lateral force threshold, out-of-bounds counting threshold, and attitude fine-tuning angle.

[0017] Furthermore, the acquisition of visual priors and coarse alignment includes: when the robot reaches the key point, coarse alignment of the end pose is performed according to the target normal reference. After completion, it switches to high-frequency force feedback closed-loop fine adjustment. The visual priors play the role of initial alignment and trend guidance.

[0018] Furthermore, the specific condition for triggering attitude fine-tuning when any axis meets the condition is: attitude fine-tuning is triggered when any axis exceeds the ±0.5N threshold for 20 consecutive control cycles.

[0019] Furthermore, the equivalent lateral force F y '=F y +μ·F z μ is the coefficient of kinetic friction, F z This represents the normal contact force of the probe.

[0020] Furthermore, it also includes dynamic updates and iterative optimization: the μ value is periodically re-estimated based on the scanning site, body surface conditions and coupling agent status, the visual prior weights are dynamically adjusted, and the control parameters are continuously optimized to ensure stability under complex curved surfaces and respiratory movements.

[0021] Furthermore, the camera samples local point clouds, sets key points at fixed intervals along the scanning path, and fits the surface normal vector to the key points and their neighborhoods as a target normal vector reference. This includes: preprocessing the RGB-D camera before sampling the local point cloud to avoid interference factors and ensure that the point cloud truly reflects the geometric shape of the abdominal surface; the sampling process is synchronized with the robotic arm's scanning path planning; the robotic arm initiates the first sampling before entering the area to be scanned, and subsequently continues sampling at preset intervals to dynamically capture changes in the surface morphology caused by breathing; when setting key points along the scanning path, the distribution is combined with the planned distribution of the abdominal scanning target area to ensure full coverage; when fitting the surface normal vector to the key points and their neighborhoods, the neighborhood point cloud is first extracted and denoised, and then the point cloud is fitted into a local plane using a fitting algorithm to calculate the normal vector as a target normal vector reference; after fitting, the quality of the target normal vector reference is evaluated, and if the deviation exceeds the range, the normal vector reference is discarded.

[0022] Furthermore, visual prior acquisition and coarse alignment are performed, including: when the robot reaches the key point, the end pose is coarsely aligned according to the target normal reference, and after completion, it switches to high-frequency force feedback closed-loop fine adjustment. The visual prior plays the role of initial alignment and trend guidance.

[0023] Furthermore, the out-of-plane lateral force F y When the plane crosses the boundary, a pitch correction around the X-axis is applied to eliminate out-of-plane deviation; the in-plane lateral force F x When the boundary is exceeded, a yaw correction around the Y-axis is applied to eliminate in-plane deviations. Each correction includes integral limiting, including:

[0024] out-of-plane lateral force F was detected y When crossing the boundary, according to the out-of-plane lateral force F y Positive and negative values ​​indicate the direction of the probe's outward deflection and the outward lateral force F. y If the positive boundary is exceeded, the robotic arm's end effector will adjust its pitch around the X-axis in the opposite direction, with an out-of-plane lateral force F. y If the negative direction exceeds the limit, adjust the pitch in the positive direction. Prioritize precise adjustment with direction correspondence to eliminate the normal deviation outside the fan surface and ensure that the adjustment matches the direction of the deviation.

[0025] In-plane lateral force F was detected x When crossing the boundary, the in-plane lateral force F x Positive and negative values ​​indicate the in-plane skewness of the probe and the in-plane lateral force F. x If the positive boundary is exceeded, the robotic arm's end effector will adjust its yaw around the Y-axis in the opposite direction, with an in-plane lateral force F. x If the negative yaw exceeds the limit, the positive yaw adjustment will specifically eliminate the normal deviation within the sector to ensure that the probe scanning surface is aligned with the normal of the body surface; each attitude fine-tuning adopts incremental adjustment, and an integral limit is set for the cumulative correction angle.

[0026] The beneficial effects of this invention are:

[0027] This invention takes "force feedback self-orthogonality" as its core, and within a unified force-position mixing main cycle, uses F z The contact force is stabilized within the target range. Only after this range is met is a scan performed along the Y-axis of the end coordinate system {E}. If the target range is not met, only Z-axis force control is applied. The six-dimensional force sensor is calibrated in the same direction as {E}, and the Z-axis is aligned with the probe normal, ensuring the mapping of "lateral force ≈ attitude error" holds true. The fixed-point working condition is based on F... x F y Approaching zero is used as the self-orthogonality criterion; for moving scan conditions, first press "μ≈Σ|F". y | / ΣF z "Perform dynamic friction coefficient calibration, and then construct equivalent lateral force F online." y '=F y +μ·F z To restore the correctness of the criteria. At the same time, in order to accelerate convergence without destroying high-frequency stability, the RGB-D surface normal vector can be obtained at low frequency and coarsely aligned with limited weights before entering each task segment. If the quality is insufficient, the weights will be automatically reduced or ignored.

[0028] The end coordinate system is calibrated in the same direction as the force sensor, and the setting of "Z-axis = probe normal" is implemented; the force-position hybrid main cycle and S-state discrimination are performed; the near-zero lateral force is used as the self-orthogonality criterion (F is used for fixed points). x F y Mobile F y The statistical calibration of the dynamic friction coefficient; robust triggering and small-angle pitch / yaw fine-tuning using "threshold band + continuous out-of-bounds counting"; and optional coarse alignment of low-frequency visual normal prior are all technical advancements of this application. Details of the technical effects also include:

[0029] Fixed-point scenario: using a six-dimensional force sensor to measure lateral force (F) x F y Using near-zero as the criterion, and combining the "threshold band + continuous out-of-bounds counting" mechanism to suppress noise interference, the probe's posture can be precisely fine-tuned.

[0030] Mobile scanning scenario: The dynamic friction coefficient μ (μ≈Σ|F) is calibrated statistically. y | / ΣF z Constructing an equivalent lateral force F y '=F y +μ·F z Eliminating the systematic bias of frictional force on the orthogonality criterion, making the lateral force nearly zero. The mapping relationship of "normal alignment" is restored to its validity during movement.

[0031] Complex surfaces and respiratory motion adaptation: Low-frequency RGB-D visual prior (2s / time) provides coarse alignment of the body surface normal, reducing tracking error in high-frequency force control closed loop; force control main loop (200H) zIt can compensate for dynamic deformation in real time and achieve dynamic maintenance of verticality.

[0032] Physical-level optimization of image quality.

[0033] Acoustic coupling and signal strength enhancement: Self-orthogonal control ensures that ultrasound waves are incident orthogonally, reducing specular reflection and interface artifacts caused by oblique incidence, and significantly enhancing the contrast and coherence of soft tissue boundaries such as the liver and gallbladder.

[0034] Systematic error suppression: In angle-sensitive modes such as elastic imaging, self-orthogonal control can reduce measurement errors caused by pressure fluctuations and incident angle deviations, and improve imaging consistency across operators and time points.

[0035] Robustness and efficiency of engineering implementation

[0036] Anti-interference design: Delayed triggering mechanism (20-cycle over-limit threshold) effectively filters force sensor noise and skin micro-deformation interference, avoiding ineffective adjustment; small angle attitude correction (3° / time) prevents Euler angle coupling oscillation and ensures control stability.

[0037] Hardware compatibility: The design of calibrating the six-dimensional force sensor in the same direction as the end coordinate system and aligning the Z-axis with the probe normal allows the force signal to be directly used for control law design without the need for complex coordinate transformations, thus reducing the difficulty of engineering implementation. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the overall system structure of the present invention;

[0039] Figure 2 This is the overall control flowchart (main state machine) of the present invention.

[0040] Figure 3 Flowchart of visual prior normal vector acquisition and coarse alignment in this invention;

[0041] Figure 4 The self-orthogonal loop branching flowchart of this invention;

[0042] Figure 5 The force-position hybrid control flowchart of this invention;

[0043] Figure 6 Flowchart of the dynamic friction coefficient μ calibration of this invention;

[0044] Figure 7 The method flowchart of the present invention. Detailed Implementation

[0045] In its specific implementation, this application discloses a self-orthogonal scanning control method for an abdominal ultrasound robot, such as... Figure 7 The method for controlling the autonomous orthogonal scanning of an abdominal ultrasound robot includes the following steps:

[0046] S100: System initialization and parameter configuration, establish the coordinate system of the robotic arm end effector, define the Z-axis as the direction from the probe normal to the body surface, the Y-axis as the main movement direction of the probe, and the X-axis as the direction within the sector.

[0047] In practice, the geometric center of the end flange of the robotic arm is taken as the origin of the end coordinate system of the robotic arm. This origin is also coaxially aligned with the center of the six-dimensional force / torque sensor and the mounting reference center of the probe fixture, ensuring that the force sensor coordinate system and the end coordinate system are completely coincident.

[0048] The Z-axis must be precisely matched to the effective scanning surface of the ultrasound probe: Using the geometric center of the scanning surface where the probe contacts the body surface as a reference, the Z-axis is set along the normal direction of this scanning surface, with the positive direction of the Z-axis pointing towards the patient's body surface. This ensures that the probe normal is completely aligned with the Z-axis, allowing for subsequent force feedback via the Z-axis (F... z When adjusting the contact force, it can directly affect the normal contact relationship between the probe and the body surface. The Y-axis, as the main direction of probe movement, needs to be parallel to the patient's body surface and located in the "out-of-plane" space outside the probe's fan-shaped area: when the plane containing the probe's fan-shaped area (i.e., the fan-shaped area formed by the propagation of ultrasound) is the XZ plane, the Y-axis extends along the tangent direction of the body surface and is perpendicular to the XZ plane, ensuring that when performing small-step displacements along the Y-axis, the probe scanning path always does not overlap with the fan-shaped area, avoiding repeated or missed scanning areas.

[0049] The X-axis is defined in the "in-plane" space where the probe sector is located, perpendicular to the Z-axis and lying in the XZ plane. Its positive direction can be set from the starting scanning end of the probe sector to the ending scanning end, so that the lateral force (F) in the X-axis direction can be monitored subsequently. x When the probe is in the fan-shaped area, it can directly reflect the attitude deviation of the probe within the fan-shaped area, providing a clear directional reference for yaw correction around the Y-axis.

[0050] In addition, after the coordinate system is established, it needs to be verified by a laser calibration tool to ensure that the coaxiality deviation between the probe normal and the Z-axis does not exceed 0.5°, the parallelism deviation between the Y-axis movement direction and the preset scanning path is controlled within 1°, and the angle error between the X, Y, and Z axes of the force sensor and the corresponding axes of the end coordinate system is less than 0.3°, so as to ensure the coordinate consistency between force signal acquisition and pose control, and can be directly used for control law design without additional coordinate transformation.

[0051] The parameter configuration includes: configuring a six-degree-of-freedom robotic arm, a six-dimensional force / torque sensor, an RGB-D camera, and a real-time controller to ensure that the force sensor coordinates are consistent with the end-effector coordinates; setting the target contact force range, lateral force threshold, out-of-bounds counting threshold, and attitude fine-tuning angle.

[0052] In practice, when configuring a six-degree-of-freedom robotic arm, a model suitable for abdominal ultrasound scanning scenarios must be selected, with a range of motion covering the entire abdominal examination area and positioning accuracy meeting the requirements of ultrasound imaging. Its end effector must be designed with a standard mounting interface for securely connecting a six-dimensional force / torque sensor, ensuring the relative position of the sensor and probe remains stable during robotic arm movement. The six-dimensional force / torque sensor is mounted between the robotic arm end effector and the probe holder using a custom fixture. After installation, coordinate calibration is required to ensure complete alignment between the sensor coordinate system and the robotic arm end effector coordinate system, guaranteeing that the collected force data directly reflects the contact force state between the probe and the body surface without requiring additional coordinate transformation.

[0053] The RGB-D camera is mounted on the side of the probe fixture. The mounting angle and height need to be adjusted to ensure that the camera can clearly capture the 3D point cloud of the body surface near the probe scanning area. Furthermore, the relationship between the camera coordinate system and the end effector coordinate system of the robotic arm needs to be established through hand-eye calibration so that the point cloud data can be accurately mapped to the end effector coordinate system for body surface normal fitting.

[0054] The real-time controller needs to have high-frequency data processing and command output capabilities, be able to simultaneously receive robotic arm pose data, six-axis force data from force sensors, and point cloud data from RGB-D cameras, and be able to stably run force-position hybrid control algorithms to ensure the continuity and real-time nature of the control loop and meet the needs of dynamic adjustment during the scanning process.

[0055] When setting parameters, the target contact force range needs to be determined in conjunction with the pressure-bearing characteristics of abdominal soft tissue. This ensures sufficient acoustic energy coupling between the probe and the body surface to obtain clear images, while avoiding excessive pressure that could cause patient discomfort or tissue damage. The lateral force threshold needs to match the detection accuracy of the force sensor to effectively identify lateral force changes caused by probe posture deviations, while eliminating misjudgments caused by sensor noise. The out-of-bounds counting threshold needs to consider interference factors such as sensor noise and micro-deformation of the skin, filtering out instantaneous interference through continuous periodic judgments, triggering posture fine-tuning only when the lateral force continuously exceeds the limit. The posture fine-tuning angle needs to balance correction efficiency and control stability, avoiding probe posture oscillation caused by excessively large single correction angles, and setting limits on cumulative fine-tuning angles to prevent over-correction from affecting imaging. Furthermore, a parameter adaptation module needs to be configured to flexibly adjust the above parameters according to the scanning site (e.g., liver, gallbladder, spleen, kidneys), body surface condition (e.g., skin smoothness, fat thickness), and coupling agent usage, ensuring reliable self-orthogonal control under different operating conditions.

[0056] The method for controlling the autonomous orthogonal scanning of an abdominal ultrasound robot includes the following steps: S200: performing visual prior acquisition and coarse alignment, sampling local point clouds with an RGB-D camera, setting key points at fixed intervals in the scanning path (specifically, setting key points at fixed intervals every 2cm), and fitting the surface normal at the key points and their neighborhood as a reference for the target normal.

[0057] In practice, when the RGB-D camera performs local point cloud sampling, the sampling area needs to be preprocessed to avoid interference factors such as coupler mirror reflection, clothing edges, and body hair, so as to ensure that the collected point cloud data can truly reflect the geometric shape of the abdominal body surface. The sampling process needs to be synchronized with the scanning path planning of the robotic arm. The first point cloud sampling is started before the robotic arm enters the area to be scanned, and subsequent sampling is carried out at preset intervals as the scanning progresses, so as to dynamically capture the changes in body surface morphology caused by respiratory movements.

[0058] When setting key points in the scanning path, the distribution of key points should be planned in conjunction with the target area of ​​the abdominal scan (such as the projection range of organs such as liver, gallbladder, spleen, and kidney) to ensure that the key points can cover the entire scanning path. The spacing between adjacent key points should take into account both the accuracy of coarse alignment and computational efficiency. This avoids both insufficient normal reference due to excessively sparse spacing and excessively dense spacing, which would increase the computational burden of point cloud processing and fitting.

[0059] When fitting the surface normal to key points and their neighborhoods, it is necessary to first extract the neighborhood point cloud within a certain range around the key point from the sampled point cloud. Then, an abnormal noise point (such as external interference points or isolated points) within the neighborhood is filtered out using a point cloud denoising algorithm. Next, a fitting algorithm suitable for local surfaces (such as the least squares method) is used to fit the neighborhood point cloud into a local plane, and then the normal vector of this plane is calculated as the target normal reference. After the fitting is completed, the quality of the target normal reference needs to be evaluated. Valid references are selected by judging the degree of deviation between the neighborhood point cloud and the fitted plane. If the deviation exceeds the set range, the normal reference of the key point is discarded to avoid deviations in subsequent coarse alignment due to point cloud quality issues, and to ensure that the target normal reference can provide a reliable directional basis for the coarse alignment of the end pose.

[0060] The process of acquiring visual priors and coarse alignment includes: when the robot reaches a key point, coarse alignment of the end-effector posture is performed based on the reference. After completion, the robot switches to high-frequency force feedback closed-loop fine adjustment. The visual priors play a role in initial alignment and trend guidance.

[0061] In practice, when the robot moves with the probe to a key point on the scanning path, it first verifies the validity of the previously fitted target normal reference, eliminating invalid references caused by temporary deformation of the body surface due to the patient's breathing, coupling agent flow, or point cloud sampling deviation. If the verification finds that the target normal reference has a large deviation, it automatically uses the most recent valid target normal reference, or appropriately reduces the weight of the visual prior, to avoid deviations in probe posture adjustment caused by incorrect references.

[0062] Subsequently, the controller calculates the adjustment amount of the robotic arm's end-effector posture based on the effective target normal reference, and gradually moves the probe normal toward the target normal using an incremental adjustment method. During the adjustment process, the contact state between the probe and the body surface is monitored in real time to prevent sudden changes in contact force or probe detachment from the body surface due to excessive posture adjustment.

[0063] After coarse alignment is completed, the deviation between the adjusted probe normal and the target normal is compared to determine whether the standard is met. If the deviation is within the preset coarse alignment accuracy range, the high-frequency force feedback closed loop is immediately switched to fine adjustment. If the deviation exceeds the range, the point cloud data of the neighborhood of the key point is extracted again and the target normal is refitted, and the coarse alignment operation is performed again.

[0064] Throughout the process, the visual prior only undertakes the initial alignment and trend guidance functions and does not participate in high-frequency real-time control. Its weight setting needs to balance the convergence speed and control stability to ensure that even if there is slight noise in the visual data, it will not interfere with the stable operation of the subsequent force control closed loop, so as to achieve the synergistic effect of "coarse alignment to shorten the convergence time and force control fine adjustment to ensure control accuracy", while avoiding lag or misjudgment in dynamic scanning due to over-reliance on visual data.

[0065] The abdominal ultrasound robot self-orthogonal scanning control method includes the following steps: S300: Force-position hybrid control main cycle starts, control cycle is started, end position and six-axis force data are read, and the end Z-direction displacement is finely adjusted based on Z-axis force feedback to keep the contact force within the target range. When the contact force meets the requirements, a small step displacement is performed along the Y-axis and the state variable S=1 is set (moving condition); if the contact force does not meet the requirements, the Y-direction displacement is paused and only Z-direction compensation is performed, and the state variable S=0 is set (fixed point condition).

[0066] In practice, when the force-position hybrid control main cycle starts, it must first be synchronized with the high-frequency control cycle of the real-time controller to ensure that each cycle is executed according to the preset rhythm, avoiding delays in data reading or command output due to cycle disorder. After starting the control cycle, the real-time pose data (including position and attitude information) of the robotic arm end effector and the six-axis force data of the six-dimensional force sensor are read synchronously first. Since the force sensor has been calibrated in the same direction as the end effector coordinate system, the read force data can be directly used for control calculation without additional coordinate transformation, ensuring the accuracy of data mapping.

[0067] When fine-tuning the Z-axis displacement of the end based on Z-axis force feedback, it is necessary to first adjust the real-time acquired Z-axis force (F) z Compare F with the preset target contact force range: If F z If the pressure is below the lower limit of the range, the robotic arm's end effector is slowly moved along the positive Z-axis, gradually increasing the contact pressure between the probe and the body surface; if F z If the pressure is higher than the upper limit of the interval, then make a slight adjustment along the negative Z-axis to reduce the contact pressure; if Fz If it is within the range, then maintain the current state. Z Towards position. An incremental adjustment strategy is used during fine-tuning, based on F... z The displacement step size is dynamically adjusted according to the magnitude of the deviation from the target range. The larger the deviation, the larger the step size is to accelerate convergence, and the smaller the deviation, the smaller the step size is to avoid overshoot, ensuring that the contact force smoothly approaches and remains within the target range.

[0068] When F is detected z When the contact force remains stable within the target contact force range (not instantaneously satisfied), the contact force is deemed to meet the requirements. At this point, the robotic arm is controlled to perform small-step displacements along the Y-axis (probe's main movement direction) of the end-effector coordinate system. After each small-step movement, the force is reconfirmed. z If the target range is still met, continue with the next small step movement and set the state variable S to 1, marking the current movement condition; if F is in the movement process... z If the target range is deviated from, immediately pause the Y-axis displacement and focus solely on performing Z-axis compensation adjustment, waiting for F... z After returning to the target range, determine whether to resume Y-axis movement based on actual needs, and set the state variable S to 0 to mark the current fixed-point condition.

[0069] Throughout the process, the end-effector pose data needs to be monitored in real time to ensure that the Y-axis displacement always follows the preset scanning path, avoiding deviation of the scanning path due to the movement error of the robotic arm. At the same time, when switching between Z-axis fine adjustment and Y-axis movement, smooth transition logic needs to be added to prevent the robotic arm movement from impacting the contact stability between the probe and the body surface during the instantaneous switching of working conditions, thereby ensuring the continuity of ultrasound imaging quality.

[0070] The abdominal ultrasound robot self-orthogonal scanning control method includes the following steps: S400: When the state variable S=0, continuously monitor the in-plane lateral force F. x out-of-plane lateral force F y When any axis meets the condition, attitude fine-tuning is triggered; specifically, attitude fine-tuning is triggered when any axis exceeds the ±0.5N threshold for 20 consecutive control cycles.

[0071] In practice, when the state variable S=0 (fixed-point condition), the controller will synchronize with the cycle of the force-position hybrid control main loop and read the in-plane lateral force F collected by the six-dimensional force sensor in real time. x Out-of-plane lateral force F y The data, because the force sensor has been calibrated in the same direction as the coordinate system of the robotic arm's end effector, reads the F... x F y It can directly reflect the alignment deviation between the probe and the body surface normal, without the need for additional coordinate transformation.

[0072] During continuous monitoring, F needs to be monitored. x and F y Perform threshold judgment separately: read the F value in real time. x Compared with the preset lateral force threshold, and simultaneously F y Compare with the same threshold. For transient out-of-bounds errors caused by force sensor noise, localized micro-deformation of the skin, or coupling agent fluctuations, a value of F is required. x and F y Set independent out-of-bounds counting variables: if the force value of a certain axis exceeds the threshold, the corresponding counting variable is incremented by 1; if the force value of that axis returns to the threshold range, the corresponding counting variable is immediately reset to 0, to avoid invalid adjustment triggered by a single instantaneous disturbance.

[0073] When the out-of-bounds count variable of any axis reaches the preset continuous out-of-bounds threshold, it is determined that there is a continuous attitude deviation on that axis, and attitude fine-tuning is triggered. Fine-tuning strictly follows the correspondence between the deviation axis and the correction direction: if it is F... y If the deviation exceeds the limit (out-of-plane deviation), the robotic arm's end effector will perform pitch correction around the X-axis, prioritizing the elimination of the probe's normal deviation outside the sector; if it is F... x If the probe goes out of bounds (in-plane deviation), the robotic arm's end effector performs yaw correction around the Y-axis to eliminate the probe's normal deviation within the sector. Each attitude fine-tuning uses incremental adjustment, and an integral limit is set for the cumulative correction angle to prevent excessive angle shift or oscillation due to continuous fine-tuning. This ensures that the probe's attitude can stably maintain a self-orthogonal state with the body surface after correction, while avoiding affecting the stability of the contact force between the probe and the body surface.

[0074] Out-of-plane lateral force F y When the boundary is exceeded, pitch correction around the X-axis is applied, prioritizing the elimination of out-of-plane deviations; lateral force F in the in-plane direction. x When the boundary is exceeded, a yaw correction around the Y-axis is applied to eliminate in-plane deviation. Each correction is subject to integral limiting to prevent oscillation.

[0075] In practice, this specifically means that when an out-of-plane lateral force F is detected... y When crossing the boundary, firstly according to F y The positive and negative directions determine the direction of the probe's deflection outside the fan-shaped area. If F y If the probe is tilted to one side outside the fan-shaped area, the robotic arm's end effector should be controlled to perform a reverse pitch adjustment around the X-axis (inside the fan-shaped area); if F y If the negative deviation is detected, a positive pitch adjustment is performed. This precise adjustment in the corresponding direction prioritizes eliminating the normal deviation of the probe outside the fan-shaped area, ensuring that the adjustment direction is perfectly matched with the deviation direction.

[0076] When the in-plane lateral force F is monitored x When crossing the boundary, the same applies to F. x The positive or negative value of F indicates the deflection of the probe within the sector.x When the robot arm crosses the boundary in the forward direction, it controls the end effector to perform reverse yaw adjustment around the Y-axis (probe normal); F x When the negative boundary is exceeded, positive yaw adjustment is performed to specifically eliminate the normal deviation of the probe in the sector, ensuring that the probe scanning surface is always aligned with the normal of the body surface.

[0077] Each attitude correction employs an incremental adjustment method, meaning that each adjustment changes only a small angle to avoid sudden changes in probe attitude due to excessively large single adjustments, which could affect the stability of contact with the body surface. Simultaneously, to prevent excessive angle deviation or oscillation caused by multiple consecutive corrections, a cumulative angle integration limit is set for each correction channel (pitch and yaw). When the cumulative correction angle of a channel reaches the set upper limit, further correction for that channel is paused, and adjustment is only considered again after the force signal feedback deviation decreases.

[0078] During the correction process, it is also necessary to monitor the Z-axis contact force F in real time. z If the change in F is due to attitude adjustment z If the target range is deviated from, immediately pause attitude correction and prioritize Z-axis force control compensation, waiting for F... z After stabilizing within the target range, the attitude correction operation resumes to ensure that attitude adjustment and contact force stability do not interfere with each other, thus maintaining the contact conditions required for ultrasound imaging. Furthermore, the corrected F value is read immediately after each correction. x or F y If the force value returns to the threshold range, the correction of the current axis is stopped to avoid continuous ineffective adjustment and further ensure the stability and efficiency of control.

[0079] The method for controlling the autonomous orthogonal scanning of an abdominal ultrasound robot includes the following steps:

[0080] S500: When the state variable S=1, during the calibration phase, a short-range scan is performed along the Y direction in a flat area, and the probe normal contact force F is recorded. z out-of-plane lateral force F y The dynamic friction coefficient is calculated, and an equivalent lateral force F is constructed during online control. y ';With equivalent lateral force F y 'Replace the original out-of-plane lateral force F y With F x Together as orthogonal criteria; (equivalent transverse force F) y '=F y +μ·F z μ is the coefficient of kinetic friction, F z (This refers to the normal contact force of the probe).

[0081] In practice, when the state variable S=1 (movement condition), the dynamic friction coefficient calibration stage is initiated first. A relatively flat area of ​​the abdominal surface is selected (avoiding areas with protrusions, wrinkles, or obvious organ projections). Initial alignment is achieved through manual assistance or visual prior alignment using a fixed-point self-orthogonal closed loop, ensuring the probe's initial position is approximately perpendicular to the flat surface, thus preventing initial posture deviations from affecting friction data acquisition. Subsequently, the robotic arm is controlled to perform a short-range uniform speed scan only along the Y-axis of the end-effector coordinate system. During the scan, Z-axis force feedback is used to strictly maintain the contact force between the probe and the surface within the target range, preventing friction calculation deviations due to contact force fluctuations.

[0082] During the scan, the out-of-plane lateral force F of each control cycle is recorded simultaneously. y In-plane lateral force F x and normal contact force F z The data is processed in real time, and the collected data is filtered: if the F value of a certain period is... z Deviation from the target range, or F x F y If a transient abnormal peak occurs (determined to be an interference signal), the data for that period is discarded, and only valid data with stable contact force and no interference are retained. After calibration, the F value in the valid data is... y Summing after taking the absolute value, for F z The estimated coefficient of dynamic friction for a single test is obtained by directly summing the two values ​​and calculating the ratio. To improve accuracy, this calibration process needs to be repeated multiple times. After removing outliers from each estimate, the average value is taken as the final coefficient of dynamic friction, μ. Simultaneously, μ needs to be judged for reasonableness. If μ exceeds the normal friction coefficient range under the action of the coupling agent between the body surface and the probe, the calibration is considered a failure, and the flatness of the scanning area and the application of the coupling agent need to be checked before re-calibrating.

[0083] During the online control phase, when the robotic arm is in the moving scanning state (S=1), the controller reads the current F value collected by the six-dimensional force sensor in real time. y With F z Data (since the force sensor and end-effector coordinate system are calibrated in the same direction, the data can be used directly). Calculate the equivalent lateral force F based on the calibrated μ. y ', that is, through F y Add μ and F z The product of the two forces will affect the dynamic friction force on F during the movement. y The systematic disturbances were explicitly canceled out. Then, F... y 'Replace the original F y , and the original F that was not affected by friction x Using the same self-orthogonality criterion: adopting the "threshold band + continuous out-of-bounds count" determination mechanism, monitoring F respectively. y 'with F xWhether it is within a preset threshold range; if any axis continuously exceeds the limit and reaches the set threshold, the corresponding attitude fine-tuning (F) is triggered. y 'When crossing the boundary, perform pitch correction around the X-axis, F x When crossing the boundary, perform yaw correction around the Y-axis to ensure that the lateral force is close to zero during movement. The validity of the "normal alignment" criterion.

[0084] In addition, the μ value needs to be dynamically managed according to the scanning conditions: if the scanning site changes (e.g., from the liver area to the kidney area) or the surface conditions change (e.g., the coupling agent dries or needs to be reapplied), the above calibration procedure needs to be repeated to update the μ value; if F is detected during the scanning process... y If the correction still results in frequent boundary crossings without significant attitude deviation, the μ value is deemed invalid. The scanning process must be paused, μ recalibrated, and then resumed to ensure the equivalent lateral force F. y It can always accurately reflect attitude deviations and avoid the decrease in orthogonal control accuracy caused by friction compensation failure.

[0085] The method for controlling the autonomous orthogonal scanning of an abdominal ultrasound robot includes the following steps:

[0086] S600: Dynamic updates and iterative optimization, updating visual priors to adapt to slow deformation caused by breathing.

[0087] In practice, when dynamically updating the visual prior, it is necessary to first synchronize with the patient's respiratory rhythm: by monitoring the respiratory signal in real time (which can be combined with external respiratory sensors or indirectly judged by extracting the deformation pattern of the body surface from continuous point cloud data), the point cloud sampling of the RGB-D camera is triggered in the relatively stable stage of the respiratory cycle (such as the end of expiration or the end of inspiration), avoiding sampling when breathing causes drastic deformation of the body surface, reducing point cloud distortion caused by instantaneous deformation, and ensuring that the collected point cloud can accurately reflect the stable shape of the abdominal body surface at that moment.

[0088] After sampling, the newly acquired local point cloud undergoes the same preprocessing and quality assessment process as the initial visual prior: first, interference points such as coupling agent reflections and hair are filtered out, and then the surface normals of key points and their neighborhoods are fitted; if the deviation between the fitted normal and the previous effective normal is within a preset range, and the deviation between the point cloud and the fitted plane meets the accuracy requirements, then the normal is updated as the new target normal reference; if the deviation is too large or the point cloud quality is substandard (such as point cloud overlap or missing points due to breathing), then the sampling result of this time is discarded, the most recent effective target normal reference is used, and the sampling frequency of subsequent visual priors is appropriately reduced to avoid invalid updates increasing the computational burden.

[0089] Meanwhile, the weight of the visual prior is dynamically adjusted according to the breathing amplitude: when the breathing amplitude is small (the body surface deformation is gentle), the weight of the visual prior in coarse alignment is appropriately increased so that the probe posture can better match the changes in the normal direction of the body surface; when the breathing amplitude is large (the body surface deformation is severe), the weight of the visual prior is reduced to reduce its interference with the high-frequency force control closed loop. At this time, more reliance is placed on force feedback to compensate for the posture deviation caused by deformation in real time, so as to ensure the stability of the self-orthogonal control.

[0090] In addition, the updating of visual prior needs to be coordinated with the main loop of force-position hybrid control: after each update, if the robot is currently in a non-moving scanning state (S=0), a quick coarse alignment is performed based on the new target normal reference, and then the return force feedback is seamlessly switched for fine adjustment; if it is in a moving scanning state (S=1), the coarse alignment is not performed for the time being, and the new normal reference is stored as the basis for subsequent coarse alignment of key points, so as to avoid the scanning path deviation or contact force fluctuation caused by adjusting the posture during movement, and finally realize the dynamic coordination of visual prior and force feedback, continuously adapting to the slow deformation of the body surface caused by breathing.

[0091] Dynamic updates and iterative optimizations include: periodically re-estimating the μ value based on the scanning site, surface conditions, and coupling agent status; dynamically adjusting visual prior weights; and continuously optimizing control parameters to ensure stability under complex surfaces and respiratory movements.

[0092] In practice, during dynamic updates and iterative optimization, the dynamic friction coefficient μ value is re-evaluated periodically: when the scanning site is switched from one organ region to another (such as from the liver region to the spleen region), or when the surface conditions change (such as from smooth skin to a wrinkled area, or when the fat thickness changes significantly), or when the coupling agent dries out and needs to be reapplied, the μ value re-evaluation process is automatically triggered.

[0093] Re-evaluation does not require selecting a separate flat area; instead, it selects a relatively flat segment within the current scan path, using the "along" approach. Y The calibration logic of "short-range scan - record force data - statistical calculation" only shortens the scan length to improve efficiency; after calculating the new μ value, it is compared with the currently used μ value. If the deviation exceeds a reasonable range, the validity of the new μ value is first verified by a small-scale trial scan (observing the equivalent lateral force F). y (Whether it can stably reflect attitude deviation) is verified before updating to the control algorithm to avoid blind replacement that could lead to distortion of the orthogonality criterion.

[0094] Secondly, there is the dynamic adjustment of visual prior weights: In addition to adjusting breathing amplitude, the weights are also optimized in real time based on the quality of the point cloud collected by the RGB-D camera. If the point cloud sampled multiple times still has a large number of interference points after denoising, or the fitted surface normal deviates significantly from the actual normal reflected by the force feedback (e.g., force feedback frequently triggers posture fine-tuning but the visual prior does not indicate the deviation), the weight of the visual prior in coarse alignment is automatically reduced to reduce its impact on posture adjustment. If the point cloud quality is stable, the fitted normal and the force feedback results are highly consistent, and the breathing amplitude is small, the weight is appropriately increased to make the coarse alignment more closely match the actual shape of the body surface, further reducing the subsequent adjustment amount of the force control closed loop.

[0095] At the same time, after the weight adjustment, the oscillation of the high-frequency force control closed loop will be monitored simultaneously. If a small oscillation occurs, the weight will be adjusted slightly to ensure the stable coordination between visual prior and force feedback.

[0096] Finally, continuous optimization of control parameters is implemented: parameters such as the target contact force range, lateral force threshold, and out-of-bounds count threshold are dynamically fine-tuned based on real-time scanning feedback. For example, when scanning deep abdominal organs (such as the kidneys), if it is found that maintaining the original contact force range makes it difficult to obtain clear images, the range will be adjusted appropriately to ensure acoustic-energy coupling. If frequent micro-deformations on the body surface cause the lateral force to briefly exceed the limit, triggering unnecessary posture fine-tuning, the lateral force threshold will be appropriately relaxed or the out-of-bounds count threshold will be extended. If it is found that posture fine-tuning still easily deviates from the orthogonal state in complex curved areas (such as the abdominal protrusion), the integral limit associated with the angle step size of a single posture fine-tuning will be slightly adjusted to make the correction more in line with the surface changes. All parameter adjustments follow the logic of "small-scale trial - monitoring feedback - confirmation of effectiveness". After each adjustment, the stability of the force signal and the force control performance related to imaging (such as the amplitude of contact force fluctuation and the speed of lateral force return to the threshold) are observed for at least one complete respiratory cycle. Only after confirming that it can improve the control stability under complex curved surfaces and respiratory movements are the adjusted parameters fixed to avoid control fluctuations caused by frequent parameter changes.

[0097] The implementation of this application is described in detail below with reference to the accompanying drawings. This invention proposes a self-orthogonal scanning control method for an abdominal ultrasound robot, aiming to maintain a constant contact force while ensuring that the normal of the ultrasound probe is consistent with the normal of the body surface in real time (self-orthogonality). This method is compatible with both fixed-point pressure imaging and path-moving scanning scenarios, solving the problems of existing force-position hybrid control being easily affected by friction during movement and difficulty in maintaining stable verticality under complex curved surfaces and respiratory movements, and improving imaging stability and repeatability without sacrificing scanning efficiency. Compared with the prior art, this invention achieves self-orthogonal control through a collaborative mechanism of "visual prior—force feedback—friction compensation—attitude fine-tuning".

[0098] The implementation process of a self-orthogonal scanning control method for abdominal ultrasound robots includes: 1. System and coordinate conventions, refer to Appendix Figure 1 The system consists of a six-degree-of-freedom robotic arm 1 and a six-dimensional force / torque sensor 3 (sampling frequency = 500Hz) mounted between a flange 2 and a probe clamp 5. z ), ultrasonic probe 6, RGB-D camera 4 mounted on probe clamp 5, and real-time controller (controlling cycle frequency = 200Hz) z Composed of ( ). Using the end effector coordinate system of the robotic arm as a unified reference: Z The axis points from the probe normal to the body surface normal direction; Y The axis is defined as the main direction of movement outside the probe sector ("out-of-plane"). X The axis is in-plane ("in-plane"). The force sensor coordinates are consistent with the end effector coordinates, ensuring that the readings can be directly used for control law design. The hardware selection and frequency configuration meet the real-time limits of common medical robot platforms, demonstrating engineering feasibility.

[0099] 2. Introduction and Mechanism of Visual Prior: Considering the curvature of abdominal soft tissue and its slow deformation during respiration, relying solely on force feedback to achieve self-orthogonality often requires a long convergence time. Therefore, this invention first samples local point clouds at a low frequency (2s / sample) using an RGB-D camera, while simultaneously setting a key point at regular intervals (2cm) along the scanning path. The surface normal is fitted to the key points and their neighborhoods along the scanning path to obtain a "target normal reference." When the scan reaches a key point, the controller first performs coarse alignment of the end effector posture based on this reference, and then switches to a high-frequency force feedback closed loop for fine-tuning.

[0100] To ensure feasibility, the visual prior only serves as an "initial alignment" and "trend guide," rather than a real-time rigid constraint: its update frequency is limited by the reliable frame rate of RGB-D, point cloud noise, and reconstruction errors; the controller assigns limited weights to the visual prior to avoid misleading information caused by transient breathing motion or coupler specular reflection. This division of labor allows the system to strike a balance between computational load and robustness, ensuring that the latency of the visual loop does not compromise the stability of the high-frequency force control closed loop. (Reference) Figure 2 , Figure 3 process.

[0101] 3. Force-position hybrid control main loop and state discrimination, refer to Figure 5Within the control cycle, the system reads the end-effector pose and six-axis force data from the previous cycle. First, it fine-tunes the end-effector Z-direction displacement using the Z-axis force as feedback, maintaining the contact force within a set range (e.g., 10N ± 1N). Once the force meets the range, it performs small-step displacement along the Y-axis to complete the scanning path tracking. If the force does not meet the range, it pauses the Y-direction displacement and only performs Z-direction compensation. The controller maintains a state variable S: S=0 when only Z-direction adjustment is performed (no horizontal movement), and S=1 when Y-direction scanning displacement exists. This state is used for subsequent selection of the self-orthogonality criterion and friction compensation, thus naturally distinguishing between fixed-point and moving conditions within a unified cycle, avoiding transient instability caused by mode switching.

[0102] 4. Fixed-point self-orthogonal control, see reference. Figure 4 In a fixed-point pressing scenario, there is no significant horizontal relative motion, and the lateral friction force can be ignored. The lateral reading of the force sensor mainly reflects the misalignment between the probe and the body surface normal. Therefore, "near-zero lateral force" can be used as the error criterion for the self-orthogonal closed loop: the controller continuously monitors F x With F y Whether it falls within a settable threshold range (e.g., ±0.5N). To suppress short-term fluctuations caused by noise and micro-deformation of the skin, an out-of-bounds counting and delayed triggering mechanism is introduced: a small-angle attitude fine-tuning is triggered only when a certain axis continuously exceeds the threshold (20 control cycles). The fine-tuning adopts incremental Euler angle correction of the current attitude: when F y When exceeding the limit, apply pitch (around the X-axis) correction, prioritizing the elimination of normal deviations in out-of-plane directions; when F x When the boundary is exceeded, a yaw (around the Y-axis) correction is applied to eliminate the deviation in the in-plane direction; the amount of each correction is limited to a small angle (3° / time) and is subject to integral limiting to prevent oscillation.

[0103] This design meets practical feasibility requirements: sensor data is directly mapped to the attitude correction channel without the need for complex modeling; delayed triggering balances response speed and disturbance rejection; small angle increments avoid Euler angle coupling amplification and singular attitude problems, and it is easy to implement on standard industrial controllers.

[0104] 5. Self-orthogonal control during path scanning, refer to Figure 6 When scanning along the Y-axis, the lateral reading of the force sensor is significantly affected by kinetic friction. If "lateral force ≈ 0" is still used as the orthogonality criterion, a systematic bias will occur. This invention achieves feasible online correction through the approach of "calibration first, compensation later".

[0105] During the calibration phase, the probe is kept approximately perpendicular within a relatively flat area, and short-range scans are performed only along the Y direction. The F value for each control cycle is recorded. y (Transverse friction component) and F z (Normal pressure), summing the data over the entire segment separately, to As an estimate of the coefficient of kinetic friction, the result is averaged after multiple repetitions. This statistic is convergent under noise and instantaneous fluctuations, does not depend on complex surface models, and is easy to implement in engineering.

[0106] During online control, if S=1, then the corrected equivalent lateral force is used. ;

[0107] Replace the original F y Enter the "threshold-count-fine-tuning" closed loop (still using the pitch channel to eliminate out-of-plane bias), while the in-plane direction continues to use F... x As a criterion (yaw channel). This compensation explicitly cancels out the systematic frictional bias caused by movement, making the lateral force "near zero". The "normal alignment" criterion is re-established in mobile scenarios, thus ensuring that the self-orthogonal closed loop remains effective during scanning. In terms of parameters, μ can be re-estimated and updated according to location, surface conditions, and coupling agent state, ensuring robustness under long-term operation.

[0108] 6. Key parameters: Force sensor sampling frequency (500Hz), control cycle frequency (200Hz), target contact force (10N), force tolerance (±1N), lateral force threshold (±0.5N), out-of-bounds counting threshold (20 cycles), fine-tuning angle step size (3° / time), visual sampling cycle (2s / time), key point interval for visual coarse alignment (2cm), and μ re-estimation strategy are all stored as configurable parameters and adjusted according to working conditions.

[0109] It is understandable that, compared to most existing ultrasonic robot scanning solutions that rely solely on force control or trajectory following, this invention takes "probe-surface normal alignment (self-orthogonality)" as a control objective: by using visual normal vector priors + high-frequency force feedback closed loops + friction compensation in moving scenarios, it continuously maintains near-normal incidence in both fixed-point and continuous scanning conditions. This mechanism directly brings benefits at the imaging physics level: improving acoustic coupling and echo signal strength, reducing specular reflection and interface artifacts caused by oblique incidence, reducing measurement deviations caused by layer deviation and "shortening effect," and significantly improving the contrast and coherence of abdominal soft tissue (liver / gallbladder / spleen / kidney, etc.) boundaries; at the same time, in modes sensitive to incident and contact stability, such as elastography / shear wave imaging, it is understandable that it reduces systematic errors caused by angle and pressure fluctuations. For clinical workflows, self-orthogonal control reduces operator dependence and rescanning rate, and improves the repeatability of standard sections and consistency across time points and operators. For robot control, self-orthogonality makes the mapping of "lateral force ≈ posture error" more observable and controllable. Combined with friction compensation, it restores the correctness of the criteria during moving scans, thereby obtaining more stable image quality and more predictable closed-loop behavior without sacrificing efficiency.

[0110] It is understandable that the synergy between visual priors and force feedback shortens convergence time without compromising stability. Placing the RGB-D surface normal as a low-frequency prior before closed-loop control allows for "coarse alignment" before entering a certain scan segment or fixed-point pressing, reducing tracking errors and attitude oscillations in the subsequent mechanical closed loop. Injecting the prior with low frequency / limited weights improves convergence speed while avoiding misleading effects caused by visual noise, breathing deformation, or specular reflection, without affecting the stability of the 200Hz-level high-frequency force control closed loop.

[0111] The friction compensation in the moving scan is understandable, restoring the correctness of the "self-orthogonal criterion". In scans moving along the Y-axis, the lateral readings are significantly affected by kinetic friction. A design of calibration followed by compensation is used: using the statistic Σ|F... y | / ΣF z Estimate the kinetic friction coefficient μ and construct the equivalent lateral force F online. y '=F y +μ·F z This makes the lateral force approximately zero. The criterion for "normal alignment" holds true again in mobile scenarios. This strategy unifies the self-orthogonality criterion and control law for fixed-point and mobile scanning, reduces scene-specific branch logic, and simplifies engineering implementation.

[0112] The robust criterion design, resistant to noise and micro-deformation, is understandable. It employs a delay triggering mechanism of "threshold band + continuous out-of-bounds counting": only when F... x or F y (or F) y The attitude fine-tuning is only triggered when the threshold is continuously exceeded. This statistical criterion effectively suppresses short-term pulses caused by force sensor noise, local skin micro-deformation, and coupling agent fluctuations, thereby avoiding frequent ineffective adjustments and improving closed-loop steady-state quality.

[0113] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0114] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for controlling the self-orthogonal scanning of an abdominal ultrasound robot, characterized in that, include: System initialization and parameter configuration: Establish the coordinate system of the robotic arm end effector, define the Z-axis as the direction from the probe normal to the body surface, the Y-axis as the main movement direction of the probe, and the X-axis as the direction within the sector; Visual prior acquisition and coarse alignment are performed. The camera samples local point clouds and sets key points at fixed intervals in the scanning path. The surface normal is fitted at the key points and their neighborhoods as a reference for the target normal. The force-position hybrid control main cycle starts, reads the end pose and six-axis force data, and fine-tunes the end Z-direction displacement based on Z-axis force feedback to keep the contact force within the target range. When the contact force meets the requirements, it performs small step displacement along the Y-axis and sets the state variable S=1. If the contact force does not meet the requirements, pause the Y-axis displacement and only perform Z-axis compensation, and set the state variable S=0; When the state variable S=0, monitor the lateral force F in the in-plane direction. x out-of-plane lateral force F y When any axis meets the condition, attitude fine-tuning is triggered; Out-of-plane lateral force F y When the plane crosses the boundary, a pitch correction around the X-axis is applied to eliminate out-of-plane deviation; the in-plane lateral force F x When the boundary is exceeded, a yaw correction around the Y-axis is applied to eliminate in-plane deviation, and each correction is subject to integral limiting. When the state variable S=1, during the calibration phase, a short-range scan is performed along the Y direction in a flat area, and the probe's normal contact force F is recorded. z out-of-plane lateral force F y The coefficient of kinetic friction is calculated to construct an equivalent lateral force F during the force feedback orthogonal phase. y ';With equivalent lateral force F y 'Replace the original out-of-plane lateral force F y With F x They serve as orthogonal criteria.

2. The method for controlling an abdominal ultrasound robot to perform self-orthogonal scanning according to claim 1, characterized in that, The key points are set at fixed intervals, specifically every 2cm.

3. The method for controlling an abdominal ultrasound robot to perform self-orthogonal scanning according to claim 1, characterized in that, The parameter configuration includes: configuring a six-degree-of-freedom robotic arm, a six-dimensional force / torque sensor, an RGB-D camera and a real-time controller, with the force sensor coordinates consistent with the end-effector coordinates; setting the target contact force range, lateral force threshold, out-of-bounds counting threshold, and attitude fine-tuning angle.

4. The method for controlling an abdominal ultrasound robot to perform self-orthogonal scanning according to claim 1, characterized in that, The process of acquiring visual priors and coarse alignment includes: when the robot reaches the key point, coarse alignment of the end-effector posture is performed based on the target normal reference. After completion, the robot switches to high-frequency force feedback closed-loop fine adjustment. The visual priors play a role in initial alignment and trend guidance.

5. The method for controlling the self-orthogonal scanning of an abdominal ultrasound robot according to claim 1, characterized in that, The attitude fine-tuning is triggered when any axis meets the following conditions: attitude fine-tuning is triggered when any axis exceeds the ±0.5N threshold for 20 consecutive control cycles.

6. The method for controlling an abdominal ultrasound robot to perform self-orthogonal scanning according to claim 1, characterized in that, Equivalent lateral force F y '=F y +μ·F z μ is the coefficient of kinetic friction, F z This represents the normal contact force of the probe.

7. The method for controlling an abdominal ultrasound robot to perform self-orthogonal scanning according to claim 1, characterized in that, It also includes dynamic updates and iterative optimization: the μ value is re-estimated periodically based on the scanning site, body surface conditions and coupling agent status, the visual prior weights are dynamically adjusted, and the control parameters are continuously optimized.

8. The method for controlling the self-orthogonal scanning of an abdominal ultrasound robot according to claim 1, characterized in that, The camera samples local point clouds, sets key points at fixed intervals along the scanning path, and fits the surface normal vector to the key points and their neighborhoods as the target normal vector reference. This includes: preprocessing the RGB-D camera before sampling local point clouds to avoid interference factors and ensure that the point clouds accurately reflect the geometric shape of the abdominal surface; the sampling process is synchronized with the robotic arm's scanning path planning; the robotic arm initiates the first sampling before entering the area to be scanned, and subsequently continues sampling at preset intervals to dynamically capture changes in the surface morphology caused by breathing; when setting key points along the scanning path, the distribution is combined with the planned distribution of the abdominal scanning target area to ensure full coverage; when fitting the surface normal vector to the key points and their neighborhoods, the neighborhood point cloud is first extracted and denoised, and then the point cloud is fitted into a local plane using a fitting algorithm to calculate the normal vector as the target normal vector reference; after fitting, the quality of the target normal vector reference is evaluated, and if the deviation exceeds the range, the normal vector reference is discarded.

9. The method for controlling an abdominal ultrasound robot to perform self-orthogonal scanning according to claim 1, characterized in that, The out-of-plane lateral force F y When the plane crosses the boundary, a pitch correction around the X-axis is applied to eliminate out-of-plane deviation; the in-plane lateral force F x When the boundary is exceeded, a yaw correction around the Y-axis is applied to eliminate in-plane deviations. Each correction includes integral limiting, including: out-of-plane lateral force F was detected y When crossing the boundary, according to the out-of-plane lateral force F y Positive and negative values ​​indicate the direction of the probe's outward deflection and the outward lateral force F. y If the positive boundary is exceeded, the robotic arm's end effector will adjust its pitch around the X-axis in the opposite direction, with an out-of-plane lateral force F. y If the negative direction exceeds the limit, adjust the pitch in the positive direction. Prioritize precise adjustment with direction correspondence to eliminate the normal deviation outside the fan surface and ensure that the adjustment matches the direction of the deviation. In-plane lateral force F was detected x When crossing the boundary, the in-plane lateral force F x Positive and negative values ​​indicate the in-plane skewness of the probe and the in-plane lateral force F. x If the positive boundary is exceeded, the robotic arm's end effector will adjust its yaw around the Y-axis in the opposite direction, with an in-plane lateral force F. x If the negative yaw exceeds the limit, the positive yaw adjustment will specifically eliminate the normal deviation within the sector to ensure that the probe scanning surface is aligned with the normal of the body surface; each attitude fine-tuning adopts incremental adjustment, and an integral limit is set for the cumulative correction angle.

Images