Planting operation robot control method based on agent fusion multi-sensor

Abstract

The application provides a planting operation robot control method based on agent fusion of multiple sensors, relates to the technical field of medical robots, and comprises the following steps: feature extraction and abnormality identification are performed on multiple sensor data by a perception layer agent, space-time registration is performed, a consistency deviation measurement value is calculated, and fused perception data are generated; a decision layer agent generates position and force control layer instructions based on the fused data, and establishes switching conditions and a cooperative constraint relationship; and the robot executes an operation in different control modes according to the instructions and the constraints. The method improves the perception accuracy and control stability, and reduces the operation risk.

Description

Technical Field

[0001] This invention relates to the field of medical robot technology, and in particular to a control method for an implantation surgery robot based on intelligent agent fusion of multiple sensors. Background Technology

[0002] Dental implant surgery is a complex and precise oral surgical procedure that traditionally relies on the surgeon's experience and manual operation, making the quality and outcome susceptible to human error. With the development of medical robotics technology, dental implant surgery robots are gradually entering clinical applications, assisting or replacing surgeons in key steps such as implant placement, improving surgical precision and safety. These robots are typically equipped with various sensors, including optical navigation systems, force sensors, and tactile sensors, to achieve real-time perception and precise control of the oral environment.

[0003] Data from a single sensor has inherent limitations and cannot fully perceive the complex oral environment. For example, optical navigation systems are affected by line-of-sight obstruction, force sensors are susceptible to external interference, and information from a single sensor is insufficient to cope with various uncertainties during surgery, which can easily lead to robot decision-making errors or inaccurate control.

[0004] Current technologies lack effective multi-sensor fusion mechanisms, and data from each sensor is often processed independently, failing to establish cross-modal constraints. When conflicting information from different sensors occurs, the system struggles to determine which information is more reliable, making it impossible to effectively utilize complementary information to improve perception accuracy, thus affecting surgical safety.

[0005] Existing robot control strategies typically employ a single control mode, primarily relying on either position control or force control, lacking intelligent switching and coordination mechanisms between the two. This makes it difficult for robots to adaptively adjust their control strategies when facing different surgical stages and tissue characteristics, failing to simultaneously guarantee positional accuracy and force interaction safety, thus limiting the application scope and performance of surgical robots. Summary of the Invention

[0006] This invention provides a control method for an implantation surgery robot based on intelligent agent fusion of multiple sensors, which can solve the problems in the prior art.

[0007] A first aspect of the present invention provides a control method for an implantation surgery robot based on agent-fused multi-sensor systems, comprising: Acquire real-time sensing data from multiple heterogeneous sensors during dental implant surgery; The perception layer agent extracts features and identifies anomalies from the real-time perception data, and the decision layer agent generates path planning and force control strategies based on the output of the perception layer agent. Spatial-temporal registration is performed between heterogeneous sensors to extract multimodal representations of the same anatomical feature from the real-time sensing data and calculate a consistency deviation metric. Based on the consistency deviation metric and the confidence level of each sensor, a correction mechanism is used to compensate the real-time sensing data by constructing cross-modal constraints to generate fused sensing data. Based on the fused perception data, the decision-making agent combines biomechanical constraints and collision safety constraints to generate position control layer commands and force control layer commands for the dental implant robot, and establishes switching conditions and cooperative constraint relationships between the position control layer and the force control layer. Based on the switching conditions and the cooperative constraint relationship, the dental implant robot performs dental implant surgery in the position-dominant control mode, with position control layer commands as the main force and force control layer commands as the safety constraint boundary, and in the force-dominant control mode, force control layer commands as the main force and position control layer commands as the trajectory constraint boundary.

[0008] The real-time perceived data is subjected to feature extraction and anomaly identification by the perception layer agent, and the decision layer agent performs path planning and force control strategy generation based on the output of the perception layer agent, including: The perception layer agent extracts position features, density distribution features, and mechanical response features from the real-time perception data, and evaluates the degree of abnormality by comparing with the preset normal range boundary, generating a perception layer output result containing feature vectors and abnormality identifiers. The decision-making agent receives the output of the perception layer and, based on the relative positional relationship between the spatial coordinates of the implantation site and the surrounding anatomical structures in the location features, generates a path planning result by constructing a spatial curve path from the current instrument position to the target position. Based on the anomaly markers, the anatomical risk areas that need to be avoided at the current oral implantation site are determined, and the path planning results are corrected by avoiding the anatomical risk areas based on the location characteristics. The decision-making agent establishes a mapping relationship between bone density and implantation torque requirements based on the density distribution characteristics. It calculates torque control target values ​​corresponding to different bone density regions based on the mapping relationship to generate a torque control strategy. Based on the real-time resistance feedback information in the mechanical response characteristics and the bone hardness information in the density distribution characteristics, it calculates feed rate control target values ​​to generate a feed rate control strategy. The force control strategy is obtained by combining the torque control strategy and the feed rate control strategy.

[0009] Spatial-temporal registration is performed between heterogeneous sensors, and multimodal representations describing the same anatomical feature are extracted from the real-time sensing data, and a consistency deviation metric is calculated, including: The timestamp information and spatial coordinate system information of the real-time sensing data are obtained. The timestamp information is time-aligned by establishing a unified time reference. The spatial coordinate system information is transformed to a unified surgical spatial coordinate system by constructing a spatial coordinate transformation matrix, thereby generating a registered multi-sensor dataset. Different modal data describing the same anatomical feature are identified from the registered multi-sensor dataset. Spatial geometric morphology representation, mechanical property representation and surface texture representation are extracted for the same anatomical feature of oral implant site to form a multimodal representation set describing the same anatomical feature. The multimodal representation set is transformed into a unified feature representation space. By constructing a cross-modal feature mapping relationship, the spatial geometric morphology representation, the mechanical property representation, and the surface texture representation are aligned in the unified feature representation space. The spatial position deviation, physical property deviation, and surface morphology deviation between the aligned different modal representations are calculated and combined to generate the consistency deviation metric.

[0010] Based on the consistency deviation metric and the confidence level of each sensor, a correction mechanism is used to compensate for the real-time sensing data by constructing cross-modal constraints, generating fused sensing data including: Based on the historical data accuracy, current operating stability, and measurement environment interference of the heterogeneous sensor, the confidence value of the heterogeneous sensor is calculated. When the consistency deviation metric exceeds the preset consistency threshold, a correction mechanism is triggered to identify inconsistent modal representations in the multimodal representation set and extract the corresponding confidence values. Based on the confidence values, the inconsistent modal representations are divided into reference modal representations and modal representations to be compensated. In a unified feature representation space, with the goal of minimizing the deviation between the corresponding feature components of the modal representation to be compensated and the reference modal representation, the compensation correction amount of the modal representation to be compensated is calculated. The compensation correction amount is superimposed on the real-time sensing data corresponding to the modal representation to be compensated, and the compensated sensor data is fused with the uncorrected real-time sensing data to generate the fused sensing data.

[0011] Based on the fused perception data, the decision-making layer agent generates position control layer commands and force control layer commands for the dental implant robot by combining biomechanical constraints and collision safety constraints, including: Bone density distribution information, bone thickness information, and surrounding anatomical structure location information of oral tissues are extracted from the fused perception data. Biomechanical bearing capacity parameters of oral tissues are established. Biomechanical constraints are obtained by limiting the operating torque and torque of the implantation robot to not exceed the biomechanical bearing capacity parameters. The current location information of the implantation device and the surgical restricted area are extracted from the fused perception data. Based on the minimum safe distance between the implantation device and the surgical restricted area, collision safety constraints are obtained. The decision-making agent verifies the collision safety constraints of the path planning results. When the path planning results do not meet the collision safety constraints, the spatial curve path of the path planning results is adjusted based on the minimum safe distance and converted into a position control layer instruction containing a position coordinate sequence and an attitude angle sequence. The decision-making agent performs biomechanical constraint verification on the force control strategy. When the torque control strategy or feed rate control strategy does not meet the biomechanical constraints, it performs amplitude limiting processing on the torque control target value and feed rate control target value based on the biomechanical bearing capacity parameter, and converts them into force control layer instructions containing force control parameters and torque control parameters.

[0012] The switching conditions and collaborative constraints between the position control layer and the force control layer are established, including: Real-time position deviation information and real-time contact force information of the implantation device are extracted from the fused sensing data. An activation threshold for the position-dominated control mode is set based on the real-time position deviation information, and an activation threshold for the force-dominated control mode is set based on the real-time contact force information. Switching conditions between the position control layer and the force control layer are established. The switching conditions stipulate that when the real-time position deviation information exceeds the activation threshold of the position-dominated control mode, the system switches to the position-dominated control mode, and when the real-time contact force information exceeds the activation threshold of the force-dominated control mode, the system switches to the force-dominated control mode. The position control motion speed is extracted from the position control layer command, and the force control target force value is extracted from the force control layer command. A coupling constraint rule is established between the position control motion speed and the force control target force value by constructing a cooperative constraint relationship. The coupling constraint rule stipulates that in the position-dominated control mode, the position control motion speed is constrained by the safety boundary of the force control target force value, and in the force-dominated control mode, the force control target force value is constrained by the trajectory boundary of the position control motion speed.

[0013] The oral implant robot, in position-dominated control mode, is dominated by position control layer commands and uses force control layer commands as safety constraint boundaries. In force-dominated control mode, it is dominated by force control layer commands and uses position control layer commands as trajectory constraint boundaries, including: The dental implant robot determines the control mode to be used in the current stage of the dental implant surgery based on the switching conditions. In the position-dominated control mode, the position control target is extracted from the position control layer command as the dominant control target, and the upper limit value of the force control target force value is extracted from the force control layer command as the safety constraint boundary. This limits the movement trajectory of the planting robot to follow the position control target and ensures that the applied force does not exceed the safety constraint boundary. When the safety constraint boundary is exceeded, the position control movement speed is reduced until it falls back into the safety constraint boundary. In the force-dominated control mode, the force control target force value is extracted from the force control layer command as the dominant control target, and the deviation range of the position control target is extracted from the position control layer command as the trajectory constraint boundary. This limits the force applied by the planting robot to follow the force control target force value and the position deviation to remain within the trajectory constraint boundary. When the position deviation exceeds the trajectory constraint boundary, the force control direction is adjusted until the position deviation returns to within the trajectory constraint boundary.

[0014] A second aspect of the present invention provides a control system for an implantation surgery robot based on agent-fused multi-sensor systems, comprising: The first unit is used to acquire real-time sensing data from multiple heterogeneous sensors during oral implant surgery; The second unit is used to extract features and identify anomalies from the real-time sensing data through the perception layer agent, and the decision layer agent performs path planning and force control strategy generation based on the output of the perception layer agent. The third unit is used to perform spatial-temporal registration between heterogeneous sensors, extract multimodal representations describing the same anatomical feature from the real-time sensing data and calculate a consistency deviation metric; based on the consistency deviation metric and the confidence level of each sensor, a correction mechanism is used to compensate the real-time sensing data by constructing cross-modal constraints to generate fused sensing data. The fourth unit is used to generate position control layer instructions and force control layer instructions for the oral implant robot based on the fused perception data, combining biomechanical constraints and collision safety constraints, and to establish switching conditions and cooperative constraint relationships between the position control layer and the force control layer. The fifth unit is used to perform oral implant surgery based on the switching conditions and the cooperative constraint relationship. In the position-dominant control mode, the oral implant robot takes the position control layer command as the main force and the force control layer command as the safety constraint boundary. In the force-dominant control mode, the force control layer command as the main force and the position control layer command as the trajectory constraint boundary.

[0015] A third aspect of the embodiments of the present invention, An electronic device is provided, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0016] Fourth aspect of the embodiments of the present invention, A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0017] The beneficial effects of this application are as follows: This invention adopts a two-layer intelligent agent architecture with a perception layer and a decision layer to extract features and identify anomalies from multi-sensor data, thereby achieving comprehensive perception and accurate judgment of the surgical environment. It overcomes the perception limitations of a single sensor in a complex oral environment and improves the system's accuracy in recognizing oral anatomical features.

[0018] This invention innovatively establishes a space-time registration and consistency deviation measurement mechanism. By fusing and compensating heterogeneous sensor data through cross-modal constraints, it effectively solves the problem of inconsistency in multi-sensor data, enhances the robustness of the system under sensor partial failure or interference, and improves the reliability of sensing data.

[0019] This invention designs a dual-mode collaborative mechanism of position control and force control. According to the control strategy of automatic switching according to the surgical stage, it ensures accurate trajectory tracking in position-dominant mode and ensures safe contact force in force-dominant mode. This achieves precise positioning and compliant operation during the surgical process, while meeting biomechanical and safety constraints, and significantly reducing surgical risks. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the control method for an implantation surgery robot based on intelligent agent fusion of multiple sensors, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the process of intelligent agent perception, decision-making, and force control strategy generation. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0023] Figure 1 This is a flowchart illustrating the control method for an implantation surgery robot based on agent-fused multi-sensor technology according to an embodiment of the present invention. Figure 1 As shown, the method includes: Acquire real-time sensing data from multiple heterogeneous sensors during dental implant surgery; The perception layer agent extracts features and identifies anomalies from the real-time perception data, and the decision layer agent generates path planning and force control strategies based on the output of the perception layer agent. Spatial-temporal registration is performed between heterogeneous sensors to extract multimodal representations of the same anatomical feature from the real-time sensing data and calculate a consistency deviation metric. Based on the consistency deviation metric and the confidence level of each sensor, a correction mechanism is used to compensate the real-time sensing data by constructing cross-modal constraints to generate fused sensing data. Based on the fused perception data, the decision-making agent combines biomechanical constraints and collision safety constraints to generate position control layer commands and force control layer commands for the dental implant robot, and establishes switching conditions and cooperative constraint relationships between the position control layer and the force control layer. Based on the switching conditions and the cooperative constraint relationship, the dental implant robot performs dental implant surgery in the position-dominant control mode, with position control layer commands as the main force and force control layer commands as the safety constraint boundary, and in the force-dominant control mode, force control layer commands as the main force and position control layer commands as the trajectory constraint boundary.

[0024] In one optional implementation, the real-time sensing data is subjected to feature extraction and anomaly identification by a perception layer agent, and the decision layer agent performs path planning and force control strategy generation based on the output of the perception layer agent, including: The perception layer agent extracts position features, density distribution features, and mechanical response features from the real-time perception data, and evaluates the degree of abnormality by comparing with the preset normal range boundary, generating a perception layer output result containing feature vectors and abnormality identifiers. The decision-making agent receives the output of the perception layer and, based on the relative positional relationship between the spatial coordinates of the implantation site and the surrounding anatomical structures in the location features, generates a path planning result by constructing a spatial curve path from the current instrument position to the target position. Based on the anomaly markers, the anatomical risk areas that need to be avoided at the current oral implantation site are determined, and the path planning results are corrected by avoiding the anatomical risk areas based on the location characteristics. The decision-making agent establishes a mapping relationship between bone density and implantation torque requirements based on the density distribution characteristics. It calculates torque control target values ​​corresponding to different bone density regions based on the mapping relationship to generate a torque control strategy. Based on the real-time resistance feedback information in the mechanical response characteristics and the bone hardness information in the density distribution characteristics, it calculates feed rate control target values ​​to generate a feed rate control strategy. The force control strategy is obtained by combining the torque control strategy and the feed rate control strategy.

[0025] like Figure 2 As shown, the method includes: The perception layer agent receives real-time perception data from an oral CT scanner, laser rangefinder, and torque sensor. For location feature extraction, the agent uses a voxel segmentation algorithm to extract the three-dimensional spatial coordinates (x=23.5mm, y=15.2mm, z=10.8mm) of the target implant site from the CT image. Simultaneously, it identifies the boundary coordinate set {(x1y1, z1), (x2, y2z2), ...} of adjacent anatomical structures, such as the mandibular canal. It calculates the minimum distance from the implant site to key anatomical structures, such as the distance to the mandibular canal, which is 2.3mm. Density distribution features are obtained by analyzing the grayscale histogram of CT values, including the average bone mineral density (850HU) of the implant area, the density gradient curve, and the proportion of bone types (e.g., 75% D2 bone and 25% D3 bone). Mechanical response features are obtained by collecting real-time resistance data when the drill bit contacts bone tissue using a torque sensor, recording the instantaneous torque value at different depths (e.g., 15 N·cm at a depth of 5mm) and the rate of change of resistance.

[0026] The anomaly identification module evaluates abnormalities by setting threshold boundaries. Location anomalies are determined based on a safe distance threshold (2.0 mm) from the implantation site to key anatomical structures; when the measured distance is less than this threshold, a location anomaly is identified. Density anomalies are determined by comparing bone mineral density values ​​to the normal reference range (400 HU-1200 HU); deviations from this range are identified as density anomalies. Mechanical anomalies are determined based on a torque mutation threshold (10 N·cm / s); when the rate of torque change exceeds this threshold, a mechanical anomaly is identified. The perception layer ultimately generates an output containing each feature vector and its corresponding anomaly identifier {normal / abnormal location, normal / abnormal density, normal / abnormal mechanical anomaly}.

[0027] After receiving the output from the perception layer, the decision-making agent performs path planning. Based on the spatial coordinates of the planting site (23.5mm, 15.2mm, 10.8mm) and the current instrument position coordinates (20.0mm, 18.5mm, 5.0mm) from the location features, it constructs a smooth spatial curve using a cubic spline interpolation algorithm, generating an initial path point set {P1, P2, ..., P...} from the current position to the target position. 10 Each path point contains three-dimensional coordinates and angle parameters, and the spacing between path points is set to 0.5mm to ensure smooth motion.

[0028] When the abnormal markers output by the perception layer indicate the presence of an anatomical risk area, the decision-making agent performs path correction. For example, if the distance to the mandibular nerve canal is detected to be less than the safety threshold (measured at 2.3 mm, close to the threshold of 2.0 mm), the agent models the risk area as a restricted area sphere with a radius of 3.0 mm and corrects the original path by introducing a virtual repulsive field algorithm. The corrected path point set {P1', P2', ..., P...} 12 Maintain a minimum safe distance of 2.5mm from the risk area and add two waypoints to achieve a smoother avoidance trajectory.

[0029] The decision-making agent establishes a mapping relationship between bone density and implantation torque requirements based on density distribution characteristics. By querying a pre-established density-torque correspondence table, it maps different detected bone regions, such as D2 type bone region density of 850 HU and D3 type bone region density of 520 HU, to corresponding torque control target values ​​(D2 type bone: 35 N·cm, D3 type bone: 25 N·cm). For regions with density variations, piecewise linear interpolation is used to calculate the torque target corresponding to the intermediate density value, forming a complete torque control strategy.

[0030] A feed rate control strategy is generated based on the combined mechanical response characteristics and density distribution characteristics. The agent inputs real-time resistance feedback information and bone stiffness information into the adaptive controller. By comparing the deviation between the current torque and the target torque (target 35 N·cm, deviation 20 N·cm), the optimal feed rate is calculated. In the high-density region (>800 HU), the baseline feed rate is set to 0.5 mm / s; in the medium-density region (500-800 HU), it is set to 0.8 mm / s; and in the low-density region (<500 HU), it is set to 0.11 mm / s. The decision-making agent integrates path planning results, torque control strategy, and feed rate control strategy to form a complete force control strategy, which includes a set of path points, the target torque value and feed rate value corresponding to each path point, to achieve real-time and precise control during the implantation process, ensuring surgical safety and accuracy.

[0031] In one alternative implementation, spatial-temporal registration is performed between heterogeneous sensors, and multimodal representations describing the same anatomical feature are extracted from the real-time sensing data, and a consistency deviation metric is calculated, including: The timestamp information and spatial coordinate system information of the real-time sensing data are obtained. The timestamp information is time-aligned by establishing a unified time reference. The spatial coordinate system information is transformed to a unified surgical spatial coordinate system by constructing a spatial coordinate transformation matrix, thereby generating a registered multi-sensor dataset. Different modal data describing the same anatomical feature are identified from the registered multi-sensor dataset. Spatial geometric morphology representation, mechanical property representation and surface texture representation are extracted for the same anatomical feature of oral implant site to form a multimodal representation set describing the same anatomical feature. The multimodal representation set is transformed into a unified feature representation space. By constructing a cross-modal feature mapping relationship, the spatial geometric morphology representation, the mechanical property representation, and the surface texture representation are aligned in the unified feature representation space. The spatial position deviation, physical property deviation, and surface morphology deviation between the aligned different modal representations are calculated and combined to generate the consistency deviation metric.

[0032] The timestamp information includes the precise time each sensor acquired data. For example, the scanning timestamp of the optical scanner is "2023-05-15 10:30:25.456", while the acquisition timestamp of the mechanical sensor is "2023-05-15 10:30:25.654". The spatial coordinate system information includes the position representation of each sensor's data in its local coordinate system. For example, the optical scanner uses a Cartesian coordinate system with the scanner as the origin, while the mechanical sensor uses a polar coordinate system with the probe tip as the origin. By selecting a time reference from the main sensor, a unified time reference system is established. A linear interpolation method is used to time-align the data from different sensors, mapping all sensor data to a unified time sampling point. For example, when the main sensor's sampling frequency is 50Hz, and the auxiliary sensor's sampling frequency is 30Hz, the auxiliary sensor's data is resampled to 50Hz using an interpolation algorithm to ensure time alignment.

[0033] For spatial registration, a spatial coordinate transformation matrix is ​​constructed, containing rotation and translation components. This matrix transforms data from different coordinate systems to a unified surgical spatial coordinate system. The transformation matrix is ​​built upon a pre-calibrated set of reference points, which are clearly defined in different coordinate systems. For example, in the dental implant environment, feature points on the teeth can be used as reference points, which can be accurately identified in both optical and CT scans. By solving for the optimal transformation relationship between these corresponding points, a coordinate transformation matrix is ​​generated with an accuracy within 0.1 mm. After applying these transformations, a registered multi-sensor dataset is generated, where all data corresponds to a unified time point and spatial coordinate system.

[0034] Key anatomical structures were located using a feature matching algorithm. For dental implant sites, the implantation area was segmented in the optical scanning data using a region growing algorithm. This area typically includes the predetermined implantation site and the alveolar bone region within a 2-3 mm radius around it. Simultaneously, the corresponding contact area was identified in the biomechanical data, and the same anatomical structures were located in the CT data. For the identified anatomical features, different modal representations were extracted: spatial geometric features, including surface curvature, roughness, and local height variations, were extracted from the optical scanning data, with typical curvature values ​​ranging from 0.1 to 5 mm. -1 Mechanical properties are extracted from mechanical detection data, including tissue stiffness, elastic modulus, and impedance characteristics. For example, the elastic modulus of healthy alveolar bone is typically in the range of 10-18 GPa. Surface texture characteristics are extracted from high-resolution images, including color distribution, texture directionality, and local density. These extracted features form a multimodal representation set describing the same anatomical feature.

[0035] Feature embedding technology is employed to map features of different dimensions and scales to a 128-dimensional feature vector space. Within this unified feature representation space, cross-modal feature mapping relationships are constructed to establish correspondences between features of different modalities. For example, high curvature regions in spatial geometry typically correspond to high hardness regions in mechanical properties; this mapping relationship is learned by analyzing correlation patterns in historical data. After feature alignment, the consistency deviation between different modal representations is calculated: spatial position deviation is obtained by calculating the Euclidean distance between geometric center points; the deviation in healthy tissue is typically less than 0.2 mm. Physical property deviation is obtained by calculating the difference in standardized mechanical properties, such as the percentage of hardness deviation. Surface morphology deviation is obtained by calculating the cosine similarity between texture feature vectors; the similarity in normal tissue should be greater than 0.85. Finally, these three types of deviations are weighted and combined to generate a comprehensive consistency deviation metric. The weights are allocated as follows: spatial position deviation 0.4, physical property deviation 0.4, and surface morphology deviation 0.2. The resulting comprehensive deviation value ranges from 0 to 1, where 0 represents complete consistency and 1 represents complete inconsistency.

[0036] By employing the aforementioned spatial-temporal registration and multimodal characterization extraction and alignment methods, the anatomical consistency of oral implant sites can be effectively assessed, providing precise navigation support for oral implant surgery. Experimental data demonstrate that this method can improve the accuracy of multi-sensor fusion to the sub-millimeter level, significantly enhancing the precision and success rate of implant placement.

[0037] In one optional implementation, based on the consistency deviation metric and the confidence level of each sensor, a correction mechanism is used to compensate the real-time sensing data by constructing cross-modal constraints, generating fused sensing data including: Based on the historical data accuracy, current operating stability, and measurement environment interference of the heterogeneous sensor, the confidence value of the heterogeneous sensor is calculated. When the consistency deviation metric exceeds the preset consistency threshold, a correction mechanism is triggered to identify inconsistent modal representations in the multimodal representation set and extract the corresponding confidence values. Based on the confidence values, the inconsistent modal representations are divided into reference modal representations and modal representations to be compensated. In a unified feature representation space, with the goal of minimizing the deviation between the corresponding feature components of the modal representation to be compensated and the reference modal representation, the compensation correction amount of the modal representation to be compensated is calculated. The compensation correction amount is superimposed on the real-time sensing data corresponding to the modal representation to be compensated, and the compensated sensor data is fused with the uncorrected real-time sensing data to generate the fused sensing data.

[0038] The confidence score of heterogeneous sensors is primarily based on three factors: historical data accuracy, current operating stability, and the degree of interference in the measurement environment. Historical data accuracy is derived by statistically analyzing the deviation between the sensor's historical output and its true value. For example, the average error rate of the sensor over the past 100 measurements can be used as a metric. If a radar sensor shows an average error rate of 3% in its historical records, its historical accuracy score is 0.97. Current operating stability is assessed by monitoring real-time parameter fluctuations of the sensor, such as voltage stability and temperature changes. A camera sensor operating within a normal temperature range (e.g., 35°C) and with voltage fluctuations not exceeding ±2% of the nominal value can achieve a stability score of 0.95. The degree of interference in the measurement environment considers the impact of external factors on sensor performance, such as the effect of lighting conditions on visual sensors or the reduction in radar performance due to rain or snow. In clear weather, the environmental score of a visual sensor is 0.98, while it drops to 0.65 in foggy conditions. By combining these three indicators, the final confidence score of the sensor is calculated using a weighted average. For example, for a specific sensor, if the weights of the three scores are 0.4, 0.3, and 0.3 respectively, the final confidence score is 0.97×0.4+0.95×0.3+0.98×0.3=0.967.

[0039] After calculating the consistency deviation metric, it needs to be compared with a preset consistency threshold. Assuming the consistency threshold is set to 0.15, a calculated consistency deviation metric of 0.21 exceeds the threshold, triggering a correction mechanism. The correction mechanism first identifies inconsistent modal representations within the multimodal representation set. For example, in an autonomous vehicle, suppose the radar sensor detects an obstacle 50 meters away, while the vision sensor estimates the distance at 42 meters, and the lidar measures 48 meters. There is a significant difference between these three measurements, identifying a large deviation between the vision sensor's output and the other two sensors. The corresponding confidence values ​​are extracted; assuming the confidence levels for radar, vision sensor, and lidar are 0.92, 0.78, and 0.96, respectively. Since lidar has the highest confidence level, followed by radar, and then the vision sensor, the modal representations of lidar and radar are considered reference modal representations, while the visual sensor's modal representation is classified as a modal representation to be compensated.

[0040] After determining the reference modal representation and the modal representation to be compensated, the compensation correction amount for the modal representation to be compensated needs to be calculated in a unified feature representation space. Taking the example above, the data from the three sensors are first mapped to the same coordinate system. Assuming the weighted average of the reference modal representation (LiDAR and radar) is 49.1 meters, and the modal representation to be compensated (visual sensor) is 42 meters, the difference between them is 7.1 meters. With the goal of minimizing this deviation, the compensation correction amount required for the visual sensor data is calculated to be +7.1 meters. In practical applications, the compensation process is more complex, requiring consideration of feature differences across multiple dimensions, not just a single distance value, but the principle remains the same: determining the compensation amount by minimizing the feature deviation between different modal representations.

[0041] After the compensation correction is calculated, it is superimposed on the real-time sensing data corresponding to the modality to be compensated. In this example, the original measurement of 42 meters from the visual sensor is increased by a compensation correction of 7.1 meters, resulting in a compensated data of 49.1 meters. This compensated visual sensor data, together with the uncorrected radar and lidar data, is used to generate the final fused sensing data through a weighted average fusion algorithm. Assuming that the weights are calculated based on the confidence levels of each sensor, the final fused obstacle distance is 49.0 meters, which is more reliable than relying solely on the result of any one sensor.

[0042] This perception data fusion method based on confidence and consistency assessment can effectively reduce errors caused by single sensor failures or environmental interference, improve the accuracy and robustness of the overall perception system, and provide more reliable environmental perception capabilities for applications such as autonomous driving and robot navigation.

[0043] In one optional implementation, based on the fused perception data, the decision-making agent generates position control layer commands and force control layer commands for the dental implant robot by combining biomechanical constraints and collision safety constraints, including: Bone density distribution information, bone thickness information, and surrounding anatomical structure location information of oral tissues are extracted from the fused perception data. Biomechanical bearing capacity parameters of oral tissues are established. Biomechanical constraints are obtained by limiting the operating torque and torque of the implantation robot to not exceed the biomechanical bearing capacity parameters. The current location information of the implantation device and the surgical restricted area are extracted from the fused perception data. Based on the minimum safe distance between the implantation device and the surgical restricted area, collision safety constraints are obtained. The decision-making agent verifies the collision safety constraints of the path planning results. When the path planning results do not meet the collision safety constraints, the spatial curve path of the path planning results is adjusted based on the minimum safe distance and converted into a position control layer instruction containing a position coordinate sequence and an attitude angle sequence. The decision-making agent performs biomechanical constraint verification on the force control strategy. When the torque control strategy or feed rate control strategy does not meet the biomechanical constraints, it performs amplitude limiting processing on the torque control target value and feed rate control target value based on the biomechanical bearing capacity parameter, and converts them into force control layer instructions containing force control parameters and torque control parameters.

[0044] Information on bone density distribution, bone thickness, and the location of surrounding anatomical structures in oral tissues is extracted from fused sensory data. In practice, CT imaging data is used to obtain the density distribution of oral bone. A bone density mapping relationship is established by analyzing CT values. For example, for D1 type bone regions, a CT value typically greater than 850 HU indicates high-density cortical bone; for D2 type bone regions, a CT value between 500-850 HU indicates moderate-density bone; for D3 type bone regions, a CT value between 250-500 HU indicates low-density bone; and for D4 type bone regions, a CT value less than 250 HU indicates very low-density bone. Simultaneously, CBCT 3D reconstruction technology is used to measure bone thickness in the oral region; for example, the bone thickness in the maxillary posterior teeth region is approximately 5-8 mm, and in the mandibular anterior teeth region, it is approximately 7-11 mm. Through optical scanning and electromagnetic navigation technology, key anatomical structures such as the inferior alveolar nerve canal and the maxillary sinus floor are located in real time, generating 3D coordinate data and marking the relative positional relationship between these structures and the surgical area.

[0045] Based on the extracted oral tissue information, biomechanical bearing capacity parameters for oral tissues were established, and corresponding torque and moment thresholds were set for different bone types. For example, the maximum bearing torque for D1 type bone was set to 45 N·cm, and the maximum drilling torque to 3.5 N·m; the maximum bearing torque for D2 type bone was set to 35 N·cm, and the maximum drilling torque to 2.8 N·m; the maximum bearing torque for D3 type bone was set to 25 N·cm, and the maximum drilling torque to 2.0 N·m; and the maximum bearing torque for D4 type bone was set to 15 N·cm, and the maximum drilling torque to 1.2 N·m. By limiting the operating torque and moment of the implantation robot to not exceed the above biomechanical bearing capacity parameters, biomechanical constraints were obtained.

[0046] The decision-making agent extracts the current position information of the implant instrument and the surgical restricted area information from the fused perception data. It uses an optical tracking system to record the real-time position of the implant instrument's tip with an accuracy of 0.1 mm. The surgical restricted area includes key anatomical structures such as the inferior alveolar nerve canal, the floor of the maxillary sinus, and the roots of adjacent teeth. These areas are marked and updated in real time through preoperative planning. Collision safety constraints are established based on the safe distance requirements between the implant instrument and the surgical restricted area. For example, a minimum safe distance of 2 mm is set for the inferior alveolar nerve canal; 1.5 mm for the floor of the maxillary sinus; and 1.8 mm for the roots of adjacent teeth.

[0047] The decision-making agent verifies the collision safety constraints of the planned surgical path. In actual operation, it calculates the shortest distance from each discrete point on the path to each restricted area, ensuring that the distance to all points is greater than the preset safety distance. When a point in the path planning result is detected to be less than the safety distance from a restricted area, a path adjustment algorithm is triggered. For example, for a path point only 1.2 mm from the inferior alveolar nerve canal, it is translated 0.8 mm away from the nerve canal, so that the final distance reaches the 2 mm safety threshold. After the path adjustment is completed, the spatial curve is converted into a sequence of position coordinates and attitude angles that the robot can execute, forming position control layer instructions. The position control layer instructions include 100 intermediate discrete position points of the implant drill from the entry point (10.5, 15.2, 8.7) mm to the target point (10.5, 15.2, 18.7) mm, and the corresponding tool attitude angles (Roll=-5°, Pitch=0°, Yaw=10°).

[0048] The decision-making agent performs biomechanical constraint verification on the force control strategy, checking whether the torque control strategy meets the biomechanical constraints. For example, in the D2 type bone region, when the originally planned torque control target value of 38 N·cm is detected, exceeding the maximum torque that this type of bone can withstand (35 N·cm), the target value is limited to 35 N·cm. Similarly, it checks whether the feed rate control strategy meets the biomechanical constraints. For example, for the D3 type bone, the originally planned feed rate is 1.8 mm / s, exceeding the safe feed rate of 1.5 mm / s for this type of bone, so the feed rate is limited to 1.5 mm / s. After limiting, the torque control target value and the feed rate control target value are converted into force control layer instructions, including the maximum allowable force parameter (25 N), force control impedance parameters (stiffness: 200 N / m, damping: 20 N·s / m), maximum torque limit (35 N·cm), and feed rate parameter (1.5 mm / s).

[0049] Through the above technical solutions, the dental implant robot can automatically generate reasonable position control commands and force control commands while ensuring biosafety and surgical precision, thus ensuring the safe and effective execution of implant surgery.

[0050] In one optional implementation, establishing the switching conditions and cooperative constraint relationships between the position control layer and the force control layer includes: Real-time position deviation information and real-time contact force information of the implantation device are extracted from the fused sensing data. An activation threshold for the position-dominated control mode is set based on the real-time position deviation information, and an activation threshold for the force-dominated control mode is set based on the real-time contact force information. Switching conditions between the position control layer and the force control layer are established. The switching conditions stipulate that when the real-time position deviation information exceeds the activation threshold of the position-dominated control mode, the system switches to the position-dominated control mode, and when the real-time contact force information exceeds the activation threshold of the force-dominated control mode, the system switches to the force-dominated control mode. The position control motion speed is extracted from the position control layer command, and the force control target force value is extracted from the force control layer command. A coupling constraint rule is established between the position control motion speed and the force control target force value by constructing a cooperative constraint relationship. The coupling constraint rule stipulates that in the position-dominated control mode, the position control motion speed is constrained by the safety boundary of the force control target force value, and in the force-dominated control mode, the force control target force value is constrained by the trajectory boundary of the position control motion speed.

[0051] Real-time positional deviation information is calculated by comparing the Euclidean distance between the current position and the desired position of the dental implant instrument, with the unit being mm. Real-time contact force information is obtained by collecting the force value generated when the implant instrument comes into contact with oral tissue using a force sensor, with the unit being N. During the extraction process, a Kalman filter algorithm is used to reduce noise in the raw data to minimize the impact of environmental interference on the measurement data. The accuracy of the filtered positional deviation can reach ±0.1 mm, and the accuracy of the force information measurement can reach ±0.05 N.

[0052] Based on the extracted real-time positional deviation information, an activation threshold for the position-dominant control mode is set. This threshold is adaptively adjusted according to the precision requirements of different dental implant procedures. For example, for delicate implant placement tasks, the positional deviation activation threshold is set to 0.5 mm, and for preliminary drilling tasks, the positional deviation activation threshold is set to 0.8 mm. When the actual measured positional deviation exceeds the corresponding threshold, it will be determined that the current task needs to prioritize ensuring positional accuracy, and the switch to the position-dominant control mode will be triggered.

[0053] Based on the extracted real-time contact force information, an activation threshold for the force-dominated control mode is set. This threshold is determined according to the characteristics of oral tissues and the implantation stage. For example, the force activation threshold is set to 1.0N for areas close to nerve tissue and 2.5N for areas contacting bone tissue. When the actual measured contact force exceeds the corresponding threshold, it is determined that the current task requires prioritizing force control safety, and a switch to the force-dominated control mode is triggered.

[0054] When establishing the switching conditions between the position control layer and the force control layer, an event-triggered switching mechanism is adopted. Specifically, it continuously monitors real-time position deviation and contact force information. When the real-time position deviation exceeds the activation threshold of the position-dominated control mode, a switching event signal is generated, and the control layer immediately switches to the position-dominated control mode. When the real-time contact force exceeds the activation threshold of the force-dominated control mode, another switching event signal is generated, and the control layer immediately switches to the force-dominated control mode. To avoid system oscillation caused by frequent switching, a minimum mode hold time of 200ms is set, meaning that the current mode must be maintained for at least 200ms after a switch before the next switching evaluation can be performed.

[0055] Once the switching conditions are established, the position control motion velocity is extracted from the position control layer commands, and the force control target force value is extracted from the force control layer commands. The position control motion velocity is represented as the velocity components along each axis in three-dimensional space, in mm / s; the force control target force value is represented as the magnitude of the contact force to be maintained along each axis, in N. During the extraction process, the original commands are interpolated and smoothed to ensure smooth changes in control quantities and avoid abrupt changes that could lead to mechanical shocks.

[0056] A collaborative constraint relationship is established, creating coupling restriction rules between the position-controlled motion speed and the force-controlled target force value. In the position-dominated control mode, the position-controlled motion speed is constrained by the safety boundary of the force-controlled target force value. Specifically, when the actual contact force detected by the force sensor approaches 80% of the force-controlled target force value, the position-controlled motion speed begins to decrease incrementally, with the adjustment ratio proportional to the difference between the contact force and the target force. For example, when the actual contact force reaches 85% of the target force, the position-controlled motion speed is reduced to 65% of the original speed; when the actual contact force reaches 90% of the target force, the position-controlled motion speed is reduced to 40% of the original speed; when the actual contact force reaches or exceeds the target force, the position-controlled motion speed is reduced to 5% of the original speed or stops completely to protect sensitive oral tissues.

[0057] In force-dominated control mode, the target force value is constrained by the trajectory boundary of the position-controlled motion speed. Specifically, a safe operating area is generated based on a preset implantation trajectory. When the force-controlled implantation instrument deviates from the boundary of this area by a preset distance threshold (e.g., 0.7mm), the target force value is automatically adjusted to bring the implantation instrument back towards the safe trajectory. The adjustment range is proportional to the deviation distance. When the deviation distance reaches 1.0mm, the target force value decreases by 50% in the deviation direction; when the deviation distance reaches 1.5mm, the target force value decreases by 80% in the deviation direction, while an auxiliary force is added in the return direction to ensure that the implant is inserted along the preset optimal trajectory.

[0058] The above technical solution enables intelligent switching and collaborative constraint between the position control layer and the force control layer, allowing dental implant instruments to ensure positional accuracy while also taking into account contact force safety. This effectively improves the autonomous operation capability and adaptability of dental implant robots in complex oral environments, making them particularly suitable for dental implant tasks that require precise positioning and are sensitive to contact forces, such as implant placement and bone drilling.

[0059] In one optional implementation, the dental implant robot, in position-dominant control mode, is dominated by position control layer commands and uses force control layer commands as safety constraint boundaries; in force-dominant control mode, it is dominated by force control layer commands and uses position control layer commands as trajectory constraint boundaries, including: The dental implant robot determines the control mode to be used in the current stage of the dental implant surgery based on the switching conditions. In the position-dominated control mode, the position control target is extracted from the position control layer command as the dominant control target, and the upper limit value of the force control target force value is extracted from the force control layer command as the safety constraint boundary. This limits the movement trajectory of the planting robot to follow the position control target and ensures that the applied force does not exceed the safety constraint boundary. When the safety constraint boundary is exceeded, the position control movement speed is reduced until it falls back into the safety constraint boundary. In the force-dominated control mode, the force control target force value is extracted from the force control layer command as the dominant control target, and the deviation range of the position control target is extracted from the position control layer command as the trajectory constraint boundary. This limits the force applied by the planting robot to follow the force control target force value and the position deviation to remain within the trajectory constraint boundary. When the position deviation exceeds the trajectory constraint boundary, the force control direction is adjusted until the position deviation returns to within the trajectory constraint boundary.

[0060] The dental implant robot first needs to determine the appropriate control mode for the current stage of the implant surgery based on switching conditions. These conditions can be based on various factors such as the stage of surgery, drilling depth, contact force feedback, and changes in tissue impedance. For example, during the initial approach to the oral cavity, if the detected force is less than 0.5N, a position-driven control mode is used; when contact is made with the alveolar bone and the detected force exceeds 0.5N but is less than 5N, the system can switch to a force-driven control mode for precise drilling. The control system determines the appropriate control mode by real-time monitoring of data from force and position sensors, combined with preset threshold parameters, and executing a mode-determining algorithm.

[0061] In position-dominated control mode, the dental implant robot extracts the position control target from the position control layer commands as the dominant control target. The position control target includes a preset sequence of three-dimensional spatial coordinate points, forming an ideal trajectory for implant placement. Simultaneously, it extracts the upper limit of the force control target force value from the force control layer commands as a safety constraint boundary. For example, the upper limit of the force value can be set to 2N during the soft tissue contact stage; 8N during the bone drilling stage; and 15N during the implant placement stage. Based on the position control target, the robot control system calculates the motor output torque through a PID controller, enabling the robotic arm end effector to precisely follow the predetermined trajectory. Simultaneously, force sensors monitor the contact force in real time. When the detected force value approaches the safety constraint boundary, such as reaching 90% of the threshold, the position control movement speed begins to decrease. When the force value exceeds the safety constraint boundary, the control system immediately reduces the position control speed by 50% and continues to decelerate until the force value falls back within the safety constraint boundary. For example, if the robot is drilling at a speed of 2 mm / s and detects that the force exceeds the safety threshold of 8 N, it immediately reduces the speed to 1 mm / s. If the force still exceeds the threshold, it continues to reduce the speed to 0.5 mm / s until the force falls back to a safe range. This mechanism ensures that the robot maintains trajectory accuracy without causing excessive pressure damage to oral tissues.

[0062] In force-dominated control mode, the dental implant robot extracts the target force value from the force control layer commands as the dominant control target. This target force value is preset according to different surgical stages and tissue types, such as 6N for cortical bone drilling, 4N for cancellous bone drilling, and 10N for implant insertion. Simultaneously, the deviation range of the position control target is extracted from the position control layer commands as the trajectory constraint boundary. For example, during drilling, the allowable position deviation range is ±0.2mm; during implant insertion, the allowable position deviation range is ±0.1mm. The control system adjusts the robotic arm output force through an impedance control algorithm to maintain the actual contact force near the target force value. It calculates the deviation between the current position and the ideal trajectory in real time. When the position deviation reaches 80% of the trajectory constraint boundary (e.g., ±0.16mm), it begins to adjust the force control direction, gradually bringing the end effector closer to the ideal trajectory. When the position deviation exceeds the trajectory constraint boundary (e.g., exceeding ±0.2mm), the target force value is temporarily reduced by 30%, and the force direction is adjusted to bring the robotic arm end effector back to the ideal trajectory. For example, if a drilling force of 6N is initially applied, and a positional deviation exceeding 0.2mm is detected, the force is temporarily reduced to 4.2N, and the direction of the applied force is adjusted to reduce the positional deviation. Once the positional deviation returns to within the trajectory constraint boundary, the force is gradually restored to the original target force value for force control.

[0063] In practical applications, the control system of the dental implant robot employs a 20ms sampling cycle to process sensor data and update control commands in real time. The control system incorporates multiple sets of safety parameter configurations, covering different patients' oral characteristics and various complex clinical situations. The control algorithm has been validated through software simulation and in vitro testing, enabling smooth switching of control modes under various conditions to ensure the safety and reliability of the surgical procedure. In clinical applications, it demonstrates high stability, with position control accuracy reaching ±0.05mm and force control accuracy reaching ±0.2N, meeting the stringent requirements of dental implant surgery.

[0064] By implementing the above control methods, the dental implant robot can intelligently adapt to the control needs of different stages of implant surgery, achieving precise implant placement while ensuring safety, effectively improving the success rate of implant surgery and the postoperative recovery effect of patients.

[0065] This invention relates to a control system for an implantation surgery robot based on intelligent agent fusion of multiple sensors, the system comprising: The first unit is used to acquire real-time sensing data from multiple heterogeneous sensors during oral implant surgery; The second unit is used to extract features and identify anomalies from the real-time sensing data through the perception layer agent, and the decision layer agent performs path planning and force control strategy generation based on the output of the perception layer agent. The third unit is used to perform spatial-temporal registration between heterogeneous sensors, extract multimodal representations describing the same anatomical feature from the real-time sensing data and calculate a consistency deviation metric; based on the consistency deviation metric and the confidence level of each sensor, a correction mechanism is used to compensate the real-time sensing data by constructing cross-modal constraints to generate fused sensing data. The fourth unit is used to generate position control layer instructions and force control layer instructions for the oral implant robot based on the fused perception data, combining biomechanical constraints and collision safety constraints, and to establish switching conditions and cooperative constraint relationships between the position control layer and the force control layer. The fifth unit is used to perform oral implant surgery based on the switching conditions and the cooperative constraint relationship. In the position-dominant control mode, the oral implant robot takes the position control layer command as the main force and the force control layer command as the safety constraint boundary. In the force-dominant control mode, the force control layer command as the main force and the position control layer command as the trajectory constraint boundary.

[0066] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0067] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0068] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A control method for an implantation surgery robot based on agent-based multi-sensor fusion, characterized in that, include: Acquire real-time sensing data from multiple heterogeneous sensors during dental implant surgery; The perception layer agent extracts features and identifies anomalies from the real-time perception data, and the decision layer agent generates path planning and force control strategies based on the output of the perception layer agent. Spatial-temporal registration is performed between heterogeneous sensors, and multimodal representations describing the same anatomical feature are extracted from the real-time sensing data, with consistency deviation metrics calculated, including: The timestamp information and spatial coordinate system information of the real-time sensing data are obtained. The timestamp information is time-aligned by establishing a unified time reference. The spatial coordinate system information is transformed to a unified surgical spatial coordinate system by constructing a spatial coordinate transformation matrix, thereby generating a registered multi-sensor dataset. Different modal data describing the same anatomical feature are identified from the registered multi-sensor dataset. Spatial geometric morphology representation, mechanical property representation and surface texture representation are extracted for the same anatomical feature of oral implant site to form a multimodal representation set describing the same anatomical feature. The multimodal representation set is transformed into a unified feature representation space. By constructing a cross-modal feature mapping relationship, the spatial geometric morphology representation, the mechanical property representation, and the surface texture representation are feature aligned in the unified feature representation space. The spatial position deviation, physical property deviation, and surface morphology deviation between the aligned different modal representations are calculated and combined to generate the consistency deviation metric. Based on the consistency deviation metric and the confidence level of each sensor, a correction mechanism is used to compensate the real-time sensing data by constructing cross-modal constraints, thereby generating fused sensing data. Based on the fused perception data, the decision-making agent combines biomechanical constraints and collision safety constraints to generate position control layer commands and force control layer commands for the dental implant robot, and establishes switching conditions and cooperative constraint relationships between the position control layer and the force control layer. Based on the switching conditions and the cooperative constraint relationship, the dental implant robot takes position control layer commands as the main force and force control layer commands as the safety constraint boundary in position-dominated control mode, and takes force control layer commands as the main force and position control layer commands as the trajectory constraint boundary in force-dominated control mode.

2. The method according to claim 1, characterized in that, The real-time perceived data is subjected to feature extraction and anomaly identification by the perception layer agent, and the decision layer agent performs path planning and force control strategy generation based on the output of the perception layer agent, including: The perception layer agent extracts position features, density distribution features, and mechanical response features from the real-time perception data, and evaluates the degree of abnormality by comparing with the preset normal range boundary, generating a perception layer output result containing feature vectors and abnormality identifiers. The decision-making agent receives the output of the perception layer and, based on the relative positional relationship between the spatial coordinates of the implantation site and the surrounding anatomical structures in the location features, generates a path planning result by constructing a spatial curve path from the current instrument position to the target position. Based on the anomaly markers, the anatomical risk areas that need to be avoided at the current oral implantation site are determined, and the path planning results are corrected by avoiding the anatomical risk areas based on the location characteristics. The decision-making agent establishes a mapping relationship between bone density and implantation torque requirements based on the density distribution characteristics, and calculates torque control target values ​​corresponding to different bone density regions according to the mapping relationship to generate torque control strategies. Based on the real-time resistance feedback information in the mechanical response characteristics and the bone hardness information in the density distribution characteristics, the feed rate control target value is calculated to generate the feed rate control strategy. The force control strategy is obtained by combining the torque control strategy and the feed rate control strategy.

3. The method according to claim 1, characterized in that, Based on the consistency deviation metric and the confidence level of each sensor, a correction mechanism is used to compensate for the real-time sensing data by constructing cross-modal constraints, generating fused sensing data including: Based on the historical data accuracy, current operating stability, and measurement environment interference of the heterogeneous sensor, the confidence value of the heterogeneous sensor is calculated. When the consistency deviation metric exceeds the preset consistency threshold, a correction mechanism is triggered to identify inconsistent modal representations in the multimodal representation set and extract the corresponding confidence values. Based on the confidence values, the inconsistent modal representations are divided into reference modal representations and modal representations to be compensated. In a unified feature representation space, with the goal of minimizing the deviation between the corresponding feature components of the modal representation to be compensated and the reference modal representation, the compensation correction amount of the modal representation to be compensated is calculated. The compensation correction amount is superimposed on the real-time sensing data corresponding to the modal representation to be compensated, and the compensated sensor data is fused with the uncorrected real-time sensing data to generate the fused sensing data.

4. The method according to claim 1, characterized in that, Based on the fused perception data, the decision-making layer agent generates position control layer commands and force control layer commands for the dental implant robot by combining biomechanical constraints and collision safety constraints, including: Bone density distribution information, bone thickness information, and surrounding anatomical structure location information of oral tissues are extracted from the fused perception data. Biomechanical bearing capacity parameters of oral tissues are established. Biomechanical constraints are obtained by limiting the operating torque and torque of the implantation robot to not exceed the biomechanical bearing capacity parameters. The current location information of the implantation device and the surgical restricted area are extracted from the fused perception data. Based on the minimum safe distance between the implantation device and the surgical restricted area, collision safety constraints are obtained. The decision-making agent verifies the collision safety constraints of the path planning results. When the path planning results do not meet the collision safety constraints, the spatial curve path of the path planning results is adjusted based on the minimum safe distance and converted into a position control layer instruction containing a position coordinate sequence and an attitude angle sequence. The decision-making agent performs biomechanical constraint verification on the force control strategy. When the torque control strategy or feed rate control strategy does not meet the biomechanical constraints, it performs amplitude limiting processing on the torque control target value and feed rate control target value based on the biomechanical bearing capacity parameter, and converts them into force control layer instructions containing force control parameters and torque control parameters.

5. The method according to claim 1, characterized in that, The switching conditions and collaborative constraints between the position control layer and the force control layer are established, including: Real-time position deviation information and real-time contact force information of the implantation device are extracted from the fused sensing data. An activation threshold for the position-dominated control mode is set based on the real-time position deviation information, and an activation threshold for the force-dominated control mode is set based on the real-time contact force information. Switching conditions between the position control layer and the force control layer are established. The switching conditions stipulate that when the real-time position deviation information exceeds the activation threshold of the position-dominated control mode, the system switches to the position-dominated control mode, and when the real-time contact force information exceeds the activation threshold of the force-dominated control mode, the system switches to the force-dominated control mode. The position control motion speed is extracted from the position control layer command, and the force control target force value is extracted from the force control layer command. A coupling constraint rule is established between the position control motion speed and the force control target force value by constructing a cooperative constraint relationship. The coupling constraint rule stipulates that in the position-dominated control mode, the position control motion speed is constrained by the safety boundary of the force control target force value, and in the force-dominated control mode, the force control target force value is constrained by the trajectory boundary of the position control motion speed.

6. The method according to claim 1, characterized in that, The oral implant robot, in position-dominated control mode, is dominated by position control layer commands and uses force control layer commands as safety constraint boundaries. In force-dominated control mode, it is dominated by force control layer commands and uses position control layer commands as trajectory constraint boundaries, including: The dental implant robot determines the control mode to be used in the current stage of the dental implant surgery based on the switching conditions. In the position-dominant control mode, the position control target is extracted from the position control layer command as the dominant control target, and the upper limit value of the force control target force value is extracted from the force control layer command as the safety constraint boundary, which limits the movement trajectory of the planting robot to follow the position control target and the applied force to not exceed the safety constraint boundary. When the safety constraint boundary is exceeded, the position control movement speed is reduced until it falls back into the safety constraint boundary. In the force-dominated control mode, the force control target force value is extracted from the force control layer command as the dominant control target, and the deviation range of the position control target is extracted from the position control layer command as the trajectory constraint boundary, which limits the applied force of the planting robot to follow the force control target force value and the position deviation to remain within the trajectory constraint boundary; When the position deviation exceeds the trajectory constraint boundary, the force control direction is adjusted until the position deviation returns to within the trajectory constraint boundary.

7. A control system for an implantation surgery robot based on intelligent agent fusion of multiple sensors, used to implement the method as described in any one of claims 1-6, characterized in that, include: The first unit is used to acquire real-time sensing data from multiple heterogeneous sensors during oral implant surgery; The second unit is used to extract features and identify anomalies from the real-time sensing data through the perception layer agent, and the decision layer agent performs path planning and force control strategy generation based on the output of the perception layer agent. The third unit is used to perform spatial-temporal registration between heterogeneous sensors, extract multimodal representations describing the same anatomical feature from the real-time sensing data, and calculate a consistency deviation metric. Based on the consistency deviation metric and the confidence level of each sensor, a correction mechanism is used to compensate the real-time sensing data by constructing cross-modal constraints, thereby generating fused sensing data. The fourth unit is used to generate position control layer instructions and force control layer instructions for the oral implant robot based on the fused perception data, combining biomechanical constraints and collision safety constraints, and to establish switching conditions and cooperative constraint relationships between the position control layer and the force control layer. The fifth unit is used to perform oral implant surgery based on the switching conditions and the cooperative constraint relationship. In the position-dominant control mode, the oral implant robot takes the position control layer command as the main force and the force control layer command as the safety constraint boundary. In the force-dominant control mode, the force control layer command as the main force and the position control layer command as the trajectory constraint boundary.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images