An adaptive extraoral scanning abutment connecting robot system based on embodied intelligence

Abstract

This invention relates to the field of medical robotics. It provides an adaptive extraoral scanning docking station robot system based on embodied intelligence. The system includes: a robot body comprising at least one multi-degree-of-freedom robotic arm; a force sensing module for real-time detection of triaxial forces and torques between the robotic arm and oral tissues; an extraoral scanning device for acquiring high-precision two-dimensional image data of the intraoral scan; an environmental sensing module for real-time acquisition of three-dimensional point cloud data of the oral environment; and a control module comprising a teaching and learning unit, an autonomous scanning unit, a pose calculation unit, and a collision detection unit. The control module is communicatively connected to the robotic arm controller, the extraoral scanning device, an RGB-D camera, and a six-dimensional force sensor. By learning from a large amount of patient data, the system can autonomously adapt to different mouth openings, dental arch morphologies, and other individual anatomical features, generating the optimal scanning path in real time without human intervention.

Description

Technical Field

[0001] This invention relates to the field of medical robot technology, and in particular to an adaptive extraoral scanning connection base robot system based on embodied intelligence. Background Technology

[0002] Dental implant restoration is the preferred treatment for tooth loss in clinical practice. The key to successful restoration lies in accurately acquiring the three-dimensional spatial position and posture information of the implant abutment, thereby customizing a highly fitting superstructure. Currently, the commonly used clinical approach involves mounting a scanner with specific geometric structures and positioning markers onto the abutment. The dentist then uses a handheld scanner to acquire images from multiple angles outside the mouth, relying on image processing and 3D reconstruction technology to calculate the pose and provide a data foundation for the design and fabrication of the restoration. While this conventional scanning mode is widely used in clinical practice, it also reveals several practical limitations. For example, patients have significant individual differences in oral anatomy, including variations in mouth opening, dental arch morphology, intraoral soft tissue distribution, and saliva secretion. For patients with limited mouth opening or unique dental arch structures, the intraoral operating space is limited, restricting the scanning angle of the device. The entire scanning process heavily relies on the dentist's clinical experience for on-the-spot adjustments, lacking a standardized operating procedure.

[0003] During the treatment process, patients need to maintain an open mouth for extended periods, during which they may exhibit subtle head movements, unconscious jaw shifts, and physiological swallowing. Furthermore, the tongue and oral soft tissues can easily shift and obscure the scanning probe, directly causing image sequence errors and poor data consistency. Doctors must frequently interrupt the procedure and reposition the probes, significantly increasing clinical operation time, exacerbating patient discomfort, and directly affecting the accuracy of subsequent abutment pose calculations. These problems are even more pronounced in complex implant cases such as multiple missing teeth or edentulous jaws. These cases often require multiple intraoral probes, with some full-mouth restorations requiring six to eight probes. Doctors need to repeatedly photograph multiple sides of each probe, resulting in a redundant and cumbersome process, low overall treatment efficiency, and fatigue due to prolonged repetitive operations, further reducing scanning consistency and causing fluctuations in data quality.

[0004] Although the industry has attempted to improve scanning stability by using automated solutions with robotic arms equipped with scanners, existing equipment still has significant technical limitations: automated equipment mostly uses pre-set fixed motion trajectories, with a rigid and singular operating mode. It cannot dynamically and adaptively adjust according to the patient's real-time changes in mouth opening or individual anatomical differences, nor can it respond promptly to sudden micro-movements in the patient's position during surgery. This easily creates scanning blind spots, resulting in incomplete data acquisition in critical areas. At the same time, the rigid movement of the robotic arm also poses a safety risk of collision and compression with intraoral soft tissues. The clinical applicability and safety of the equipment are limited, making it difficult to meet the diagnostic and treatment needs of different complex cases and hindering its widespread application in clinical practice. Summary of the Invention

[0005] To address the above issues, this invention teaches the robot, through extensive instruction, the operational strategies employed by doctors to handle different patients, environments, and micro-movements. This enables the robot to autonomously find the optimal scanning angle based on information such as the patient's current mouth opening, oral environment, and scanning body position, and to adapt to the patient's micro-movements in real time. Ultimately, it completes accurate measurements of all scanning bodies, solving the problem of scanning failure or decreased accuracy caused by patient micro-movements in traditional methods.

[0006] According to an embodiment of the present invention, an adaptive external scanning connection base robot system based on embodied intelligence is provided.

[0007] In a first aspect of the invention, an adaptive external scanning docking station robot system based on embodied intelligence is provided. The system includes: Robot body: includes at least one multi-degree-of-freedom robotic arm, which has zero-force dragging function; Force sensing module: includes a six-dimensional force sensor, installed between the end of the robotic arm and the extraoral scanning device, used to detect the triaxial force and triaxial torque of the robotic arm in contact with the oral tissue in real time; Extraoral scanning device: Fixedly mounted on the end effector of the robotic arm, used to acquire high-precision two-dimensional image data of the scanned body inside the oral cavity; Environmental perception module: including an RGB-D camera fixedly mounted on the end effector of the robotic arm, used to acquire three-dimensional point cloud data of the oral cavity environment in real time; Scanning body: rectangular structure, with a threaded interface at the bottom that matches the implant connection abutment, and asymmetrically arranged circular markings printed on all four sides; Control module: Implemented by a high-performance industrial computer, which includes a teaching and learning unit, an autonomous scanning unit, a pose calculation unit and a collision detection unit; the control module is connected to the robotic arm controller, the external scanning device, the RGB-D camera and the six-dimensional force sensor.

[0008] Furthermore, the robotic arm is equipped with a connection interface at its end, and each joint of the robotic arm has a built-in high-precision encoder to obtain the position and pose of the end in the base coordinate system through forward kinematics calculation.

[0009] Furthermore, the specific steps of the aforementioned teaching and learning unit are as follows: The doctor switched the robotic arm to teaching mode through the control software, so that all joints of the robotic arm were in a state of zero force control. The doctor drags the robotic arm to perform multiple complete scans on various head models and the scanning objects inside the mouths of several different patients, collecting data on different mouth openings, different dental arch shapes, different numbers of scanning objects, and different scanning angles. The control module 600 synchronously acquires multimodal data at a frequency of 100Hz: encoder data from each joint of the robotic arm, which, after forward kinematics calculation, yields the pose sequence of the end effector in the base coordinate system. High-precision two-dimensional image frame sequence acquired by the external scanning device 300 3D point cloud data acquired by an RGB-D camera; Extracting state features from multimodal data Including: robotic arm pose Patient's mouth opening Scan body position closest distance Oral obstruction point cloud ; State characteristics As input, the doctor's next movement command. As output, training samples are constructed; An improved generative adversarial imitation learning algorithm is used to train the policy network; after training, the policy network is saved as an embodied intelligent policy model.

[0010] Furthermore, the improved generative adversarial imitation learning algorithm specifically involves: constructing a discriminator network based on a diffusion model, representing state-action pairs as follows: , where the state This includes the patient's mouth opening, scan posture, three-dimensional oral cavity structure information, and movements. These are the control commands for the robotic arm; Its diffusion loss function is defined as: , in, These represent category labels, corresponding to expert-taught data and robot-generated data, respectively. It is Gaussian noise. The noise predicted by the diffusion model; Construct a diffusion discriminant function based on diffusion loss. : , The output range of the diffusion discriminator is , representing the probability that the state-action pair belongs to the expert-taught behavior; during policy learning, a reward function is constructed based on the diffusion discriminator: , By maximizing the reward function, the robotic arm is driven to generate a scanning trajectory that more closely resembles the doctor's instruction.

[0011] Furthermore, the autonomous scanning unit is used to acquire the current state characteristics of the patient's oral environment in real time during autonomous scanning mode, input the state characteristics into the embodied intelligent strategy model, generate an adaptive scanning path that adapts to the current individual characteristics of the patient, and control the robotic arm to carry the extraoral scanning device to scan the scanning body along the adaptive scanning path.

[0012] Furthermore, the process of generating the adaptive scan path is modeled as a sequence decision problem: at time... The system state is represented as , Indicates the current pose of the robotic arm's end effector; Embodied intelligent policy model output in state The following action probability distribution: , Among them, actions This represents the pose increment of the robotic arm's end effector in three-dimensional space; Actions are sampled based on this distribution: , The pose of the robotic arm's end effector is updated using a robot kinematics model.

[0013] in, Given the positive kinematics function of the robotic arm, the motion sequence is obtained by performing multi-step recursion on the above process within a closed-loop control framework. And further generate the corresponding robotic arm end effector trajectory. This forms a continuous adaptive scanning path.

[0014] Furthermore, the pose calculation unit is used to calculate the precise pose of all scanned objects step by step from multiple angles based on the multi-view high-precision two-dimensional images acquired by the extraoral scanning device.

[0015] Furthermore, the collision detection unit is used to calculate the spatial distance between the robotic arm and the oral tissue in real time based on the three-dimensional point cloud data collected by the environmental perception module. When the distance is less than a preset safety threshold, an obstacle avoidance command is generated. The force sensor monitors the contact force in real time. If it is greater than the set threshold, an emergency stop is executed and the arm retracts.

[0016] Furthermore, the calculation process for the spatial distance between the robotic arm and the oral tissue is as follows: The 3D point cloud data collected by the environmental perception module is preprocessed, including noise reduction, voxel filtering and coordinate system calibration, and the point cloud data is uniformly converted to the robot arm base coordinate system. The processed point cloud data is segmented, the effective point set representing the surface of oral soft tissue is extracted, and the point cloud of non-target areas is removed. Construct a spatial index structure to accelerate nearest neighbor search; For the robotic arm's end effector, the minimum Euclidean distance to the oral tissue point cloud set is calculated point by point to obtain the current minimum spatial interval: for the robotic arm Its point cloud set to oral tissue The minimum distance is defined as: .

[0017] In a second aspect of the invention, an adaptive external scanning connection base robot method based on embodied intelligence is provided. The method includes: Step S01: Place the robotic arm equipped with an extraoral scanning device and an RGB-D camera into teaching mode; Step S02: The doctor drags the robotic arm to perform multiple complete scans on various head models and the scanning bodies inside the mouths of multiple different patients. The doctor synchronously records the multimodal data during each scan at a preset frequency and extracts the state features from it. Step S03: Using the state features as input and the doctor's next-move instruction as output, construct a training sample set and train a deep neural network model using an imitation learning algorithm as an embodied intelligence strategy model. Step S04: Place the robotic arm in autonomous scanning mode, collect and extract current state features through an RGB-D camera and an extraoral scanning device, input the current state features into the embodied intelligent strategy model, and generate an initial adaptive scanning path that adapts to the individual characteristics of the current patient. Step S05: Control the robotic arm to carry the external scanning device to move along the adaptive scanning path, and perform closed-loop control at a preset frequency during the movement until all scanning body poses are completed.

[0018] This invention teaches the robot to learn from doctors' operational strategies for dealing with different patients, environments, and micro-movements through extensive teaching. In practical applications, the robot can autonomously find the optimal scanning angle based on information such as the patient's current mouth opening, oral environment, and scanning body position, and adapt to the patient's micro-movements in real time. Ultimately, it completes the accurate measurement of all scanning bodies, solving the problem of scanning failure or decreased accuracy caused by patient micro-movements in traditional methods.

[0019] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of the present invention, nor is it intended to restrict the scope of the invention. Other features of the invention will become readily apparent from the following description.

[0020] The beneficial effects of this invention are: 1. By learning from a large amount of patient data, the system can autonomously adapt to individual anatomical features such as different mouth openings and dental arch shapes, and generate the optimal scanning path in real time without manual intervention. This not only overcomes the influence of patient differences, but also significantly reduces the average scanning time from 118 seconds to 42 seconds, and the scanning results are highly consistent. 2. During the scanning process, the system monitors the patient's status in real time through an RGB-D camera and an extraoral scanning device. When changes in the patient's status characteristics are detected, such as slight head movements, jaw movements, or tongue or intraoral tissues obscuring the scanning object, the embodied intelligent strategy model can adjust the scanning path in real time to ensure scanning continuity and intelligence. This solves the problem of scanning failure or decreased accuracy caused by slight patient movements in traditional methods. 3. By using an RGB-D camera for real-time ranging and a force sensor for real-time contact monitoring, a dual safety redundancy is formed, achieving a zero collision rate in clinical testing and ensuring the safety of patients and equipment. Attached Figure Description

[0021] The above and other features, advantages, and aspects of the various embodiments of the present invention will become more apparent from the accompanying drawings and the following detailed description. Wherein: Figure 1 A block diagram of an adaptive external scanning docking station robot system based on embodied intelligence according to an embodiment of the present invention is shown; Figure 2 A flowchart illustrating the specific implementation of the teaching and learning phase according to an embodiment of the present invention is shown; Figure 3 A flowchart illustrating a specific implementation of the autonomous scanning phase according to an embodiment of the present invention is shown; Figure 4 A flowchart of an adaptive external scanning connection base robot method based on embodied intelligence according to an embodiment of the present invention is shown. Detailed Implementation

[0022] 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.

[0023] According to an embodiment of the present invention, an adaptive extraoral scanning docking station robot system based on embodied intelligence is proposed. By allowing the robot to learn the operating strategies of doctors in response to different patients, different environments, and different micro-movement situations through extensive teaching, the robot can autonomously find the optimal scanning angle based on information such as the patient's current mouth opening, oral environment, and scanning body position in practical applications, and adapt to the patient's micro-movements in real time, ultimately completing the accurate measurement of all scanning bodies. This solves the problem of scanning failure or decreased accuracy caused by patient micro-movements in traditional methods.

[0024] The principles and spirit of the present invention will be explained in detail below with reference to several representative embodiments.

[0025] Figure 1 This is a block diagram of an adaptive external scanning connection base robot system based on embodied intelligence, according to an embodiment of the present invention. The system includes: Robot body: includes at least one multi-degree-of-freedom robotic arm, which has zero-force dragging function; Force sensing module: includes a six-dimensional force sensor, installed between the end of the robotic arm and the extraoral scanning device, used to detect the triaxial force and triaxial torque of the robotic arm in contact with the oral tissue in real time; Extraoral scanning device: Fixedly mounted on the end effector of the robotic arm, used to acquire high-precision two-dimensional image data of the scanned body inside the oral cavity; Environmental perception module: including an RGB-D camera fixedly mounted on the end effector of the robotic arm, used to acquire three-dimensional point cloud data of the oral cavity environment in real time; Scanning body: rectangular structure, with a threaded interface at the bottom that matches the implant connection abutment, and asymmetrically arranged circular markings printed on all four sides; Control module: Implemented by a high-performance industrial computer, which includes a teaching and learning unit, an autonomous scanning unit, a pose calculation unit and a collision detection unit; the control module is connected to the robotic arm controller, the external scanning device, the RGB-D camera and the six-dimensional force sensor.

[0026] To provide a clearer explanation of the aforementioned adaptive external scanning connection base robot system based on embodied intelligence, a specific embodiment will be described below. However, it is worth noting that this embodiment is only for better illustrating the present invention and does not constitute an improper limitation of the present invention.

[0027] The following example will be used to further illustrate an adaptive external scanning connection base robot system based on embodied intelligence.

[0028] Robot body: includes at least one multi-degree-of-freedom robotic arm, which has zero-force dragging function.

[0029] In this embodiment, a six-degree-of-freedom collaborative robotic arm 100 is used. Its end effector has a connection interface, and each joint of the robotic arm has a built-in high-precision encoder that can provide real-time feedback of joint angles. The end effector's pose in the base coordinate system is obtained through forward kinematics calculation. The robotic arm has a zero-force dragging function, allowing it to be easily dragged by a doctor in teaching mode.

[0030] Force sensing module 200: includes a six-dimensional force sensor, installed between the end of the robotic arm 100 and the extraoral scanning device 300, for real-time detection of the triaxial force and triaxial torque of the robotic arm in contact with oral tissue.

[0031] Extraoral scanning device 300: fixedly installed on the end effector of the robotic arm, used to acquire high-precision two-dimensional image data of the intraoral scan body.

[0032] The device includes at least two cameras and a fill light to obtain the precise pose of the scanned object.

[0033] Environmental perception module 400: includes an RGB-D camera fixedly mounted on the end effector of the robotic arm, used to acquire three-dimensional point cloud data of the oral cavity environment in real time.

[0034] The RGB-D camera uses structured light / ToF principle to output 3D point cloud data in real time at a frame rate of over 30Hz.

[0035] Scanner 500: A cuboid structure with a threaded interface at the bottom that matches the implant connection abutment. Asymmetrically arranged circular markers are printed on all four sides. The geometric dimensions and marker distribution of the scanner are known, allowing the establishment of a local coordinate system for the scanner. The inventor's published patent (publication number: CN115607320A) discloses an extraoral scanning connection abutment pose measuring instrument and pose parameter determination method; therefore, the specific structure and processing method of the extraoral scanning device 300 and scanner 500 will not be elaborated here.

[0036] Control module 600: Implemented by a high-performance industrial computer, it internally includes a teaching and learning unit, an autonomous scanning unit, a pose calculation unit, and a collision detection unit. Control module 600 communicates with the robotic arm controller, the external scanning device 300, the RGB-D camera, and the six-dimensional force sensor.

[0037] The teaching and learning unit is used to record data when doctors drag the robotic arm to perform scanning operations in the manual teaching mode, to build a diverse teaching dataset containing different patient mouth openings, different dental arch shapes, different numbers of scanning bodies, and different scanning angles, and to obtain an embodied intelligent strategy model through imitation learning algorithm training.

[0038] During each teaching session, the control module 600 synchronously acquires multimodal data, including: encoder data for each joint of the robotic arm, high-precision two-dimensional image frame sequences acquired by the extraoral scanning device 300, and three-dimensional point cloud data acquired by the RGB-D camera. State features extracted from the above data include robotic arm pose, patient mouth opening, scanning body pose, distance between the robotic arm end effector and oral tissues, and point cloud data of oral obstructions.

[0039] like Figure 2 As shown, the specific implementation process of the demonstration and learning phase is described in detail.

[0040] The doctor switched the robotic arm to "teaching mode" via the control software. At this time, all joints of the robotic arm are in a state of zero force control and can be easily dragged.

[0041] The doctor drags the robotic arm to perform multiple complete scans of various head models and the oral cavity scans of several different patients. Teaching scenarios cover: Different mouth opening ranges: patients with limited mouth opening (approximately 30mm), patients with normal mouth opening (approximately 50mm), and patients with larger mouth opening (approximately 60mm or more). Different dental arch shapes: normal dental arch, narrow dental arch, wide dental arch; Different numbers of scanning bodies: single implant, multiple implants, edentulous jaw (6-8 scanning bodies); Different scanning angles: scanning paths starting from different points and angles; Each head model and each patient was taught 3-5 times, for a total of no less than 500 teaching sessions, forming a diverse teaching dataset.

[0042] During each teaching session, the control module 600 synchronously acquires the following multimodal data at a frequency of 100Hz: The encoder data of each joint of the robotic arm is used to calculate the pose sequence of the end effector in the base coordinate system after forward kinematics. P t ; High-precision two-dimensional image frame sequence acquired by the extraoral scanning device 300 ; 3D point cloud data acquired by an RGB-D camera; The teaching and learning unit extracts state features from multimodal data. ,include: robotic arm pose : The position of the robotic arm's end effector; Patient's mouth opening Extract the vertical distance between the upper and lower jaws from the RGB-D camera point cloud; Scan body position : Obtain the three-dimensional spatial coordinates of the scanned body within the oral cavity from the pose calculation unit; closest distance The closest distance between the end effector of the robotic arm and the oral tissue; Oral obstruction point cloud Obstacle point cloud information obtained from RGB-D camera point clouds; State characteristics As input, the doctor's next movement command. As output, training samples are constructed. The Generative Adversarial Imitation Learning (GAIL) algorithm is used to train the policy network. GAIL trains the policy network through adversarial training between the generator (policy network) and the discriminator, enabling the policy network to learn the doctor's decision-making logic under different states. After training, the policy network is saved as an embodied intelligent policy model.

[0043] In the imitation learning process, the discriminator in traditional generative adversarial imitation learning is improved by constructing a discriminator network based on a diffusion model to enhance stability and robustness in complex oral environments. To address uncertainties in oral scenarios such as patient micro-movements, soft tissue occlusion, and point cloud noise, state-action pairs are represented as... , where the state This includes the patient's mouth opening, scan posture, three-dimensional oral cavity structure information, and movements. This refers to the control commands for the robotic arm. A diffusion discriminator is constructed based on a diffusion model. By introducing a diffusion denoising process with conditional variables, probabilistic modeling of state-action pairs is performed. The diffusion loss function is defined as: , in, These represent category labels, corresponding to expert-taught data and robot-generated data, respectively. It is Gaussian noise. This is the noise predicted by the diffusion model.

[0044] Based on the above diffusion loss, a diffusion discriminant function is constructed. This is used to determine how closely the current state-action pair approximates the expert teaching distribution, and its expression is: , The output range of the diffusion discriminator is , representing the probability that the state-action pair belongs to the expert-taught behavior. During policy learning, a reward function is constructed based on the diffusion discriminator: , By maximizing the reward function, the robotic arm is driven to generate a scanning trajectory that more closely resembles the doctor's instruction.

[0045] Compared to traditional GAIL discriminators, the diffusion-based discriminator can provide smoother and more stable discrimination signals under complex conditions such as narrow oral cavity space, high-noise point clouds, and patient micro-movements, thereby improving the stability and generalization ability of policy learning.

[0046] The autonomous scanning unit is used to acquire the current state characteristics of the patient's oral environment in real time in autonomous scanning mode, input the state characteristics into the embodied intelligent strategy model, generate an adaptive scanning path that adapts to the current individual characteristics of the patient, and control the robotic arm to carry the extraoral scanning device to scan the scanning body along the adaptive scanning path.

[0047] Specifically, the process of generating the adaptive scan path can be modeled as a sequence decision problem. At time [time value missing]... The system state is represented as , This indicates the current pose of the robotic arm's end effector.

[0048] Embodied intelligent policy model output in state The following action probability distribution: , Among them, actions This represents the pose increment of the robotic arm's end effector in three-dimensional space. The system samples the motion based on this distribution: , Subsequently, the pose of the robotic arm's end effector was updated using the robot's kinematics model:

[0049] in, Let be the positive kinematics function of the robotic arm. The motion sequence is obtained by performing multi-step recursion on the above process within a closed-loop control framework. And further generate the corresponding robotic arm end effector trajectory. This forms a continuous adaptive scanning path.

[0050] like Figure 3As shown, the specific implementation process of the autonomous scanning phase is described in detail.

[0051] The doctor switched the robotic arm to "autonomous mode" using the control software.

[0052] The system first controls the robotic arm to move to a safe "observation position," and then uses an RGB-D camera and an extraoral scanning device 300 to acquire a 3D point cloud model and initial 2D image frames of the current patient's oral cavity environment, extracting initial state features: the current position of the robotic arm's end effector. P 0; Current patient's mouth opening O 0; the three-dimensional spatial position of each scanning body within the oral cavity (Not all scanning rods); closest distance D 0: The closest distance between the robotic arm's end effector and oral tissues; distribution of obstacles within the oral cavity. C 0 (adjacent teeth, soft tissue, etc.).

[0053] The current state characteristics are input into the embodied intelligent strategy model, and the model outputs an initial scanning path adapted to the individual characteristics of the current patient. Path 0.

[0054] The robotic arm began along Path 0 motion. During motion, the system performs the following closed-loop control at a frequency of 200Hz: The RGB-D camera and the external scanning device acquire and calculate the updated 3D point cloud data and the 3D spatial pose of the scanned volume in real time, and extract the current state features. State t .

[0055] When patient micro-movements (such as head movement, jaw movement, tongue or tissue obscuring the scan area) are detected that cause changes in state characteristics, the updated state characteristics will be used. State t The embodied intelligent strategy model is input, and the model outputs adjusted motion commands in real time to achieve adaptive adjustment during the scanning process.

[0056] The collision detection unit calculates the closest distance between the robotic arm and the oral tissue in real time. d t ,like d t If the path is less than 5mm, an avoidance mechanism is triggered, and the Dynamic Motion Primitives (DMP) method is used to replan the remaining path online.

[0057] The system monitors data from the six-dimensional force sensor in real time. If the contact force exceeds 1N, it will execute an emergency stop and retract.

[0058] The system automatically acquires the scanning progress of the external scanning device on the scanned object, and finally completes the scanning of all scanning rods completely and autonomously.

[0059] The pose calculation unit is used to calculate the precise pose of all scanned objects step by step from multiple angles based on the multi-view high-precision two-dimensional images acquired by the extraoral scanning device.

[0060] The collision detection unit calculates the spatial distance between the robotic arm and the oral tissue in real time based on the 3D point cloud data collected by the environmental perception module. When the distance is less than a preset safety threshold, an obstacle avoidance command is generated. Simultaneously, a force sensor monitors the contact force in real time; if it exceeds a set threshold, an emergency stop and retraction are executed. This achieves dual safety redundancy, resulting in a 0% collision rate in clinical testing.

[0061] The distance between the robotic arm and the oral tissue was calculated using the following method: The 3D point cloud data collected by the environmental perception module is preprocessed, including noise reduction, voxel filtering and coordinate system calibration, and the point cloud data is uniformly converted to the robot arm base coordinate system to eliminate the external parameter error between the sensor and the robot arm.

[0062] The processed point cloud data is segmented to extract the effective point set representing the surface of oral soft tissue, and non-target area point clouds (such as instruments, background structures, etc.) are removed to improve the accuracy and real-time performance of distance calculation.

[0063] A spatial index structure is constructed to accelerate nearest neighbor search. Based on this, for the robotic arm end effector, the minimum Euclidean distance to the oral tissue point cloud set is calculated point by point to obtain the current spatial minimum interval.

[0064] Specifically, for robotic arms Its point cloud set to oral tissue The minimum distance is defined as: .

[0065] This embodiment illustrates the system's adaptive capability in response to subtle patient movements through a typical scenario.

[0066] Scenario description: During the scan, the patient experienced slight jaw movement due to swallowing, causing the mouth opening to decrease from 50mm to 45mm, while the tongue partially obscured the scanning area.

[0067] System response: The RGB-D camera monitors in real time at a frequency of 30Hz and detects changes in state characteristics within the next sampling cycle (approximately 33ms) after a micro-motion occurs.

[0068] The updated state features (mouth opening 45m, obstacle point cloud) are input into the embodied intelligent policy model, and the model outputs the adjusted motion command within about 5ms.

[0069] According to the adjustment instructions, the robotic arm adjusts the scanning angle from the originally planned large angle to an angle that adapts to the new opening, and at the same time corrects the scanning path to align with the unobstructed surface of the object being scanned.

[0070] The entire process is completed in about 50ms, with almost no sensation from the patient, and the scan data is continuous and complete, requiring no restart.

[0071] This embodiment also describes in detail the specific implementation of the collision detection unit.

[0072] The collision detection unit performs the following steps at a frequency of 50Hz: The system receives raw 3D point cloud data acquired by an RGB-D camera 410, performs voxel filtering downsampling, removes outliers, and obtains a denoised point cloud.

[0073] Based on the current pose of the robotic arm, extract the point cloud within a 50mm radius around the end effector (including the external scanning device 200) of the robotic arm as the region of interest for collision detection. C ROI .

[0074] Calculate the key points (predefined collision detection points) on the end effector of the robotic arm. C ROI Minimum distance of each point in d min .

[0075] Security assessment and response: If d min >100mm, normal movement; if 50mm <d min <100mm, decelerate, reducing the current speed by 50%; if d min If the distance is less than 50mm, obstacle avoidance is triggered, the current motion is paused, and a new local path is generated by calling the dynamic motion primitive method.

[0076] Once visual obstacle avoidance is triggered, data from the six-dimensional force sensor is monitored simultaneously. If the force sensor detects a contact force greater than 1N, the system will further decelerate or retract to ensure absolute safety.

[0077] Through the aforementioned dual safety protection mechanisms, this system can achieve zero-collision autonomous scanning.

[0078] Monitoring module 700: Connected to control module 600, it is used to monitor the digital twin model of the robotic arm body during teaching and autonomous scanning, the images acquired in real time by the external scanning device, the point cloud information acquired by the RGB-D camera, and the real-time scanning of the three-dimensional model of the body posture.

[0079] To verify the technical effect of the present invention, five edentulous head models were selected (six scanning bodies were installed in each head model), and each model was scanned four times. Micro-motion and scanning body occlusion interference were introduced during the scanning process. The following three methods were used to obtain the pose data of the connecting abutment: Group A: Doctors using handheld scanners; Group B: Preset path robotic arm scanning (the robotic arm moves along a fixed trajectory without adaptive adjustment, and uses an independent RGB-D camera for environmental perception). Group C: The system of this invention.

[0080]

[0081] Experimental results show that the system of this invention outperforms existing technologies in terms of scanning efficiency, measurement accuracy, completion rate, and safety. In particular, through functional division of labor—the external scanning device focuses on high-precision pose measurement, while the RGB-D camera focuses on real-time environmental sensing—efficient, accurate, and safe autonomous scanning is achieved. Through teaching and dynamic adjustment, zero-collision autonomous scanning is achieved while maintaining extremely high consistency and accuracy.

[0082] Based on the same inventive concept, this invention also proposes an adaptive external scanning connection base robot method based on embodied intelligence, such as... Figure 4 As shown, the method includes: Step S01: Place the robotic arm equipped with an extraoral scanning device and an RGB-D camera into teaching mode; Step S02: The doctor drags the robotic arm to perform multiple complete scans on various head models and the scanning bodies inside the mouths of multiple different patients. The doctor synchronously records the multimodal data during each scan at a preset frequency and extracts the state features from it. Step S03: Using the state features as input and the doctor's next-move instruction as output, construct a training sample set and train a deep neural network model using an imitation learning algorithm as an embodied intelligence strategy model. Step S04: Place the robotic arm in autonomous scanning mode, collect and extract current state features through an RGB-D camera and an extraoral scanning device, input the current state features into the embodied intelligent strategy model, and generate an initial adaptive scanning path that adapts to the individual characteristics of the current patient. Step S05: Control the robotic arm to carry the external scanning device to move along the adaptive scanning path, and perform closed-loop control at a preset frequency during the movement until all scanning body poses are completed; Closed-loop control includes: Status characteristics are updated in real time via an RGB-D camera and an external scanning device; When a change in patient status characteristics is detected due to slight movement, the updated status characteristics are input into the embodied intelligent strategy model to adjust the scanning path in real time. Force is monitored in real time using an RGB-D camera and a six-dimensional force sensor. When the detection distance and contact force exceed the threshold, an emergency stop or retraction is executed.

[0083] While the spirit and principles of the invention have been described with reference to several specific embodiments, it should be understood that the invention is not limited to the disclosed specific embodiments, and the division of aspects does not imply that features in these aspects cannot be combined for benefit; such division is merely for ease of description. The invention is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

[0084] Regarding the limitation of the scope of protection of this invention, those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solution of this invention are still within the scope of protection of this invention.

Claims

1. An adaptive external scanning connection base robot system based on embodied intelligence, characterized in that, The system includes: Robot body: includes at least one multi-degree-of-freedom robotic arm, which has zero-force dragging function; Force sensing module: includes a six-dimensional force sensor, installed between the end of the robotic arm and the extraoral scanning device, used to detect the triaxial force and triaxial torque of the robotic arm in contact with the oral tissue in real time; Extraoral scanning device: Fixedly mounted on the end effector of the robotic arm, used to acquire high-precision two-dimensional image data of the scanned body inside the oral cavity; Environmental perception module: including an RGB-D camera fixedly mounted on the end effector of the robotic arm, used to acquire three-dimensional point cloud data of the oral cavity environment in real time; Scanning body: rectangular structure, with a threaded interface at the bottom that matches the implant connection abutment, and asymmetrically arranged circular markings printed on all four sides; Control module: Implemented by a high-performance industrial computer, which includes a teaching and learning unit, an autonomous scanning unit, a pose calculation unit and a collision detection unit; the control module is connected to the robotic arm controller, the external scanning device, the RGB-D camera and the six-dimensional force sensor.

2. The adaptive external scanning connection base robot system based on embodied intelligence according to claim 1, characterized in that, The robotic arm has a connection interface at its end and each joint of the robotic arm has a built-in high-precision encoder. The position and pose of the end in the base coordinate system are obtained through forward kinematics calculation.

3. The adaptive external scanning connection base robot system based on embodied intelligence according to claim 1, characterized in that, The specific steps of the teaching and learning unit are as follows: The doctor switched the robotic arm to teaching mode through the control software, so that all joints of the robotic arm were in a state of zero force control. The doctor drags the robotic arm to perform multiple complete scans on various head models and the scanning objects inside the mouths of several different patients, collecting data on different mouth openings, different dental arch shapes, different numbers of scanning objects, and different scanning angles. The control module 600 synchronously acquires multimodal data at a frequency of 100Hz: encoder data from each joint of the robotic arm, which, after forward kinematics calculation, yields the pose sequence of the end effector in the base coordinate system. High-precision two-dimensional image frame sequence acquired by the external scanning device 300 3D point cloud data acquired by an RGB-D camera; Extracting state features from multimodal data Including: robotic arm pose Patient's mouth opening Scan body position closest distance Oral obstruction point cloud ; State characteristics As input, the doctor's next movement command. As output, training samples are constructed; An improved generative adversarial imitation learning algorithm is used to train the policy network; after training, the policy network is saved as an embodied intelligent policy model.

4. The adaptive external scanning connection base robot system based on embodied intelligence according to claim 3, characterized in that, The improved generative adversarial imitation learning algorithm is specifically as follows: A discriminator network is constructed based on a diffusion model, representing state-action pairs as... , where the state This includes the patient's mouth opening, scan posture, three-dimensional oral cavity structure information, and movements. These are the control commands for the robotic arm; Its diffusion loss function is defined as: ， in, These represent category labels, corresponding to expert-taught data and robot-generated data, respectively. It is Gaussian noise. The noise predicted by the diffusion model; Construct a diffusion discriminant function based on diffusion loss. : ， The output range of the diffusion discriminator is , representing the probability that the state-action pair belongs to the expert-taught behavior; during policy learning, a reward function is constructed based on the diffusion discriminator: ， By maximizing the reward function, the robotic arm is driven to generate a scanning trajectory that more closely resembles the doctor's instruction.

5. The adaptive external scanning connection base robot system based on embodied intelligence according to claim 1, characterized in that, The autonomous scanning unit is used to acquire the current state characteristics of the patient's oral environment in real time in autonomous scanning mode, input the state characteristics into the embodied intelligent strategy model, generate an adaptive scanning path that adapts to the current individual characteristics of the patient, and control the robotic arm to carry the extraoral scanning device to scan the scanning body along the adaptive scanning path.

6. The adaptive external scanning connection base robot system based on embodied intelligence according to claim 5, characterized in that, The process of generating the adaptive scan path is modeled as a sequence decision problem: at time... The system state is represented as , Indicates the current pose of the robotic arm's end effector; Embodied intelligent policy model output in state The following action probability distribution: ， Among them, actions This represents the pose increment of the robotic arm's end effector in three-dimensional space; Actions are sampled based on this distribution: ， The pose of the robotic arm's end effector is updated using a robot kinematics model. in, Given the positive kinematics function of the robotic arm, the motion sequence is obtained by performing multi-step recursion on the above process within a closed-loop control framework. And further generate the corresponding robotic arm end effector trajectory. This forms a continuous adaptive scanning path.

7. The adaptive external scanning connection base robot system based on embodied intelligence according to claim 1, characterized in that, The pose calculation unit is used to calculate the precise pose of all scanned objects step by step from multiple angles based on the multi-view high-precision two-dimensional images acquired by the extraoral scanning device.

8. The adaptive external scanning connection base robot system based on embodied intelligence according to claim 1, characterized in that, The collision detection unit is used to calculate the spatial distance between the robotic arm and the oral tissue in real time based on the three-dimensional point cloud data collected by the environmental perception module. When the distance is less than a preset safety threshold, an obstacle avoidance command is generated. The force sensor monitors the contact force in real time. If it is greater than the set threshold, an emergency stop is executed and the arm retracts.

9. The adaptive external scanning connection base robot system based on embodied intelligence according to claim 1, characterized in that, The calculation process for the spatial distance between the robotic arm and the oral tissue is as follows: The 3D point cloud data collected by the environmental perception module is preprocessed, including noise reduction, voxel filtering and coordinate system calibration, and the point cloud data is uniformly converted to the robot arm base coordinate system. The processed point cloud data is segmented, the effective point set representing the surface of oral soft tissue is extracted, and the point cloud of non-target areas is removed. Construct a spatial index structure to accelerate nearest neighbor search; For the robotic arm's end effector, the minimum Euclidean distance to the oral tissue point cloud set is calculated point by point to obtain the current minimum spatial interval: for the robotic arm Its point cloud set to oral tissue The minimum distance is defined as: 。 10. A method for an adaptive external scanning connection base robot based on embodied intelligence, characterized in that, Using an adaptive extraoral scanning connection base robot system based on embodied intelligence as described in any one of claims 1-9, the method includes: Step S01: Place the robotic arm equipped with an extraoral scanning device and an RGB-D camera into teaching mode; Step S02: The doctor drags the robotic arm to perform multiple complete scans on various head models and the scanning bodies inside the mouths of multiple different patients. The doctor synchronously records the multimodal data during each scan at a preset frequency and extracts the state features from it. Step S03: Using the state features as input and the doctor's next-move instruction as output, construct a training sample set and train a deep neural network model using an imitation learning algorithm as an embodied intelligence strategy model. Step S04: Place the robotic arm in autonomous scanning mode, collect and extract current state features through an RGB-D camera and an extraoral scanning device, input the current state features into the embodied intelligent strategy model, and generate an initial adaptive scanning path that adapts to the individual characteristics of the current patient. Step S05: Control the robotic arm to carry the external scanning device to move along the adaptive scanning path, and perform closed-loop control at a preset frequency during the movement until all scanning body poses are completed.

Images