Robot grabbing and positioning control method and system

Abstract

The invention relates to the technical field of robot grabbing and positioning control, in particular to a robot grabbing and positioning control method and system, and the method comprises the following steps: obtaining the actual microscopic position and posture data of a robot end effector relative to a target object; calculating an instantaneous deviation value between the robot end effector and the target object according to the actual microscopic position and attitude data and a theoretical expected position and attitude of the robot end effector; when the instantaneous deviation value exceeds a set tolerance range, generating a compensation motion instruction according to the instantaneous deviation value; according to the compensation motion instruction, a robot end effector is driven to carry out micro-attitude correction, so that the robot end effector is aligned with the target object; and after the robot end effector completes micro attitude correction and is aligned with the target object, grabbing operation is executed. And the grabbing and positioning precision and stability can be improved.

Description

Technical Field

[0001] This invention relates to the technical field of robot grasping and positioning control, and specifically to a robot grasping and positioning control method and system. Background Technology

[0002] In modern industrial production, automated robot systems play a central role in the high-speed, continuous grasping and placement of small, complex, or sensitive objects. These systems are typically equipped with advanced sensing and actuation components to achieve extremely high positioning accuracy and flexible operation. However, in the long-term, high-intensity industrial environment, even well-designed robots cannot avoid slight performance degradation of their internal components due to continuous wear and tear or environmental influences. For example, after prolonged operation, the position feedback signal of the servo motor encoder of a robot joint may exhibit slight and intermittent jitter. This jitter may be within the system's preset noise tolerance range in most cases, but in high-frequency motion modes, it may occasionally exceed the preset threshold. This causes the servo drive, attempting to accurately follow the motion trajectory, to misinterpret the noisy signal as an actual position deviation, resulting in excessive or unnecessary correction movements. Consequently, the actual output position of the robot joint exhibits high-frequency oscillations and minute tracking errors at the microscopic level. Summary of the Invention

[0003] The purpose of this invention is to address the aforementioned shortcomings by proposing a robot grasping and positioning control method and system.

[0004] The present invention adopts the following technical solution: A robot grasping and positioning control method, the method comprising the following steps: When the robot end effector approaches the target object to a set distance, the actual microscopic position and attitude data of the robot end effector relative to the target object are obtained by the local sensing unit installed on the robot end effector. Based on the actual microscopic position and attitude data, as well as the theoretical expected position and attitude of the robot end effector, calculate the instantaneous deviation between the robot end effector and the target object; When the instantaneous deviation exceeds the set tolerance range, a compensation motion command is generated based on the instantaneous deviation. According to the compensation motion command, drive the robot end effector to perform micro-orientation correction so that the robot end effector is aligned with the target object; After the robot's end effector completes micro-orientation correction and aligns with the target object, it performs the grasping operation.

[0005] This technical solution enables real-time perception, calculation, and correction of microscopic deviations between the robot's end effector and the target object, thereby significantly improving the accuracy and stability of grasping and positioning. It effectively avoids grasping failures or excessive grasping force caused by inaccurate positioning, and solves the problem of decreased reliability in robot grasping and positioning in existing technologies.

[0006] This application also discloses a robot grasping and positioning control system, applied to a robot grasping and positioning control method, the system comprising: The local perception module acquires the actual microscopic position and attitude data of the robot end effector relative to the target object when the robot end effector approaches the target object to a set distance. The deviation calculation module calculates the instantaneous deviation between the robot end effector and the target object based on the actual microscopic position and attitude data, as well as the theoretical expected position and attitude of the robot end effector. The instruction generation module generates a compensation motion instruction based on the instantaneous deviation when the instantaneous deviation exceeds the set tolerance range. The correction module drives the robot's end effector to perform micro-orientation correction according to the compensation motion command, so as to align the robot's end effector with the target object; The grasping execution module performs the grasping operation after the robot's end effector completes micro-orientation correction and aligns with the target object.

[0007] Through modular design, various functions can be efficiently integrated, providing hardware and software support for robot grasping and positioning, thereby achieving high-precision and high-stability grasping operations.

[0008] This application enables higher target alignment, significantly improving the stability of the grasping operation and avoiding the risk of objects slipping or changing posture during grasping or handling. It increases the success rate and stability of robot grasping and positioning, optimizes production cycle time, reduces operating costs, and brings significant economic benefits and technological advantages to automated production lines.

[0009] To further understand the features and technical content of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are for reference and illustration only and are not intended to limit the present invention. Attached Figure Description

[0010] Figure 1 This is a flowchart of a robot grasping and positioning control method according to the present invention; Figure 2 This is a schematic diagram of the structure of a robot grasping and positioning control system according to the present invention. Detailed Implementation

[0011] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention. Furthermore, the accompanying drawings of the present invention are for simple illustrative purposes only and are not depictions of actual dimensions; this is stated in advance. The following embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the scope of protection of the present invention.

[0012] This embodiment provides a robot grasping and positioning control method and system, combined with Figure 1 and Figure 2 As shown.

[0013] refer to Figure 1 A robot grasping and positioning control method, the method includes the following steps: When the robot end effector approaches the target object to a set distance, the actual microscopic position and attitude data of the robot end effector relative to the target object are obtained by the local sensing unit installed on the robot end effector. Based on the actual microscopic position and attitude data, as well as the theoretical expected position and attitude of the robot end effector, calculate the instantaneous deviation between the robot end effector and the target object; When the instantaneous deviation exceeds the set tolerance range, a compensation motion command is generated based on the instantaneous deviation. According to the compensation motion command, drive the robot end effector to perform micro-orientation correction so that the robot end effector is aligned with the target object; After the robot's end effector completes micro-orientation correction and aligns with the target object, it performs the grasping operation.

[0014] "Robot end effector" refers to the part at the very end of a robot arm that directly contacts or manipulates a target object, such as a robotic gripper, suction cup, or tool. Its theoretical expected position and orientation refer to the precise spatial position and orientation that the robot end effector should achieve under ideal conditions, according to a pre-defined task plan.

[0015] "Local sensing unit" refers to a sensor module installed on the end effector of a robot for near-range, high-precision perception of target objects or environmental features, such as miniature vision sensors, laser rangefinders, and ultrasonic sensors.

[0016] "Actual microscopic position and attitude data" refers to the real spatial position and orientation information of the robot's end effector relative to the target object at the microscopic scale, which is acquired by the local sensing unit. This data usually has high resolution and accuracy.

[0017] "Instantaneous deviation" refers to the instantaneous difference between the actual microscopic position and orientation of the robot end effector and the theoretically expected position and orientation. It quantifies the degree of alignment between the robot end effector and the target object.

[0018] "Setting the tolerance range" refers to the maximum instantaneous deviation that is allowed between the robot's end effector and the target object. If the deviation exceeds this range, it is considered that correction is required.

[0019] "Compensation motion command" refers to motion control command that is calculated based on instantaneous deviation and used to drive the robot's end effector to make minor adjustments.

[0020] "Micro-attitude correction" refers to the fine, small-amplitude position and attitude adjustments made by the robot's end effector after receiving a compensation motion command, in order to eliminate instantaneous deviations and achieve precise alignment with the target object.

[0021] "Grasping operation" refers to the action of picking up an object by gripping, adsorption, or other means after the robot's end effector has completed micro-orientation correction and aligned with the target object.

[0022] The robot grasping and positioning control method of this application uses a local sensing unit that can be a high-resolution miniature camera. When the robot's end effector is about 5-10 mm away from the target object, the camera captures a local image of the target object and analyzes the feature points in the image using image processing algorithms to calculate the actual microscopic position and orientation of the robot's end effector relative to the target object. Alternatively, the local sensing unit can be a laser rangefinder array. By emitting a laser beam and receiving the reflected signal, it accurately measures the distance between the robot's end effector and various points on the surface of the target object, thereby constructing a local three-dimensional point cloud of the target object and extracting the actual microscopic position and orientation data from it.

[0023] After acquiring the actual microscopic position and attitude data, the system compares this data with the pre-defined theoretical expected position and attitude. The theoretical expected position and attitude are typically generated by the task planning system based on the CAD model of the target object and the grasping point information. By calculating the Euclidean distance, angle difference, or quaternion difference between the actual data and the expected data, a six-degree-of-freedom instantaneous deviation can be obtained, which includes all inconsistencies in position and attitude.

[0024] If the calculated instantaneous deviation exceeds a preset tolerance range of 0.1 mm or 0.5 degrees in a certain direction, the system will generate a series of minute joint motion commands based on the magnitude and direction of the deviation, using a preset kinematic model and control algorithm. These commands are designed to guide the robot's end effector to move in the direction that reduces the deviation. Alternatively, the generation of compensating motion commands can also be based on fuzzy control or neural network algorithms. By learning historical deviation data and corresponding corrective actions, it can intelligently generate compensating commands that are more adaptable to complex situations.

[0025] The generated compensating motion commands are sent to the robot controller, which converts these commands into servo motor control signals for each joint of the robot. The servo motors drive the joints to make minute rotations, thereby enabling the robot's end effector to make fine adjustments to its position and orientation. This correction process is usually gradual and may require multiple iterations. After each iteration, the actual microscopic position and orientation data are reacquired, and the instantaneous deviation is recalculated until the deviation falls within a set tolerance range.

[0026] Once the instantaneous deviation is successfully controlled within the set tolerance range, it indicates that the robot's end effector has been precisely aligned with the target object. At this point, the system will trigger the gripping execution module, such as closing the robotic gripper or activating the suction cup, to stably grip the target object.

[0027] The robot grasping and positioning control method of this application introduces a local sensing unit to perform real-time, high-precision sensing of the actual microscopic position and posture of the robot end effector relative to the target object, and calculates the instantaneous deviation based on this, thereby generating a compensating motion command to drive the robot end effector to perform micro-posture correction, and finally achieves precise alignment with the target object.

[0028] This application further proposes a step for generating a compensating motion command based on the instantaneous deviation, including: Install local sensing units and torque sensors on the robot's end effector; When the robot's end effector approaches the target object to the critical contact zone, it performs a micro-pressure probe action; During the micro-pressure probe operation, the optical orientation data of the local sensing unit and the physical torque data of the torque sensor are acquired simultaneously. By comparing the instantaneous pose deviation reported by optical pose data with the trend of contact torque change fed back by physical torque data, the authenticity of the instantaneous pose deviation or the presence of measurement noise can be determined. When the instantaneous pose deviation is determined to be real, a compensation motion command is generated based on the instantaneous pose deviation. When it is determined that the instantaneous pose deviation has measurement noise, the optical pose data with measurement noise is corrected to obtain the corrected optical pose data, and a compensation motion command is generated based on the corrected optical pose data.

[0029] Specifically, a local sensing unit and a torque sensor are installed on the robot's end effector. The local sensing unit acquires the microscopic position and orientation data of the robot's end effector relative to the target object, while the torque sensor senses the potential contact force or torque between the robot's end effector and the target object. The critical contact zone refers to the area where the robot's end effector and the target object are about to make physical contact or have already made slight contact. Micro-probing refers to the robot's end effector making a probing movement towards the target object with a very small force or displacement to trigger or sense slight contact. During the micro-probing action, the optical orientation data from the local sensing unit and the physical torque data from the torque sensor are acquired synchronously, ensuring the temporal correspondence between the two types of sensor data.

[0030] The reliability of optical data can be cross-validated by comparing the instantaneous pose deviation reported by optical pose data with the trend of contact torque changes reported by physical torque data. For example, when optical pose data indicates a significant deviation between the robot's end effector and the target object, and the torque sensor simultaneously reports a significant change in contact torque, this usually indicates that the optical pose deviation is real. Conversely, if the optical pose data indicates a deviation, but the torque sensor does not report a corresponding change in contact torque, it may indicate the presence of measurement noise in the optical pose data. When the instantaneous pose deviation is determined to be real, a compensating motion command is directly generated based on this deviation. When the instantaneous pose deviation is determined to contain measurement noise, the optical pose data containing measurement noise needs to be corrected to eliminate or reduce the influence of the noise, thereby obtaining corrected optical pose data, and a more accurate compensating motion command is generated based on this.

[0031] This application's solution effectively addresses the issues of inaccurate measurements or noise that may occur with a single local sensing unit in complex environments by introducing a torque sensor and micro-pressure probe actions, combined with the synchronous acquisition and comparison of optical pose data and physical torque data. Specifically, when the robot's end effector approaches the target object to the critical contact zone, a micro-pressure probe action is executed, enabling the torque sensor to detect the slight contact force or torque between the robot and the target object. At this time, the synchronously acquired optical pose data and physical torque data provide two independent and complementary information streams. The optical pose data provides high-precision spatial position and attitude information, while the physical torque data provides direct evidence of physical contact. By comparing and analyzing these two types of data, it can be determined whether the instantaneous pose deviation reported by the optical pose data truly reflects the relative positional relationship between the robot's end effector and the target object, or whether it is merely measurement noise caused by environmental interference or sensor noise. This multimodal data fusion and verification mechanism allows for an effective assessment of the reliability of the instantaneous pose deviation before generating compensation motion commands, thereby avoiding corrections based on erroneous data and improving the accuracy and reliability of subsequent micro-pose corrections.

[0032] In some preferred embodiments, it is assumed that a robotic end effector needs to grasp a precision electronic component with a surface featuring minute irregularities. As the end effector approaches the component, the local sensing unit may experience slight fluctuations in optical pose data due to surface reflections or minor shadows, leading to uncertainty in the calculation of instantaneous pose deviations. At this point, by performing a micro-probing motion, the end effector lightly touches the component surface with a minimal force. The torque sensor immediately reports a tiny change in contact torque, while the local sensing unit continues to provide optical pose data. The system synchronously compares these two sets of data: if the deviation reported by the optical pose data matches the trend of contact torque changes reported by the torque sensor (e.g., the optical data shows a leftward deviation, while the torque sensor detects an increase in contact force on the right), the deviation is determined to be real, and a compensating motion command is generated accordingly. Conversely, if the deviation reported by the optical pose data does not match the trend of contact torque changes reported by the torque sensor (e.g., the optical data shows a deviation, but the torque sensor does not detect any change in contact torque), measurement noise is identified in the optical pose data. At this point, the system will correct the optical pose data, for example, through filtering or prediction based on historical data, and then generate compensating motion commands based on the corrected data. In this way, even when there is uncertainty in the optical data, accurate and reliable compensating motion commands can be generated, thereby achieving precise grasping of precision electronic components.

[0033] This application further proposes that the steps for generating compensated motion commands also include: Install non-contact force sensors on the robot's end effector; When the robot's end effector approaches the target object to the non-contact critical zone, it executes a micro-motion command; During the execution of micro-motion commands, the optical posture data of the local sensing unit, the non-contact force data of the non-contact force sensor, and the physical torque data of the torque sensor are acquired simultaneously. By comparing the instantaneous pose deviation reported by optical pose data with the non-contact force change trend fed back by non-contact force sensor and the contact torque change trend fed back by physical torque data, the authenticity of the instantaneous pose deviation or the presence of measurement noise can be determined. When the instantaneous pose deviation is determined to be real, a compensation motion command is generated based on the instantaneous pose deviation. When it is determined that the instantaneous pose deviation has measurement noise, the optical pose data with measurement noise is corrected to obtain the corrected optical pose data, and a compensation motion command is generated based on the corrected optical pose data.

[0034] Specifically, a non-contact force sensor is a sensor capable of sensing the presence of a target object or the weak interaction force between it and a robot end effector without physical contact. Examples include capacitive sensors, ultrasonic sensors, infrared sensors, or vision-based force estimation systems. Its purpose is to provide additional environmental awareness information before the robot end effector comes into contact with the target object, aiding in determining the authenticity of pose deviations. The non-contact critical zone can be understood as the area where the robot end effector is close to the target object but has not yet made physical contact. Within this area, the non-contact force sensor can effectively sense the presence of the target object or its weak influence on the robot end effector. In practical applications, micro-motion commands refer to the small, controlled movements performed by the robot end effector within the non-contact critical zone, such as small displacements or posture adjustments along a preset direction. The purpose is to induce perceptible non-contact force or pose changes between the target object and the robot end effector by introducing controlled micro-movements, thereby providing a dynamic reference for sensor data to more accurately determine the authenticity of instantaneous pose deviations.

[0035] Furthermore, simultaneously acquiring optical pose data from the local sensing unit, non-contact force data from the non-contact force sensor, and physical torque data from the torque sensor means that the system simultaneously collects information streams from different sensors during the execution of micro-motion commands. This multimodal data synchronous acquisition provides comprehensive information for subsequent integrated judgment. Specifically, judging the authenticity of the instantaneous pose deviation or the presence of measurement noise by comparing the instantaneous pose deviation reported by the optical pose data with the trend of non-contact force changes reported by the non-contact force sensor and the trend of contact torque changes reported by the physical torque data involves the system performing cross-validation and fusion analysis of these three different types of data. For example, when the optical pose data reports an instantaneous pose deviation, if the non-contact force data and / or physical torque data also show a trend consistent with the deviation (e.g., non-contact force or contact torque increases with the increase of pose deviation), it can enhance confidence in the authenticity of the pose deviation. Conversely, if the optical pose data reports a deviation, but other sensor data do not show corresponding changes, it may indicate the presence of measurement noise or error in the optical pose data. When the instantaneous pose deviation is determined to be real, a compensating motion command is generated based on the instantaneous pose deviation. This means that the system confirms that the deviation reflected in the optical pose data actually exists and calculates the command used to correct the posture of the robot's end effector. When the instantaneous pose deviation is determined to have measurement noise, the optical pose data containing measurement noise is corrected to obtain corrected optical pose data. A compensating motion command is then generated based on the corrected optical pose data. This means that after the system identifies non-realistic deviations in the optical pose data, it will activate the corresponding correction algorithm (such as filtering, smoothing, or fusion correction based on other sensor data) to eliminate or reduce the influence of noise, thereby obtaining more accurate pose data, and generating a compensating motion command based on this.

[0036] This application's solution involves additionally installing a non-contact force sensor on the robot's end effector. When the robot's end effector approaches the target object to the non-contact critical zone, it executes micro-motion commands. This allows for the simultaneous acquisition of optical pose data from the local sensing unit, non-contact force data from the non-contact force sensor, and physical torque data from the torque sensor, even before physical contact occurs or when contact is extremely weak. The introduction of the non-contact force sensor enables the system to obtain additional mechanical sensing information even when there is a weak interaction between the robot's end effector and the target object, but stable contact has not yet been established. By comprehensively comparing and cross-validating the instantaneous pose deviation reported by the optical pose data with the non-contact force change trend reported by the non-contact force sensor and the contact torque change trend reported by the torque sensor, the authenticity of the instantaneous pose deviation can be determined more comprehensively and robustly. This multimodal data fusion method effectively compensates for the limitations of a single sensor under specific working conditions. Especially before the critical contact zone or under weak contact conditions, non-contact force sensors can provide early warning and auxiliary verification, thereby avoiding misjudgments caused by insensitive torque sensor data or noise in optical data.

[0037] In some preferred embodiments, this application is implemented as follows: Assume a robot end effector needs to grasp a smooth, easily damaged precision electronic component. When the robot end effector approaches the component to the non-contact critical zone (e.g., 5mm to 1mm from the target object), the system initiates a micro-motion command, causing the robot end effector to slowly approach at an extremely low linear velocity (e.g., 0.1mm / s) in a direction perpendicular to the target object's surface, while simultaneously performing minor posture adjustments. During this period, a local sensing unit (e.g., a high-precision vision sensor) continuously acquires optical pose data, a non-contact force sensor (e.g., a capacitive proximity sensor) monitors the weak electric field changes between the robot and the target object in real time, and a torque sensor (e.g., a six-dimensional force sensor) is prepared to sense any very early contact forces that may occur. When the local sensing unit reports a small instantaneous pose deviation, the system immediately cross-checks the data from the non-contact force sensor and the torque sensor. If the non-contact force sensor displays a change in non-contact force signal consistent with the pose deviation (e.g., a regular change in capacitance as the pose deviation increases), and the torque sensor has not detected a significant contact torque, the system determines that the pose deviation is genuine and generates a corresponding compensating motion command. Conversely, if the optical pose data reports a deviation, but neither the non-contact force sensor nor the torque sensor shows a corresponding change, the system determines that the optical pose data contains measurement noise and initiates a correction algorithm to filter the optical pose data. Then, it generates a compensating motion command based on the corrected data. In this way, even in extremely delicate and sensitive grasping tasks, the accuracy of pose deviation judgment can be ensured, thereby achieving high-precision grasping and positioning.

[0038] The steps to determine the authenticity of instantaneous pose deviation or the presence of measurement noise include: Before comparing the instantaneous pose deviation reported by the optical pose data with the non-contact force change trend fed back by the non-contact force sensor and the contact torque change trend fed back by the physical torque data, the optical pose data of the local sensing unit, the non-contact force data of the non-contact force sensor and the physical torque data of the torque sensor are timestamped. By analyzing the timestamp information in the data streams of the local sensing unit, the non-contact force sensor and the torque sensor, the microsecond-level time delay or jitter between the data streams of the local sensing unit, the non-contact force sensor and the torque sensor is identified and compensated, and the timestamped optical pose data, non-contact force data and physical torque data are obtained. Based on the timestamped optical pose data, non-contact force data, and physical torque data, calculate the cross-correlation coefficient or dynamic synchronization index between instantaneous pose deviation, non-contact force change trend, and contact torque change trend; Based on the cross-correlation coefficient or dynamic synchronization index, determine the authenticity of the instantaneous pose deviation or whether there is measurement noise.

[0039] Specifically, timestamp alignment refers to the precise time synchronization of multi-source data streams from local sensing units, non-contact force sensors, and torque sensors. Its purpose is to eliminate or compensate for microsecond-level time delays or jitter caused by differences in sensor sampling frequencies, data transmission delays, or inconsistencies in internal system processing times. In practical applications, high-precision timestamps can be embedded in each sensor data packet, and these timestamps can be used for interpolation, resampling, or delay compensation algorithms before data fusion to ensure that all relevant data points correspond precisely on the time axis. For example, linear interpolation or spline interpolation methods can be used to unify data with different sampling rates onto the same time reference.

[0040] Cross-correlation coefficients, or dynamic synchronization indices, can be understood as statistical measures of the similarity or synchronization degree between different data sequences. Cross-correlation coefficients quantify the linear correlation between two signals at different time lags; the closer the value is to 1 or -1, the stronger the correlation. Dynamic synchronization indices capture synchronization behavior in nonlinear or time-varying systems, for example, through phase synchronization, generalized synchronization, or lag synchronization methods. Their purpose is to provide an objective, quantitative basis for judging the authenticity of pose deviations by quantifying the intrinsic relationship between instantaneous pose deviations and changes in physical forces / torques. If the pose deviation and changes in physical forces / torques are highly synchronized in time and have a reasonable physical correlation, the pose deviation is considered real; conversely, if the synchronization is poor or there is no physical correlation, measurement noise may exist.

[0041] This application's solution addresses the inconsistency in the time dimension of multi-source sensor data by introducing timestamp alignment processing. By accurately identifying and compensating for microsecond-level time delays or jitter between the data streams of local sensing units, non-contact force sensors, and torque sensors, it ensures that all data points represent the same physical state at the same moment during subsequent comparison and analysis. This avoids misjudgments caused by time asynchrony, enabling accurate and meaningful comparisons between the instantaneous pose deviation reported by the optical pose data and the non-contact force change trend fed back by the non-contact force sensor and the contact torque change trend fed back by the physical torque data.

[0042] Based on this, by calculating the cross-correlation coefficients or dynamic synchronization indices between instantaneous pose deviation, non-contact force variation trends, and contact torque variation trends, this application can quantify the intrinsic correlation between these different physical quantities. When a robot end effector experiences a real, minute pose deviation from the target object, it is usually accompanied by detectable changes in non-contact force or contact torque, and these changes should be highly synchronized in time and have a physical causal relationship. Cross-correlation coefficients or dynamic synchronization indices can capture this synchronicity and correlation, thus providing an objective, quantitative basis for judging the authenticity of the instantaneous pose deviation. For example, if the pose change indicated by optical pose data and the force / torque change indicated by the force / torque sensor are highly synchronized in time and in the same direction, it can be confirmed that the pose deviation is a real physical phenomenon; conversely, if there is a lack of synchronicity or correlation between the two, the pose deviation is likely caused by measurement noise.

[0043] In some preferred embodiments, it is assumed that a robotic end effector is approaching a precision electronic component for grasping. During the execution of micro-motion commands, a local sensing unit continuously outputs optical pose data, a non-contact force sensor provides non-contact force data on the component's surface, and a torque sensor monitors the minute torques that the end effector may experience. Because these three sensors may have different data refresh rates and internal processing delays, their data streams may have microsecond-level time deviations when unprocessed.

[0044] To accurately determine the accuracy of instantaneous pose deviations in optical pose data reports, the three types of data are first timestamped. For example, each sensor's data packet includes a high-precision hardware timestamp. In the data fusion module, by analyzing these timestamps, it is identified that the data stream from the local sensing unit has a 50-microsecond delay relative to the torque sensor's data stream, while the data stream from the non-contact force sensor has a 30-microsecond lead. Based on these delay and lead information, the system resamples and interpolates the data, aligning all data points to a unified time axis to obtain timestamp-aligned optical pose data, non-contact force data, and physical torque data.

[0045] Subsequently, based on the timestamp-aligned data, the cross-correlation coefficients between instantaneous pose deviations and the trends of non-contact force and contact torque changes are calculated. For example, if the optical pose data reports a small displacement of the end effector in a certain direction, and at the same time point, the non-contact force sensor detects a significant increase in the non-contact force in that direction, while the torque sensor detects a corresponding small torque change, and the cross-correlation coefficient between these changes reaches 0.9 or higher (indicating a high positive correlation), then the system determines that the instantaneous pose deviation is real. Conversely, if the optical pose data reports a pose deviation, but the non-contact force or torque data do not change synchronously, or the cross-correlation coefficient is very low (e.g., below 0.3), then the pose deviation is likely caused by measurement noise and requires correction. In this way, the system can reliably identify the true pose deviation, thereby generating accurate compensation motion commands to ensure that the robot's end effector can accurately align with the target object.

[0046] This application further proposes a step for correcting optical pose data containing measurement noise to obtain corrected optical pose data, including: The monitoring results are obtained by monitoring external environmental parameters, surface characteristics of target objects, or signal quality of local sensing units. Based on the monitoring results, the correction algorithm parameters are dynamically adjusted to obtain the adjusted correction algorithm parameters. Based on the adjusted correction algorithm parameters, the optical pose data containing measurement noise is corrected to obtain the corrected optical pose data.

[0047] Specifically, external environmental parameters may include, but are not limited to, ambient light intensity, temperature, humidity, and airborne particulate matter concentration. These parameters can be acquired in real time by environmental sensors integrated into or near the robot's end effector. Target object surface characteristics may include reflectivity, roughness, color, and texture, which can be evaluated using the optical imaging or spectral analysis functions of the local sensing unit. The signal quality of the local sensing unit refers to its signal-to-noise ratio, image sharpness, and data integrity, which can be obtained by analyzing the raw sensor output data. The monitoring results are a quantitative representation of the current environment and sensor status, derived from a comprehensive evaluation of the above parameters. For example, when the light intensity is too high or too low, or when the target object surface is too smooth or too rough, the optical pose data acquired by the local sensing unit may be more susceptible to noise interference. A decrease in the signal quality of the local sensing unit also indicates that its output data may contain more noise.

[0048] Furthermore, dynamically adjusting the calibration algorithm parameters refers to modifying the internal parameters of the algorithm used to calibrate optical pose data in real time based on the monitoring results. These parameters may include, but are not limited to, the type of filter (e.g., mean filter, Gaussian filter, median filter), the size of the filter window, the threshold, the number of iterations, and the weighting coefficients. For example, when a large change in ambient light intensity is detected, the cutoff frequency or gain of the filter can be adjusted; when the surface texture of the target object is complex, the size of the filter window can be reduced to retain more details; when the signal-to-noise ratio of the local sensing unit decreases, the filtering intensity can be increased. Through this dynamic adjustment, the calibration algorithm can better adapt to the current working conditions, thereby improving the accuracy and robustness of the calibration.

[0049] This application's solution addresses the problem of poor calibration performance in complex and variable environments caused by traditional fixed-parameter calibration algorithms. This is achieved by monitoring external environmental parameters, target object surface characteristics, and the signal quality of the local sensing unit, and dynamically adjusting the calibration algorithm parameters based on the monitoring results. Since the generation mechanism and characteristics of measurement noise are not static but closely related to multiple factors, real-time acquisition of information on these influencing factors and adaptive adjustment of the calibration algorithm accordingly allows for more precise optimization of the calibration process based on the current situation. For example, when changes in ambient light cause specific types of noise in the optical data of the local sensing unit, the system can identify this change and automatically switch to or adjust to filtering parameters more suitable for handling that type of noise, rather than blindly applying a universal set of parameters. This adaptive mechanism ensures that the calibration algorithm is always in an optimal or near-optimal operating state, thereby effectively improving the suppression of measurement noise.

[0050] In some preferred embodiments, it is assumed that the robot needs to pick up various types of electronic components on different production lines. These components may have different surface gloss levels (e.g., matte plastic casings, highly reflective metal surfaces) and colors, and the ambient lighting conditions on the production lines may also vary over time or shifts.

[0051] Before or during the grasping task, the system continuously monitors the following parameters: External environmental parameters: The ambient light sensor mounted on the robotic arm acquires the real-time light intensity of the current working area. Target object surface characteristics: While acquiring pose data, the local sensing unit also analyzes the reflectivity and texture features of the target object's surface. For example, for highly reflective surfaces, it may detect local overexposure or light spots caused by specular reflection. Signal quality of the local sensing unit: The system evaluates the signal-to-noise ratio and sharpness of the image output by the local sensing unit in real time.

[0052] The system will dynamically adjust the correction algorithm parameters when it detects the following situations: Situation 1: When the ambient light intensity suddenly increases, causing high-frequency noise or local saturation in the photoelectric pose data of the local sensing unit, the system will adjust the filter type of the correction algorithm from mean filtering to Gaussian filtering based on the monitoring results, and increase the standard deviation of the filter (i.e., the filter window size) to more effectively smooth high-frequency noise and suppress the influence of saturation regions. Situation 2: When the robot needs to grasp objects with complex micro-textures (e.g., frosted surfaces), if an excessively large filter window is used, it may misjudge the real texture features as noise and filter them out, resulting in pose information distortion. In this case, the system will dynamically reduce the filter window size and adjust the feature retention threshold based on the analysis results of the target object's surface texture, so as to retain the real geometric features of the object's surface to the maximum extent while removing noise. Scenario 3: When the performance of the internal electronic components of the local sensing unit deteriorates slightly due to prolonged operation or increased ambient temperature, resulting in a lower signal-to-noise ratio of the output signal, the system will dynamically increase the number of iterations of the correction algorithm or adjust the weighting coefficients based on the signal quality monitoring results to enhance the correction strength and thus compensate for the impact of the sensor performance degradation.

[0053] Through the above dynamic adjustments, the correction algorithm can always remain efficient and accurate even under changing environmental and object conditions, ensuring that the corrected optical pose data can truly reflect the microscopic relative relationship between the robot's end effector and the target object, thereby providing reliable input for the subsequent generation of compensation motion commands, and ultimately achieving high-precision robot grasping and positioning control.

[0054] The steps for correcting optical pose data containing measurement noise to obtain corrected optical pose data include: Before performing correction, acquire the microscopic surface feature data of the target object; Based on microscopic surface feature data, identify the microscopic textures or irregular geometric features present on the surface of the target object; Based on the identified micro-textures or irregular geometric features, the filter window size or feature retention threshold of the correction algorithm is dynamically adjusted to avoid misjudging real micro-surface features as measurement noise. Based on the adjusted correction algorithm parameters and the adjusted correction algorithm's filter window size or feature retention threshold, the optical pose data containing measurement noise is corrected to obtain the corrected optical pose data.

[0055] Specifically, acquiring microscopic surface feature data of a target object refers to scanning or photographing the surface of the target object using high-resolution imaging equipment (e.g., miniature cameras, laser scanners, or structured light projectors) to obtain information such as its geometry, texture, and roughness at the micrometer or even nanometer level. This data can include height maps, normal maps, or texture images. Identifying the microscopic textures or irregular geometric features on the target object's surface based on this data can be understood as analyzing the acquired microscopic surface feature data using image processing algorithms (e.g., edge detection, feature point extraction, texture analysis, or surface roughness calculation) to distinguish the inherent microstructures on the target object's surface that are crucial for grasping and positioning, such as tiny grooves, protrusions, particles, or specific surface patterns, from random measurement noise. In practical applications, the filter window size or feature retention threshold of the correction algorithm is dynamically adjusted based on the identified microscopic textures or irregular geometric features. The purpose is to enable the correction algorithm to intelligently distinguish between true features and measurement noise. For example, when fine micro-textures are detected on the surface of a target object, the filter window size can be reduced or the feature retention threshold increased to avoid smoothing out these real textures. Conversely, when the surface is relatively smooth, the filter window size can be appropriately increased or the feature retention threshold decreased to more effectively remove noise. Therefore, based on the adjusted correction algorithm parameters and the adjusted filter window size or feature retention threshold, optical pose data with measurement noise is corrected. This ensures that the correction process removes noise while preserving the true micro-features of the target object's surface to the maximum extent, resulting in more accurate and reliable corrected optical pose data.

[0056] In some preferred embodiments, suppose a robot needs to grasp a precision electronic component with micron-level grooves on its surface. As the robot's end effector approaches the component, the optometry data acquired by the local sensing unit may be affected by ambient light fluctuations or sensor noise, resulting in measurement noise in the data. To correct for this noise, firstly, a high-resolution miniature camera is used to acquire microscopic images of the electronic component's surface. These images contain the precise location and depth information of the grooves, i.e., microscopic surface feature data. Next, image processing algorithms analyze these images to identify these grooves as genuine microscopic texture features. Subsequently, when filtering and correcting the noisy optometry data acquired by the local sensing unit, the correction algorithm dynamically adjusts its filtering window size based on the identified groove features. For example, it reduces the filtering window size in the grooved area or increases the feature retention threshold to ensure that this grooved information is not misjudged as noise and filtered out. In smooth areas outside the grooves, conventional filtering parameters can be used for noise removal. Finally, based on the adaptively adjusted correction algorithm parameters, the optical pose data containing measurement noise was corrected. The resulting optical pose data not only removed noise but also accurately preserved the microscopic marking information on the surface of the electronic component. This high-precision correction data was then used to generate compensating motion commands, driving the robot's end effector to perform micro-pose corrections, enabling it to accurately align and grasp the delicate electronic component, avoiding grasping failure or damage due to loss of detail.

[0057] The steps for correcting optical pose data containing measurement noise to obtain corrected optical pose data include: A dynamic spectral analysis module is integrated into the local sensing unit to obtain the spectral characteristics of the microscopic features of the target object's surface; Update the feature preservation threshold of the correction algorithm based on spectral characteristics; Update the filtering parameters of the correction algorithm based on the spectral characteristics; Based on the adjusted correction algorithm parameters, the updated feature retention threshold, and the updated filtering parameters, and according to the adjusted correction algorithm's filtering window size or feature retention threshold, the optical pose data containing measurement noise is corrected to obtain the corrected optical pose data.

[0058] Specifically, optical pose data typically includes the position and orientation information of the robot's end effector relative to the target object. Position data: usually represented by three components in a Cartesian coordinate system: X, Y, and Z. These components can be treated as independent time-series data, and one-dimensional filtering algorithms can be directly applied. Orientation data: usually represented by quaternions. Euler angles (Roll, Pitch, Yaw) or rotation matrix representation are used. However, Euler angles suffer from gimbal locking, and linear filtering may corrupt their physical meaning; rotation matrices are redundant; quaternions are a commonly used attitude representation, but directly applying linear filtering to their four components may violate their unit quaternion constraint (i.e., ...). ),in, is the scalar part of the quaternion. The three components of the vector part of a quaternion cause distortion of attitude information. Therefore, attitude data correction requires special processing.

[0059] Correspondingly, examples of location data correction algorithms (taking Gaussian filtering and Butterworth filtering as examples) show that various filtering algorithms can be used to correct location data (each component of X, Y, and Z).

[0060] One approach is to use Gaussian filtering (time-domain filtering). Gaussian filtering is a commonly used linear smoothing filter whose weight coefficients follow a Gaussian distribution. It effectively suppresses random noise that follows a normal distribution, while also minimizing edge blurring. Discrete Gaussian function: ;in, Indicates the index of the distance from the center point of the filter window; It is a key parameter of the Gaussian filter, which determines the smoothness and effective range of the filter. Larger filters produce stronger smoothing effects but may lose more detail; conversely, smaller filters produce weaker smoothing effects but retain more detail. Filter calculation: For a data point in a location data sequence... Its corrected value It can be obtained by weighted averaging the points in its neighborhood: ;in, Indicates the sampling time of the location data sequence. This is the relative offset index within the window. The weights are calculated using a Gaussian function and satisfy the following: The filter window size is , The radius of the filtering window; dynamically adjusted: when fine micro-textures or irregular geometric features are detected on the surface of the target object, the size of the filtering window is reduced (i.e., the size of the filtering window is decreased) to avoid misjudging these real features as measurement noise and filtering them out. or reduce To retain more detail, when the signal quality of the local sensing unit is low (e.g., signal-to-noise ratio decreases), the filter window size can be increased or enlarged to remove noise more effectively. To enhance filtering strength, a "feature retention threshold" can be used as an adaptive mechanism. For example, if the data gradient change in a certain region exceeds a preset threshold, the filtering strength in that region can be reduced or edge-preserving filtering can be used to avoid smoothing out true features.

[0061] Secondly, a Butterworth filter (frequency domain filter) is used. The Butterworth filter is a commonly used analog or digital filter characterized by a flat frequency response in the passband and rapid attenuation in the stopband, often used to remove noise within a specific frequency range. Digital Butterworth filter transfer function: ;in, and These are filter coefficients. and This refers to the filter order. These coefficients are determined by the filter type, order, and cutoff frequency or bandwidth. Filter calculation: First, the time series of location data is transformed to the frequency domain using a Discrete Fourier Transform or Fast Fourier Transform. In the frequency domain, the data is multiplied by the filter transfer function. , This is the frequency domain representation of the original location data. The filtered frequency domain data is then converted back to the time domain using an inverse Fourier transform. Dynamic adjustments include: Filter type: Based on the frequency characteristics of the measured noise, a low-pass filter, high-pass filter, or band-pass / band-stop filter can be selected. Order: Determines the steepness of the filter's transition band; the order can be increased when more stringent separation of noise and signal is required. Cutoff frequency or bandwidth: For example, when high-frequency jitter noise is detected, the cutoff frequency of the low-pass filter can be lowered; when spectral characterization reveals material properties associated with noise at a specific frequency, the cutoff frequency can be optimized accordingly.

[0062] Special considerations for attitude data correction: Due to the nonlinear characteristics of attitude data, directly applying the aforementioned linear filtering algorithms may not be suitable. One approach is to convert to rotation vectors or angular velocity filtering: Quaternions can be converted to rotation vectors (represented by axis angles) or their angular velocities can be calculated. These quantities are then filtered, and then integrated or converted back to quaternions. Rotation vectors and angular velocities can be approximated as linear within a small range, thus linear filtering can be applied. Another approach is state estimation-based filtering: For high-precision attitude correction, extended Kalman filters or unscented Kalman filters are typically used. These filters can handle nonlinear system models and measurement noise, providing optimal attitude estimates by fusing sensor measurements (such as optical pose data) and the system motion model through prediction and update steps. Although their internal formulas are complex, their core idea is to use the system dynamic model and measurement model for iterative estimation and adjust the estimated values ​​based on the measurement residuals.

[0063] Integrating a dynamic spectral analysis module into the local sensing unit refers to embedding or connecting a hardware module capable of spectral analysis within the local sensing unit of the robot's end effector. This dynamic spectral analysis module can include a miniature spectrometer, multispectral or hyperspectral imaging sensor, etc., and its main function is to capture the reflection, absorption, or emission spectral information of the target object's surface at different wavelengths. Through this module, the spectral characteristics of the target object's surface microstructure can be obtained; these characteristics directly reflect the object's material composition, surface roughness, coatings, or contaminants, and other physicochemical properties.

[0064] Furthermore, updating the feature retention threshold of the correction algorithm based on spectral characteristics refers to intelligently adjusting the boundary used to distinguish true features from noise during data processing using the acquired spectral data. For example, if spectral analysis shows that a certain micro-region has a specific material spectral fingerprint, and this fingerprint is usually associated with important surface features, the feature retention threshold of that region can be lowered accordingly to ensure that even slight geometric changes can be identified as true features rather than measurement noise. Simultaneously, updating the filtering parameters of the correction algorithm based on spectral characteristics refers to dynamically adjusting the specific parameters of the filters used for smoothing or denoising based on the surface characteristics revealed by the spectral data, such as the filter type (e.g., Gaussian filter, median filter), order, cutoff frequency, or bandwidth. For example, for textured regions with specific spectral responses, filtering parameters more suitable for preserving the texture details can be used, while for regions with uniform spectral responses, stronger filtering can be used to remove random noise.

[0065] In some preferred embodiments, suppose the robot needs to grasp a precision electronic component with a multi-layered composite material surface. The component's surface may contain minute scratches or coating defects that are difficult to discern with the naked eye. These are critical features that need to be accurately identified and avoided, but their geometry may resemble artifacts generated by sensor noise. Traditional geometry-based correction methods may misclassify these critical microscopic features as noise and filter them out, or misclassify noise as features, leading to unnecessary corrections.

[0066] Using the scheme of this application, the dynamic spectral analysis module integrated on the local sensing unit synchronously acquires the spectral characteristics of the surface micro-features of the electronic component when the robot's end effector approaches it. For example, this module can detect subtle but stable differences in the spectral reflectance of the scratched area and the surrounding intact coating area at a specific wavelength, or unique absorption peaks exhibited by the coating defect area. Based on these unique spectral characteristics, the system determines that these geometric "anomalies" are real material or surface structural features, rather than random noise. Subsequently, the feature retention threshold of the correction algorithm is dynamically updated according to this spectral information to ensure that these micro-features with specific spectral fingerprints are not incorrectly filtered out. At the same time, the filtering parameters are also adjusted according to the spectral characteristics, for example, using filter types and parameters that can effectively suppress random noise but retain specific spectral features, thereby obtaining highly accurate corrected optical pose data that retains key micro-features. Finally, the robot can perform micro-pose correction based on this precise pose data and successfully perform grasping operations, avoiding damage to critical areas.

[0067] This application further proposes steps for updating the filter parameters of the correction algorithm, including: Integrate spectral reference standards on the local sensing unit; Before or during the acquisition of the spectral characteristics of the microscopic features of the target object's surface, the spectral reference standard is periodically scanned to obtain the actual spectral response data of the spectral reference standard. The actual spectral response data of the spectral reference standard is compared with the preset reference spectral response data to identify the spectral response deviation of the local sensing unit. Generate spectral response calibration parameters based on the spectral response deviation; The spectral characteristics are corrected using spectral response calibration parameters to obtain the corrected spectral characteristics; The filtering parameters of the correction algorithm are updated based on the corrected spectral characteristics.

[0068] Specifically, integrating a spectral reference standard onto a local sensing unit refers to embedding or attaching a standard material or device with known and stable spectral response characteristics into the physical structure of the local sensing unit. This spectral reference standard can be a white board, a grayscale card, or a calibration plate with specific absorption or reflection peaks. Its purpose is to provide a reliable and repeatable benchmark for the local sensing unit to evaluate its own spectral measurement performance.

[0069] The periodic spectral scanning of the spectral reference standard to obtain its actual spectral response data can be understood as the dynamic spectral analysis module of the local sensing unit performing one or more spectral measurements on the integrated spectral reference standard within a specific time interval during the robot's grasping and positioning control process, or before each task begins. This scanning allows us to obtain the local sensing unit's response to known spectral characteristics in its current operating state, i.e., the actual spectral response data.

[0070] In practical applications, comparing the actual spectral response data of a spectral reference standard with preset baseline spectral response data to identify the spectral response deviation of a local sensing unit involves comparing the data obtained through scanning with the standard spectral response data of the spectral reference standard under ideal conditions. This comparison can be performed using methods such as difference analysis, correlation analysis, or spectral matching algorithms. Its purpose is to quantify the measurement error or drift of the local sensing unit at different wavelengths, thereby identifying its spectral response deviation.

[0071] Furthermore, generating spectral response calibration parameters based on the spectral response deviation refers to calculating a set of parameters to correct subsequent spectral measurement results of the local sensing unit based on the identified deviation. These calibration parameters can be gain factors, offsets, nonlinear correction curves, or matrices, etc., and their purpose is to compensate for the spectral measurement errors of the local sensing unit itself.

[0072] Therefore, correcting the spectral characteristics using spectral response calibration parameters yields corrected spectral characteristics. This involves processing the raw spectral data after the local sensing unit acquires the spectral characteristics of the target object's surface microscopic features, using the previously generated calibration parameters. This correction eliminates or significantly reduces the influence of the local sensing unit's own response bias on the measurement results, thus obtaining corrected spectral characteristics that more closely approximate the true spectral characteristics of the target object.

[0073] Finally, updating the filtering parameters of the correction algorithm based on the corrected spectral characteristics means using the calibrated and corrected, more accurate spectral characteristics as input to adjust the filtering parameters of the correction algorithm. For example, based on the specific texture or material information reflected in the corrected spectral characteristics, the type, order, cutoff frequency, or bandwidth of the filter can be set more precisely to achieve more accurate identification and removal of measurement noise.

[0074] In some preferred embodiments, this application is implemented as follows: Assume that the spectral response curve of the dynamic spectral analysis module of the local sensing unit integrated on the robot's end effector exhibits a slight nonlinear deviation at the time of manufacture. To address this issue, a spectral reference standard made of a specific material is integrated inside or near the local sensing unit. This standard has a known and stable reflectance spectrum in the visible and near-infrared bands. Each time the robot starts or at regular intervals, the system triggers the local sensing unit to perform a rapid spectral scan of the spectral reference standard. For example, the scan acquires the actual reflectance spectral data of the standard in the wavelength range of 400 nm to 1000 nm. Subsequently, this actual reflectance spectral data is compared with the preset reference reflectance spectral data of the standard under ideal conditions. By calculating the root mean square error between the two or using a spectral matching algorithm, the spectral response deviation of the local sensing unit at different wavelengths can be identified, for example, a 5% measurement gain error at a specific wavelength. Based on these identified deviations, the system generates a set of spectral response calibration parameters, such as a polynomial calibration function or a lookup table. When the local sensing unit subsequently acquires the spectral characteristics of the target object's surface micro-features, these raw spectral data are immediately corrected using the calibration parameters to obtain the corrected spectral characteristics. For example, if the raw spectral data is overestimated by the local sensing unit by 5% at a certain wavelength, the correction process will reduce it by 5%. Finally, the filtering parameters of the correction algorithm, such as the cutoff frequency of a Gaussian filter or the window size of a median filter, are dynamically adjusted and updated based on these corrected, more accurate spectral characteristics. For example, if the corrected spectral characteristics show that the target object's surface has very fine micro-texture, the filter's cutoff frequency might be set higher to preserve these details and avoid misclassifying them as noise. In this way, it is ensured that the updating of the filtering parameters is based on true spectral information, thereby improving the correction effect for measurement noise in optical pose data.

[0075] The steps for updating the filter parameters of the correction algorithm include: Integrate multispectral or hyperspectral imaging modules on local sensing units; Using multispectral or hyperspectral imaging modules, spectral data of the microscopic features of the target object's surface can be acquired in multiple discrete or continuous wavelength ranges. Feature extraction is performed on the spectral data to obtain spectral features; Based on spectral characteristics, identify nonlinear, multimodal, or rapid transient change patterns in spectral data; Based on the identified change patterns, the filtering parameters of the correction algorithm are dynamically adjusted and updated. The filtering parameters include the filter type, order, cutoff frequency, or bandwidth. After updating the filtering parameters of the correction algorithm, the corrected optical pose data is subjected to real-time residual analysis through the local sensing unit to obtain the residual analysis results. Based on the residual analysis results, evaluate the correction effect of the correction algorithm after updating the filter parameters; When the evaluated correction effect does not reach the preset accuracy threshold, the filtering parameters of the correction algorithm are iteratively optimized based on the residual analysis results until the residual fluctuation meets the preset threshold.

[0076] A "multispectral or hyperspectral imaging module" refers to a sensor system capable of capturing the spectral information of a target object's reflection, transmission, or emission across multiple specific or continuous wavelength ranges. Multispectral imaging modules typically acquire data within several discrete wide bands, while hyperspectral imaging modules acquire data within hundreds of continuous narrow bands, thus providing richer spectral information. The aim is to obtain a more detailed spectral fingerprint of the target object's surface microstructure, enabling more accurate identification of its physical and chemical properties.

[0077] "Spectral data" refers to a set of data acquired through multispectral or hyperspectral imaging modules that reflects the optical response intensity of a target object's surface at different wavelengths. This data can reveal the microscopic characteristics of the target object's surface, such as material composition, texture, and roughness.

[0078] Feature extraction refers to extracting representative and discriminative information from raw spectral data, such as the shape of the spectral curve, peak positions, absorption valley depths, and slopes. This can be achieved through various signal processing and machine learning methods, including principal component analysis, independent component analysis, and wavelet transform. Its purpose is to reduce the dimensionality of high-dimensional spectral data and highlight key information related to microscopic features.

[0079] "Nonlinear, multimodal, or rapid transient change modes" refer to complex patterns of change in spectral data. Nonlinear modes may manifest as a non-linear relationship between the spectral response and certain physical quantities; multimodal modes may indicate the presence of multiple peaks or valleys in the spectral data, reflecting the presence of various components or structures; rapid transient change modes may correspond to rapid local changes in the microstructure of the target object's surface or environmental conditions. Identifying these modes helps in a deeper understanding of the complexity of the target object's surface.

[0080] "Filtering parameters" refer to the specific configuration of the filter used in the correction algorithm, such as the filter type (e.g., Butterworth filter, Chebyshev filter, Wiener filter, etc.), order (which determines the steepness of the filter), cutoff frequency, or bandwidth (which determines the frequency range that is passed or blocked). The purpose of dynamically adjusting these parameters is to enable the filter to optimally remove measurement noise while preserving the microscopic feature information in the true optical pose data.

[0081] Real-time residual analysis refers to the analysis of the difference between the corrected optical pose data and a reference (such as the original data or more reliable data obtained through other methods) immediately after the correction algorithm updates the filtering parameters. The residual is the deviation between the corrected data and the true or expected value. By analyzing the statistical characteristics of the residuals (such as mean, variance, and fluctuation range), the performance of the correction algorithm can be evaluated.

[0082] "Preset accuracy threshold" refers to the acceptable error range or performance standard set when evaluating the correction effect. When the residual analysis results show that the correction effect has not reached this threshold, it means that there is still room for improvement in the current filtering parameters.

[0083] "Iterative optimization" refers to the process of repeatedly adjusting filter parameters and re-calibrating and evaluating them based on the results of residual analysis. This process continues until the fluctuation of the residuals meets a preset threshold, meaning the calibration effect meets the requirements. This is usually achieved through optimization algorithms (such as gradient descent, genetic algorithms, etc.) to find the optimal combination of filter parameters.

[0084] This application's solution integrates a multispectral or hyperspectral imaging module onto a local sensing unit, enabling the acquisition of richer and more detailed spectral data on the microscopic features of the target object's surface. Given that this spectral data may contain complex nonlinear, multimodal, or rapidly transient change patterns, this application further extracts features from the spectral data and identifies these complex patterns, thereby achieving a more comprehensive and in-depth understanding of the target object's surface microscopic properties. It is precisely through the identification of these complex patterns that the filtering parameters of the correction algorithm can be dynamically adjusted and updated, including the filter type, order, cutoff frequency, or bandwidth, ensuring that the filter accurately adapts to the specific features of the target object's surface. This effectively suppresses measurement noise while preserving the microscopic details in the true optical pose data to the greatest extent possible.

[0085] Building upon this foundation, this application introduces a real-time residual analysis and iterative optimization mechanism. After updating the filtering parameters, real-time residual analysis of the corrected optical pose data allows for immediate evaluation of the current filtering parameter correction effect. If the evaluation result does not reach the preset accuracy threshold, the filtering parameters are iteratively optimized based on the residual analysis results. This closed-loop feedback control mechanism ensures that the filtering parameters are continuously adjusted to the optimal state until the residual fluctuation meets the preset threshold, thereby effectively solving the problem in traditional methods where filtering parameters are difficult to accurately adapt to complex microscopic features, and significantly improving the correction accuracy and stability of optical pose data.

[0086] The key innovations of this application lie in the explanation of iteration, the dynamic adjustment and iterative optimization mechanism, and the dynamic adjustment of the calibration algorithm parameters, which ensure the adaptability and robustness of the calibration process. Firstly, the utilization of monitoring results: external environmental parameters (such as light intensity) affect the noise level of the optical sensor. For example, insufficient lighting may increase image noise, requiring a stronger filter. Secondly, the surface characteristics of the target object (such as micro-texture, irregular geometric features, and spectral characteristics) affect the recognition of true features. When fine textures are identified, the filter window size is reduced or the feature retention threshold is increased to avoid smoothing out these features. Spectral characteristics can further guide the selection of filter parameters; for example, adjusting the filter bandwidth for the spectral response of a specific material. Thirdly, the signal quality of the local sensing unit directly reflects the noise level. Poor signal quality requires stronger filtering. Fourthly, spectral response calibration parameters: after calibration using spectral reference standards, the corrected spectral characteristics more accurately reflect the target object, thus providing a more reliable basis for updating the filter parameters.

[0087] Regarding real-time residual analysis and iterative optimization: After initial correction, real-time residual analysis is performed on the corrected optical pose data using a local sensing unit. Residual refers to the difference between the corrected data and a reference (e.g., the original data or more reliable data obtained through fusion with other sensors). Residual calculation: For example, calculating the root mean square error or maximum deviation between the corrected pose data and the uncorrected original data. Evaluation of correction effect: Comparing the residual analysis results with a preset accuracy threshold. Iterative optimization: If the evaluated correction effect does not meet the preset accuracy threshold, the filtering parameters of the correction algorithm (such as filter type, order, cutoff frequency, or bandwidth) are iteratively optimized based on the residual analysis results (e.g., frequency components, amplitude, distribution characteristics of the residuals). This process continues until the residual fluctuation meets the preset threshold, ensuring that the corrected optical pose data achieves the required accuracy. For example, if significant high-frequency components still exist in the residuals, it may be necessary to further reduce the cutoff frequency of the low-pass filter or increase its order.

[0088] As one specific implementation, suppose a robot end effector needs to grasp a precision electronic component with a complex micro-texture on its surface. When the robot end effector approaches the target object to the non-contact critical zone, the multispectral imaging module integrated on the local sensing unit starts to work, acquiring spectral data of the electronic component's surface in the visible and near-infrared bands (e.g., 450nm, 550nm, 650nm, 800nm, 950nm).

[0089] First, feature extraction is performed on the acquired multispectral data, such as by calculating the ratios between different bands and the normalized difference vegetation index, to obtain spectral features that reflect surface texture and material properties. Then, based on these spectral features, a pre-trained machine learning model (such as a support vector machine or neural network) is used to identify whether there are nonlinear absorption peaks or multimodal reflection modes in the spectral data caused by specific materials or microstructures.

[0090] For example, if a distinct absorption peak is identified in the 800nm ​​band, and the shape of this peak exhibits non-linear characteristics, this may indicate the presence of a specific polymer coating on the surface of the target object. Based on this identified non-linear variation pattern, the filtering parameters of the correction algorithm are dynamically adjusted. Specifically, the filter type might be changed from a simple mean filter to a Wiener filter, which is more suitable for processing non-linear signals, and its cutoff frequency might be adjusted to preserve the microscopic features represented by the absorption peak while filtering out high-frequency random noise.

[0091] After the filter parameters are updated, the local sensing unit immediately performs real-time residual analysis on the corrected optical pose data. Assume the analysis shows a root mean square error (RMSE) of 0.05 mm between the corrected pose data and the reference pose data obtained through a high-precision laser scanner, while the preset accuracy threshold is 0.02 mm. Since the preset threshold is not met, the system will fine-tune the filter order and bandwidth based on the residual analysis results, for example, using a gradient descent algorithm, and perform correction and residual analysis again. This iterative process continues until the RMS error is reduced to below 0.02 mm, and the residual fluctuation meets the preset threshold, thus ensuring that the final optical pose data used for micro-pose correction has extremely high accuracy and reliability.

[0092] refer to Figure 2 This application proposes a robot grasping and positioning control system, applied to a robot grasping and positioning control method. The system includes: The local perception module acquires the actual microscopic position and attitude data of the robot end effector relative to the target object when the robot end effector approaches the target object to a set distance. The deviation calculation module calculates the instantaneous deviation between the robot end effector and the target object based on the actual microscopic position and attitude data, as well as the theoretical expected position and attitude of the robot end effector. The instruction generation module generates a compensation motion instruction based on the instantaneous deviation when the instantaneous deviation exceeds the set tolerance range. The correction module drives the robot's end effector to perform micro-orientation correction according to the compensation motion command, so as to align the robot's end effector with the target object; The grasping execution module performs the grasping operation after the robot's end effector completes micro-orientation correction and aligns with the target object.

[0093] Specifically, the local perception module can be understood as a collection of sensors integrated into the robot's end effector, designed to provide high-precision local environmental perception capabilities. For example, this module may include, but is not limited to, miniature vision sensors, LiDAR, structured light sensors, or ultrasonic sensors. These sensors are used to capture, in real time, the surface features and contour information of the target object, as well as the actual microscopic position and orientation data of the robot's end effector relative to the target object, as the robot's end effector approaches the target object to a set distance. This data forms the basis for subsequent precise alignment and grasping.

[0094] The deviation calculation module can be a standalone processor unit or a software module integrated into the robot controller. Its purpose is to process and analyze the real-time data acquired by the local sensing module. In practical applications, this module receives actual microscopic position and attitude data and compares it with the theoretically expected position and attitude of the robot's end effector, either pre-set or obtained through path planning, to accurately calculate the instantaneous deviation between the two. This deviation can be a positional deviation (such as offsets in the X, Y, and Z axes) or an attitude deviation (such as rotational angle deviations around the X, Y, and Z axes).

[0095] The instruction generation module is responsible for generating corresponding compensation motion instructions based on the instantaneous deviation output by the deviation calculation module. When the instantaneous deviation exceeds the preset tolerance range, the module will generate a series of tiny motion instructions based on the magnitude and direction of the deviation using a preset control algorithm (such as PID control, fuzzy control, or model-based control algorithm). These instructions are designed to guide the robot's end effector to move in a direction that is precisely aligned with the target object.

[0096] The correction module is the execution part of the robot control system. Its purpose is to convert the compensation motion commands generated by the command generation module into the actual physical motion of the robot's end effector. Specifically, this module performs fine micro-attitude corrections by driving the joint motors or micro-motion mechanisms of the robot's end effector. For example, by adjusting the fine-tuning mechanism of the robot's wrist or utilizing the redundant degrees of freedom of the robot body, sub-millimeter or sub-degree adjustments to the position and attitude of the end effector can be achieved until the robot's end effector is precisely aligned with the target object.

[0097] The grasping execution module is a crucial part of the system for ultimately completing the task. Its purpose is to safely and effectively perform grasping operations after the robot's end effector is precisely aligned with the target object. For example, this module can control the grippers, suction cups, or other grasping tools on the robot's end effector to perform actions such as closing, adhering, or inserting, thereby firmly grasping the target object.

[0098] The content disclosed above is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.

Claims

1. A robot grasping and positioning control method, characterized in that, The method includes the following steps: When the robot end effector approaches the target object to a set distance, the actual microscopic position and attitude data of the robot end effector relative to the target object are obtained by the local sensing unit installed on the robot end effector. Based on the actual microscopic position and attitude data, as well as the theoretical expected position and attitude of the robot end effector, calculate the instantaneous deviation between the robot end effector and the target object; When the instantaneous deviation exceeds the set tolerance range, a compensation motion command is generated based on the instantaneous deviation. According to the compensation motion command, drive the robot end effector to perform micro-orientation correction so that the robot end effector is aligned with the target object; After the robot's end effector completes micro-orientation correction and aligns with the target object, it performs the grasping operation.

2. The robot grasping and positioning control method as described in claim 1, characterized in that, The steps for generating compensation motion commands based on instantaneous deviation include: Install local sensing units and torque sensors on the robot's end effector; When the robot's end effector approaches the target object to the critical contact zone, it performs a micro-pressure probe action; During the micro-pressure probe operation, the optical orientation data of the local sensing unit and the physical torque data of the torque sensor are acquired simultaneously. By comparing the instantaneous pose deviation reported by optical pose data with the trend of contact torque change fed back by physical torque data, the authenticity of the instantaneous pose deviation or the presence of measurement noise can be determined. When the instantaneous pose deviation is determined to be real, a compensation motion command is generated based on the instantaneous pose deviation. When it is determined that the instantaneous pose deviation has measurement noise, the optical pose data with measurement noise is corrected to obtain the corrected optical pose data, and a compensation motion command is generated based on the corrected optical pose data.

3. The robot grasping and positioning control method as described in claim 2, characterized in that, The steps for generating compensated motion commands also include: Install non-contact force sensors on the robot's end effector; When the robot's end effector approaches the target object to the non-contact critical zone, it executes a micro-motion command; During the execution of micro-motion commands, the optical posture data of the local sensing unit, the non-contact force data of the non-contact force sensor, and the physical torque data of the torque sensor are acquired simultaneously. By comparing the instantaneous pose deviation reported by optical pose data with the non-contact force change trend fed back by non-contact force sensor and the contact torque change trend fed back by physical torque data, the authenticity of the instantaneous pose deviation or the presence of measurement noise can be determined. When the instantaneous pose deviation is determined to be real, a compensation motion command is generated based on the instantaneous pose deviation. When it is determined that the instantaneous pose deviation has measurement noise, the optical pose data with measurement noise is corrected to obtain the corrected optical pose data, and a compensation motion command is generated based on the corrected optical pose data.

4. The robot grasping and positioning control method as described in claim 3, characterized in that, The steps to determine the authenticity of instantaneous pose deviation or the presence of measurement noise include: Before comparing the instantaneous pose deviation reported by the optical pose data with the non-contact force change trend fed back by the non-contact force sensor and the contact torque change trend fed back by the physical torque data, the optical pose data of the local sensing unit, the non-contact force data of the non-contact force sensor and the physical torque data of the torque sensor are timestamped. By analyzing the timestamp information in the data streams of the local sensing unit, the non-contact force sensor and the torque sensor, the microsecond-level time delay or jitter between the data streams of the local sensing unit, the non-contact force sensor and the torque sensor is identified and compensated, and the timestamped optical pose data, non-contact force data and physical torque data are obtained. Based on the timestamped optical pose data, non-contact force data, and physical torque data, calculate the cross-correlation coefficient or dynamic synchronization index between instantaneous pose deviation, non-contact force change trend, and contact torque change trend; Based on the cross-correlation coefficient or dynamic synchronization index, determine the authenticity of the instantaneous pose deviation or whether there is measurement noise.

5. The robot grasping and positioning control method as described in claim 3, characterized in that, The steps for correcting optical pose data containing measurement noise to obtain corrected optical pose data include: The monitoring results are obtained by monitoring external environmental parameters, surface characteristics of target objects, or signal quality of local sensing units. Based on the monitoring results, the correction algorithm parameters are dynamically adjusted to obtain the adjusted correction algorithm parameters. Based on the adjusted correction algorithm parameters, the optical pose data containing measurement noise is corrected to obtain the corrected optical pose data.

6. The robot grasping and positioning control method as described in claim 5, characterized in that, The steps for correcting optical pose data containing measurement noise to obtain corrected optical pose data include: Before performing correction, acquire the microscopic surface feature data of the target object; Based on microscopic surface feature data, identify the microscopic textures or irregular geometric features present on the surface of the target object; Based on the identified micro-textures or irregular geometric features, the filter window size or feature retention threshold of the correction algorithm is dynamically adjusted to avoid misjudging real micro-surface features as measurement noise. Based on the adjusted correction algorithm parameters and the adjusted correction algorithm's filter window size or feature retention threshold, the optical pose data containing measurement noise is corrected to obtain the corrected optical pose data.

7. The robot grasping and positioning control method as described in claim 6, characterized in that, The steps for correcting optical pose data containing measurement noise to obtain corrected optical pose data include: A dynamic spectral analysis module is integrated into the local sensing unit to obtain the spectral characteristics of the microscopic features of the target object's surface; Update the feature preservation threshold of the correction algorithm based on spectral characteristics; Update the filtering parameters of the correction algorithm based on the spectral characteristics; Based on the adjusted correction algorithm parameters, the updated feature retention threshold, and the updated filtering parameters, and according to the adjusted correction algorithm's filtering window size or feature retention threshold, the optical pose data containing measurement noise is corrected to obtain the corrected optical pose data.

8. The robot grasping and positioning control method as described in claim 7, characterized in that, The steps for updating the filter parameters of the correction algorithm include: Integrate spectral reference standards on the local sensing unit; Before or during the acquisition of the spectral characteristics of the microscopic features of the target object's surface, the spectral reference standard is periodically scanned to obtain the actual spectral response data of the spectral reference standard. The actual spectral response data of the spectral reference standard is compared with the preset reference spectral response data to identify the spectral response deviation of the local sensing unit. Generate spectral response calibration parameters based on the spectral response deviation; The spectral characteristics are corrected using spectral response calibration parameters to obtain the corrected spectral characteristics; The filtering parameters of the correction algorithm are updated based on the corrected spectral characteristics.

9. A robot grasping and positioning control method as described in claim 8, characterized in that, The steps for updating the filter parameters of the correction algorithm include: Integrate multispectral or hyperspectral imaging modules on local sensing units; Using multispectral or hyperspectral imaging modules, spectral data of the microscopic features of the target object's surface can be acquired in multiple discrete or continuous wavelength ranges. Feature extraction is performed on the spectral data to obtain spectral features; Based on spectral characteristics, identify nonlinear, multimodal, or rapid transient change patterns in spectral data; Based on the identified change patterns, the filtering parameters of the correction algorithm are dynamically adjusted and updated. The filtering parameters include the filter type, order, cutoff frequency, or bandwidth. After updating the filtering parameters of the correction algorithm, the corrected optical pose data is subjected to real-time residual analysis through the local sensing unit to obtain the residual analysis results. Based on the residual analysis results, evaluate the correction effect of the correction algorithm after updating the filter parameters; When the evaluated correction effect does not reach the preset accuracy threshold, the filtering parameters of the correction algorithm are iteratively optimized based on the residual analysis results until the residual fluctuation meets the preset threshold.

10. A robot grasping and positioning control system, applied to the robot grasping and positioning control method as described in claim 1, characterized in that, The system includes: The local perception module acquires the actual microscopic position and attitude data of the robot end effector relative to the target object when the robot end effector approaches the target object to a set distance. The deviation calculation module calculates the instantaneous deviation between the robot end effector and the target object based on the actual microscopic position and attitude data, as well as the theoretical expected position and attitude of the robot end effector. The instruction generation module generates a compensation motion instruction based on the instantaneous deviation when the instantaneous deviation exceeds the set tolerance range. The correction module drives the robot's end effector to perform micro-orientation correction according to the compensation motion command, so as to align the robot's end effector with the target object; The grasping execution module performs the grasping operation after the robot's end effector completes micro-orientation correction and aligns with the target object.

Images