Multi-sensor data detection and processing method, device and robot

Abstract

The present invention discloses a detection and processing method, device and robot for multi-sensor data, including: receiving observation data from multiple sensors of the robot; calculating and acquiring the posture data of the robot according to the data preprocessed from the observation data of multiple sensors; using the acquired posture data to detect and evaluate the data state, adjusting the credibility of each sensor data in multiple sensors according to the detection and evaluation results of the data state, and correcting the observation data according to the credibility; acquiring optimized posture data according to the adjusted credibility and the corrected observation data, and reversely adjusting the credibility according to the error results between the optimized posture data; and feeding back the reversely adjusted credibility and the optimized posture data. By adopting the above scheme, valid data of different sensors can be extracted according to the data state of the sensor for calculation to ensure the accuracy of the algorithm, and the entire algorithm can be fault-tolerant to abnormal data, ensuring the stability and safety of the robot.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a method, device and robot for detecting and processing multi-sensor data. Background Art

[0002] As robot technology matures, related robot products are gradually entering the public's field of vision. The emergence of robots not only liberates people's hands, allowing everyone to get rid of boring and repetitive work, but also improves the efficiency of production and life.

[0003] Many existing robots use multiple sensor data fusion technologies, that is, multiple sensors are installed on the body to detect specific data, and the positioning problem of the robot in special scenes is solved in a targeted manner. Under the existing processing framework and architecture, the data processing method usually adopted is to judge the status of the data received and processed by the interface, that is, to detect data before the relevant algorithm runs, and throw out the problematic data if anomalies are found to ensure the normal and stable operation of the robot.

[0004] However, the existing sensor data detection method is too simplistic in its judgment of data. Some data cannot be judged to be abnormal through simple data detection and time detection, or may affect the accuracy of the algorithm. Summary of the invention

[0005] The main purpose of the present invention is to disclose a method, device and robot for detecting and processing multi-sensor data, so as to at least solve the problems that the existing sensor data detection method used in the related technology is too single in judging the data, and some data cannot be judged whether there are abnormalities through simple data self-detection and time detection, or may affect the accuracy of the algorithm.

[0006] According to one aspect of the present invention, a detection and processing method for multi-sensor data is provided.

[0007] The detection and processing method of multi-sensor data according to the present invention includes: receiving observation data from multiple sensors of the robot; calculating and acquiring the posture data of the robot based on the data preprocessed from the observation data of the above multiple sensors; using the above acquired posture data to detect and evaluate the data status, adjusting the credibility of each sensor data of the above multiple sensors according to the detection and evaluation results of the data status, and correcting the above observation data according to the above credibility; acquiring optimized posture data according to the adjusted credibility and the corrected observation data, and reversely adjusting the credibility according to the error result between the above optimized posture data; and feeding back the reversely adjusted credibility and the above optimized posture data.

[0008] According to another aspect of the present invention, a detection and processing device for multi-sensor data is provided. The detection and processing device for multi-sensor data according to the present invention includes: a receiving module for receiving observation data from multiple sensors of a robot; a calculation and acquisition module for calculating and acquiring the posture data of the robot according to the data preprocessed from the observation data of the above multiple sensors; a detection and processing module for detecting and evaluating the data state using the posture data acquired above, adjusting the credibility of each sensor data in the above multiple sensors according to the evaluation result of the data state, and correcting the above observation data according to the above credibility; a reverse adjustment module for acquiring optimized posture data according to the adjusted credibility and the corrected observation data, and reversely adjusting the credibility according to the error result between the above optimized posture data; a feedback module for feeding back the reversely adjusted credibility and the above optimized posture data to the calculation and acquisition module.

[0009] According to yet another aspect of the present invention, a robot is provided.

[0010] The robot according to the present invention comprises: a memory and a processor, wherein the memory is used to store computer-executable instructions; and the processor is used to execute the computer-executable instructions stored in the memory, so that the robot executes any one of the above methods.

[0011] According to the present invention, the observation data of multiple sensors are positively detected and evaluated in real time, and the credibility of each sensor data in multiple sensors is adjusted according to the detection and evaluation results of the data status, and the above observation data is corrected according to the credibility; the optimized posture data is obtained according to the adjusted credibility and the corrected observation data, and the above credibility is reversely adjusted according to the error result between the above optimized posture data; then the reversely adjusted credibility and the above optimized posture data are fed back as the basis for the next calculation. The availability of sensor data is judged by a parallel method of positive detection and evaluation and reverse adjustment feedback. This method has high real-time performance, and can extract valid data of different sensors according to the data status of the sensor for calculation to ensure the accuracy of the algorithm. The observation data is corrected during the execution process, which can make the entire algorithm fault-tolerant to abnormal data and ensure the stability and safety of the robot. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a flow chart of a method for detecting and processing multi-sensor data according to an embodiment of the present invention;

[0013] Figure 2 is a structural block diagram of a detection and processing device for multi-sensor data according to an embodiment of the present invention;

[0014] Figure 3is a structural block diagram of a detection and processing device for multi-sensor data according to a preferred embodiment of the present invention;

[0015] Figure 4 is a structural block diagram of a robot according to an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The specific implementation of the present invention is described in detail below with reference to the accompanying drawings.

[0017] According to an embodiment of the present invention, a method for detecting and processing multi-sensor data is provided.

[0018] Figure 1 FIG. 1 is a flow chart of a method for detecting and processing multi-sensor data according to an embodiment of the present invention. Figure 1 As shown, the detection and processing method of the multi-sensor data includes:

[0019] Step S101: receiving observation data from multiple sensors of the robot;

[0020] Step S102: Calculate and obtain the position and posture data of the robot according to the preprocessed data of the observation data of the above multiple sensors;

[0021] Step S103: using the acquired posture data to detect and evaluate the data state, adjusting the credibility of the data of each sensor in the multiple sensors according to the detection and evaluation results of the data state, and correcting the observation data according to the credibility;

[0022] Step S104: obtaining optimized posture data according to the adjusted credibility and the corrected observation data, and reversely adjusting the credibility according to the error result between the optimized posture data;

[0023] Step S105: Feedback the reversely adjusted credibility and the optimized posture data.

[0024] According to an embodiment of the present invention, the observation data of multiple sensors are positively detected and evaluated in real time, the credibility of each sensor data in the multiple sensors is adjusted according to the detection and evaluation results of the data status, and the above observation data is corrected according to the credibility; the optimized posture data is obtained according to the adjusted credibility and the corrected observation data, and the above credibility is reversely adjusted according to the error result between the above optimized posture data; then the reversely adjusted credibility and the above optimized posture data are fed back as the basis for the next calculation. The availability of sensor data is judged by a parallel method of positive detection and evaluation and reverse adjustment feedback. This method has high real-time performance, and can extract valid data of different sensors according to the data status of the sensor for calculation to ensure the accuracy of the algorithm. The observation data is corrected during the execution process, which can make the entire algorithm fault-tolerant to abnormal data and ensure the stability and safety of the robot.

[0025] Preferably, after step S101 and before step S102, the following processing may also be included: performing time anomaly processing on the above-mentioned observation data according to the timestamp information of the above-mentioned observation data, wherein the above-mentioned time anomaly processing includes at least one of the following: data time disorder processing, data delay processing, data loss processing, and data jump processing; performing data out-of-bounds detection processing on the above-mentioned observation data to determine whether the relevant value of the above-mentioned observation data is within a predetermined range; performing data volatility processing on the above-mentioned observation data, wherein the data volatility processing includes: determining the mean change rate of adjacent data tuples, if there is a data tuple whose mean change rate is greater than a predetermined change rate threshold, then by comparing the means of multiple groups of partial data in the data tuple, detecting and eliminating abnormal data within the data tuple. Align the collection time of the observation data according to the collection frequency of the observation data of each sensor.

[0026] The following processing can be performed at the data receiving end:

[0027] (1) Data time anomaly processing (e.g., time disorder, delay, loss, jump)

[0028] This part mainly ensures the accuracy of data by simply judging the observed data and its timestamp. Once a problem occurs, the data will be adjusted or thrown out at the receiving end. For example, if the time is out of order, the corresponding observed data will be discarded. If there is a delay, loss, jump, etc., the corresponding observed data can be marked or recorded.

[0029] (2) Data out-of-bounds processing (for example, data out-of-bounds, data pointer out-of-bounds)

[0030] When data is accessed, data boundary detection may be added to determine whether the relevant values ​​of the above observed data are within a predetermined range, and to determine whether there is a possibility of data or data pointer crossing the boundary.

[0031] (3) Data volatility processing (e.g., estimating the magnitude of data fluctuations)

[0032] Using statistical methods, the fluctuation degree of data changes is recorded in real time, and the variance and standard deviation of data changes are recorded at the same time. The data situation is judged based on the robot's preset running speed and rotation angle to determine whether there is any abnormality in the current data.

[0033] For example, assuming that s(ti) is the sensor data collected at a certain time, k sensor data are set as a calculation tuple according to the motion characteristics, then each k data will generate the mean Smeans of the tuple.

[0034] Smeans = (1 / k)*∑ k i＝1 s(ti)

[0035] According to the robot's motion properties, it can be determined that in a short time, the mean of adjacent tuples will not change significantly. At this time, if the change rate of the mean of a tuple and the mean of the adjacent tuple is greater than a predetermined threshold, it is necessary to mark the tuple with a large mean change, marking the tuple as likely to contain abnormal data, and then filter out the data by using the local data mean inside the tuple. For example, randomly extract multiple groups of partial data from the tuple, calculate the mean of the above multiple groups of partial data, and then detect abnormal data based on the mean change rate of each group of partial data, and remove the detected abnormal data.

[0036] (4) Data alignment processing (for example, data memory length alignment)

[0037] (5) Data length alignment processing (e.g., data size, data queue size) and unified data types.

[0038] (6) Data time alignment processing (e.g., timestamp alignment)

[0039] Since the observation data of each sensor are collected at different frequencies, the collection timestamps of the observation data are aligned to ensure that the data of each tuple can be compared for mean calculation.

[0040] Preferably, the above step S102 may further include: for each sensor in the first group of sensors (for example, a sensor group including an inertial sensor IMU and / or an odometer), integrating and solving the observation data corresponding to the sensor according to the motion-related information maintained in the sensor (for example, information associated with speed, such as linear velocity, angular velocity, acceleration, etc.), and obtaining the first position data within a predetermined time period; fusing the first position data corresponding to one or more sensors in the above first group of sensors, and assigning the fused first position data as the initial value to the second group of sensors (for example, a sensor group including a lidar), and for each sensor in the second group of sensors, solving and obtaining the second position data corresponding to the sensor according to the motion-related information maintained in the sensor (for example, angular velocity, acceleration, etc.).

[0041] In the process of optimal implementation, the following model can be used for solution:

[0042] 1. Data pre-calculation model (for example, single sensor data integration, multi-sensor combination integration, etc.).

[0043] For single sensor integration, a set of linear velocity, angular velocity and other information is maintained for different sensors. Pre-integration and solution are performed based on each received sensor data. Taking IMU as an example, pre-integration and solution are performed based on the received IMU data to estimate the current rotation state of the robot. From the i-th moment, the P (displacement) and Q (rotation) at the j-th moment are obtained by integrating the IMU observation data. The calculation results of the IMU data tuple are compared with the calculation results of the same time tuple of other sensors (which can be an odometer, another IMU, or other sensors). If the integration result is obviously deviated, it is considered that the data of the IMU sensor is abnormal. The internal data local integration method can be continued to calculate the mean velocity and the mean angular velocity, and the abnormal data can be detected by the calculation results.

[0044] 2. Recursive observation solution model (for example, given a priori, a model is formulated recursively. The model selection needs to be selected based on the last error situation. The initial default selection is the uniform speed model)

[0045] Initially, the uniform speed model is selected for recursive solution. When data anomalies occur, the model will perform partial elimination of the sensor observation data combination based on the abnormal situation of the sensor. That is, in the combined observation part, the sensors with obvious problems are eliminated, and the remaining sensors are relied on for observation and solution, and state change records.

[0046] In the preferred implementation process, when the observation data of the sensor is found to be abnormal, the individual data anomalies can be recorded, for example, the anomalies of a single frame of data can be recorded; the statistical data anomalies can be recorded, for example, by recording the anomalies of sensor data for a period of time, the sensor itself can be evaluated to determine whether there is a problem with the sensor itself and whether adjustment or manual inspection is required.

[0047] Preferably, in the above-mentioned step S103, adjusting the credibility of each sensor data in the above-mentioned multiple sensors according to the evaluation result of the data status can further include: for each sensor in the above-mentioned second group of sensors, determining the error score corresponding to the sensor according to the matching degree between the historical posture data corresponding to the sensor and the second posture data corresponding to the sensor, and looping this step until multiple error scores are obtained; adjusting the credibility of each sensor data in the above-mentioned first group of sensors and the above-mentioned second group of sensors according to the multiple error scores.

[0048] In the preferred implementation process, the credibility of each sensor data is adjusted according to the evaluation results of the data status. For example, according to the situation of a single sensor, the sensor statistics, and the error score corresponding to the sensor, the credibility of the sensor data during use is adjusted. When the error score is high, a higher credibility will be given accordingly during the optimization solution. When the error score is low, the credibility of the sensor will be reduced. In this way, the accuracy and reliability of the optimization results are ensured.

[0049] For example, the error score Score corresponding to the sensor can be obtained in the following way:

[0050]

[0051] Where K is a constant, e i is the matching degree between the historical point cloud data and the current point cloud pose data (the above-mentioned second pose data) corresponding to the sensor, and n is the number of laser point cloud data points.

[0052] For example, the robot system has a lidar sensor, an IMU, and an odometer. The matching degree between the historical point cloud data and the current point cloud pose data (the second pose data) corresponding to the lidar sensor is used to obtain two error scores Score1 and Score2. These two scores can be used to adjust the credibility of the corresponding IMU sensor or odometer. If the Score1 score value is lower than the predetermined value, the credibility of the corresponding sensor is lowered by a predetermined amount (for example, 0.2).

[0053] Preferably, correcting the above-mentioned observation data according to the above-mentioned credibility further includes: when the current credibility is less than a predetermined credibility threshold, according to the sensor type corresponding to the above-mentioned current credibility, selecting to execute the corresponding observation data correction processing method, wherein the above-mentioned observation data correction processing method includes at least one of the following: directly eliminating one or more frames of sensor observation data, recalculating the calculation results based on the motion-related information maintained by the sensor, and replacing the original abnormal observation values ​​with the calculation results.

[0054] During the preferred implementation process, detection, update and replacement can be performed based on the specific data of the sensor, and the data can be effectively updated and replaced based on the rating results. The main thing is to correct and adjust the front-end observation data based on the rating results. If a certain sensor data is abnormal, it will be based on the sensor characteristics. Taking IMU data as an example, if a frame of sensor has a problem, then the frame data will be directly discarded to ensure that the integration results of other data are normal. For example, if the odometer data is abnormal, it will be recalculated based on the maintained speed and angular velocity, and the original odometer observation result will be replaced by the calculated result.

[0055] Preferably, reversely adjusting the above-mentioned credibility according to the error result between the above-mentioned optimized posture data can further include: selecting reference benchmark data from the above-mentioned optimized posture data, determining the error between other data in the above-mentioned optimized posture data except the above-mentioned reference benchmark data and the reference benchmark data, and reversely adjusting the above-mentioned credibility according to the size of the error.

[0056] In the preferred implementation process, the currently optimized posture data b1 is used as the reference benchmark data for the next time, and the errors between the other data b2 and b3 in the posture data except b1 and b1 are calculated respectively, and the above credibility is reversely adjusted according to the size of the error. If the error is small, the original credibility of the corresponding sensor data is maintained. If the error is large, the credibility of the corresponding sensor data is lowered by a predetermined amount (for example, 0.2). Through the above processing, the credibility of the sensor data can be reversely adjusted by the optimized posture data. And, after further optimization of the credibility of the sensor data is performed, step S105 is executed, that is, the credibility of the optimized posture data and the current latest sensor data is fed back, and steps S102 to S105 perform corresponding processing based on the latest data after feedback.

[0057] According to an embodiment of the present invention, a detection and processing device for multi-sensor data is provided.

[0058] Figure 2 is a structural block diagram of a detection and processing device for multi-sensor data according to an embodiment of the present invention. Figure 2As shown, the detection and processing device for multi-sensor data includes: a receiving module 20, which is used to receive observation data from multiple sensors of the robot; a calculation and acquisition module 22, which is used to calculate and acquire the posture data of the robot based on the data preprocessed from the observation data of the above multiple sensors; a detection and processing module 24, which is used to detect and evaluate the data state using the posture data acquired above, adjust the credibility of each sensor data in the above multiple sensors according to the evaluation result of the data state, and correct the above observation data according to the above credibility; a reverse adjustment module 26, which is used to acquire the optimized posture data according to the adjusted credibility and the corrected observation data, and reversely adjust the credibility according to the error result between the above optimized posture data; a feedback module 28, which is used to feed back the reverse adjusted credibility and the above optimized posture data to the calculation and acquisition module.

[0059] According to an embodiment of the present invention, the detection and processing device for multi-sensor data performs real-time forward detection and evaluation of the observation data of the multi-sensors, adjusts the credibility of each sensor data in the multiple sensors according to the detection and evaluation results of the data status, and corrects the observation data according to the credibility; obtains the optimized posture data according to the adjusted credibility and the corrected observation data, and reversely adjusts the credibility according to the error result between the optimized posture data; then feeds back the reversely adjusted credibility and the optimized posture data as the basis for the next calculation. The parallel method of forward detection and evaluation and reverse adjustment feedback is adopted to judge the availability of sensor data. This method has high real-time performance, and can extract valid data of different sensors according to the data status of the sensor for calculation to ensure the accuracy of the algorithm. The observation data is corrected during the execution process, which can make the entire algorithm fault-tolerant to abnormal data and ensure the stability and safety of the robot.

[0060] Preferably, if Figure 3As shown, the detection and processing device for multi-sensor data also includes: a preprocessing module 30, the preprocessing module 30 includes: a time anomaly processing submodule 300, which is used to perform time anomaly processing on the above-mentioned observation data according to the timestamp information of the above-mentioned observation data, wherein the above-mentioned time anomaly processing includes at least one of the following: data time disorder processing, data delay processing, data loss processing, and data jump processing; a data cross-border detection processing submodule 302, which is used to perform data cross-border detection processing on the above-mentioned observation data to determine whether the data value of the above-mentioned observation data is within a predetermined range; a data volatility processing submodule 304, which is used to perform data volatility processing on the above-mentioned observation data, wherein the data volatility processing includes: determining the mean change rate of adjacent data tuples, if there is a data tuple whose mean change rate is greater than a predetermined change rate threshold, then by comparing the means of multiple groups of partial data in the data tuple, the abnormal data inside the data tuple is detected and eliminated. A time alignment module 306 is used to align the collection timestamps of the observation data according to the collection frequency of the observation data of each sensor.

[0061] Preferably, if Figure 3 As shown, the calculation and acquisition module 22 may further include: a first solver module 220, which is used to integrate and solve the observation data corresponding to each sensor in the first group of sensors according to the motion-related information maintained in the sensor (for example, information associated with speed, such as linear velocity, angular velocity, acceleration, etc.), and obtain the first position data within a predetermined time period; a second solver module 222, which is used to fuse the first position data corresponding to one or more sensors in the above-mentioned first group of sensors, assign the fused first position data as the initial value to the second group of sensors, and for each sensor in the second group of sensors, solve and obtain the second position data corresponding to the sensor according to the motion-related information maintained in the sensor.

[0062] Preferably, if Figure 3 As shown, the detection processing module 24 may further include: an acquisition submodule 240, for determining, for each sensor in the second group of sensors, an error score corresponding to the sensor according to a degree of matching between the historical posture data corresponding to the sensor and the second posture data corresponding to the sensor, and executing this step in a loop until a plurality of the error scores are obtained; and an adjustment submodule 242, for adjusting the credibility of the data of each sensor in the first group of sensors and the second group of sensors according to the plurality of the error scores.

[0063] According to an embodiment of the present invention, a robot is also provided.

[0064] Figure 4 is a structural block diagram of a robot according to an embodiment of the present invention. Figure 4As shown, the robot according to the present invention includes: a memory 40 and a processor 42, wherein the memory 40 is used to store computer execution instructions; the processor 42 is used to execute the computer execution instructions stored in the memory, so that the robot executes the detection and processing method of multi-sensor data provided in the above embodiment.

[0065] The processor 42 may be a central processing unit (CPU). The processor 42 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0066] The memory 40, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions / modules corresponding to the robot repositioning method in the embodiment of the present invention. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions and modules stored in the memory.

[0067] The memory 40 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required by at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 40 may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0068] The one or more modules are stored in the memory 40 and when executed by the processor 42, the following steps are performed: Figure 1 The detection and processing method of multi-sensor data in the illustrated embodiment.

[0069] For details of the above robots, please refer to Figure 1 and Figure 2 The corresponding related descriptions and effects in the illustrated embodiments can be understood and will not be repeated here.

[0070] In summary, with the help of the above implementation mode provided by the present invention, the credibility of the internal data of the algorithm is judged by performing real-time forward observation and reverse feedback in parallel, the sensor data is judged by observing whether the received data group is normal, and the abnormal data is diagnosed and analyzed in real time and efficiently. The system can also perform fault tolerance processing on the data with some abnormalities, and judge the sensor status according to the statistical results. This method has high real-time performance and can extract valid data from different sensors according to the status of the sensors for calculation to ensure the accurate operation of the algorithm and improve the accuracy. At the same time, it can also make the entire algorithm fault-tolerant to abnormal data and ensure the stability and safety of the robot.

[0071] The above disclosures are only several specific embodiments of the present invention; however, the present invention is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A detection and processing method for multi-sensor data, characterized in that: include: Receive observation data from multiple sensors of the robot; The robot's posture data is calculated based on the preprocessed data of the observation data of the multiple sensors, wherein for each sensor in the first group of sensors, the observation data corresponding to the sensor is integrated and solved according to the motion related information maintained in the sensor to obtain the first posture data within a predetermined time period, the first posture data corresponding to one or more sensors in the first group of sensors are fused, and the fused first posture data is assigned as an initial value to the second group of sensors, and for each sensor in the second group of sensors, the second posture data corresponding to the sensor is solved and obtained according to the motion related information maintained in the sensor; The acquired posture data is used to detect and evaluate the data state, and the credibility of each sensor data in the multiple sensors is adjusted according to the detection and evaluation result of the data state, and the observation data is corrected according to the credibility, wherein adjusting the credibility of each sensor data in the multiple sensors according to the evaluation result of the data state includes: for each sensor in the second group of sensors, according to the matching degree between the historical point cloud posture data corresponding to the sensor and the second posture data corresponding to the sensor, determining the error score corresponding to the sensor, repeating this step until a plurality of the error scores are obtained, and adjusting the credibility of each sensor data in the first group of sensors and the second group of sensors according to the plurality of the error scores; Acquire optimized posture data according to the adjusted credibility and the corrected observation data, and reversely adjust the credibility according to the error result between the optimized posture data; The reversely adjusted credibility and the optimized posture data are fed back.

2. The method according to claim 1, characterized in that After receiving the observation data from the multiple sensors of the robot, and before calculating and acquiring the posture data of the robot according to the preprocessed data of the observation data of the multiple sensors, the method further includes at least one of the following: Performing time anomaly processing on the observation data according to the timestamp information of the observation data, wherein the time anomaly processing includes at least one of the following: data time disorder processing, data delay processing, data loss processing, and data jump processing; Performing data out-of-bounds detection processing on the observed data to determine whether a relevant value of the observed data is within a predetermined range; Performing data volatility processing on the observed data, wherein the data volatility processing includes: determining the mean change rate of adjacent data tuples, and if there is a data tuple whose mean change rate is greater than a predetermined change rate threshold, detecting and eliminating abnormal data within the data tuple by comparing the means of multiple groups of partial data within the data tuple; According to the collection frequency of the observation data of each sensor, the collection timestamps of the observation data are aligned.

3. The method according to claim 1, characterized in that Correcting the observed data according to the credibility includes: When the current credibility is less than a predetermined credibility threshold, the corresponding observation data correction processing method is selected to be executed according to the sensor type corresponding to the current credibility, wherein the observation data correction processing method includes at least one of the following: directly eliminating one or more frames of sensor observation data, recalculating the calculation results based on the motion-related information maintained by the sensor, and using the calculation results to replace the original abnormal observation values.

4. The method according to claim 1, characterized in that: Reversely adjusting the credibility according to the error result between the optimized posture data includes: Select reference benchmark data from the optimized posture data, determine the error between other data in the optimized posture data except the reference benchmark data and the reference benchmark data, and reversely adjust the credibility according to the size of the error.

5. A detection and processing device for multi-sensor data, characterized in that: include: A receiving module, used for receiving observation data from multiple sensors of the robot; A calculation and acquisition module, used for calculating and acquiring the posture data of the robot according to the data preprocessed from the observation data of the multiple sensors, wherein the calculation and acquisition module comprises: a first solver module, used for performing integral calculation on the observation data corresponding to each sensor in the first group of sensors according to the motion related information maintained in the sensor, and acquiring the first posture data within a predetermined time period; and a second solver module, used for fusing the first posture data corresponding to one or more sensors in the first group of sensors, assigning the fused first posture data as an initial value to the second group of sensors, and for each sensor in the second group of sensors, calculating and acquiring the second posture data corresponding to the sensor according to the motion related information maintained in the sensor; a detection processing module, configured to detect and evaluate data status using the acquired posture data, adjust the credibility of each sensor data in the multiple sensors according to the evaluation result of the data status, and correct the observation data according to the credibility, wherein the detection processing module comprises: an acquisition submodule, configured to determine, for each sensor in the second group of sensors, an error score corresponding to the sensor according to the matching degree between the historical posture data corresponding to the sensor and the second posture data corresponding to the sensor, and to execute this step repeatedly until a plurality of the error scores are acquired; and an adjustment submodule, configured to adjust the credibility of each sensor data in the first group of sensors and the second group of sensors according to the plurality of the error scores; A reverse adjustment module, used for obtaining optimized posture data according to the adjusted credibility and the corrected observation data, and reversely adjusting the credibility according to the error result between the optimized posture data; A feedback module is used to feed back the reversely adjusted credibility and the optimized posture data to the calculation acquisition module.

6. The device according to claim 5, characterized in that Also includes: A preprocessing module, wherein the preprocessing module comprises at least one of the following: A time anomaly processing submodule, configured to perform time anomaly processing on the observation data according to the timestamp information of the observation data, wherein the time anomaly processing includes at least one of the following: data time disorder processing, data delay processing, data loss processing, and data jump processing; A data out-of-bounds detection processing submodule, used to perform data out-of-bounds detection processing on the observed data to determine whether the data value of the observed data is within a predetermined range; A data volatility processing submodule is used to perform data volatility processing on the observed data, wherein the data volatility processing includes: determining the mean change rate of adjacent data tuples, and if there is a data tuple whose mean change rate is greater than a predetermined change rate threshold, detecting and eliminating abnormal data within the data tuple by comparing the means of multiple groups of partial data within the data tuple; The time alignment module is used to align the collection timestamps of the observation data according to the collection frequency of the observation data of each sensor.

7. A robot comprising: A memory and a processor, characterized in that: The memory is used to store computer-executable instructions; The processor is used to execute the computer-executable instructions stored in the memory, so that the robot performs the method according to any one of claims 1 to 4.

Images