Improved MESO collision detection method based on collaborative robot joint moment real-time difference

Abstract

The invention discloses an improved MESO collision detection method based on collaborative robot joint torque real-time difference, and belongs to the field of robot collision detection. According to the method, firstly, through single-joint friction force identification, Stribeck model parameters capable of accurately fitting friction force characteristics of a low-speed area are obtained; in the offline stage, an improved extended state observer is adopted to estimate an external torque, the torque and the speed are differentiated, non-collision data sets and collision data sets are collected, the mean value and the covariance of the non-collision data sets and the collision data sets are counted, the mahalanobis distance of the two types of data sets is calculated to divide sub-domains, and then an index basis function coefficient is determined. In the online stage, the measurement quantity is obtained in real time, the mahalanobis distance and the primary function value of the current period are calculated, the current confidence coefficient is updated by combining the confidence coefficient of the previous period, and whether collision happens or not is judged by comparing the collision probability with the non-collision probability. According to the method, the problems of false detection caused by a joint reversing moment peak and external moment estimation noise, detection blind areas of a traditional threshold method and the like are effectively solved, and the accuracy and reliability of collision detection are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of safety detection in industrial applications of collaborative robots, specifically involving an improved MESO collision detection method based on real-time differential joint torques of collaborative robots. Background Technology

[0002] Collision detection technology is an important component of robot application safety research, and it is widely used in industrial and daily life scenarios. To avoid accidental damage caused by collisions or interference between robots and people or the environment in production and daily life, collision recognition and detection functions are required in scenarios such as automated operation of industrial robots, operation of service robots, automatic charging of robotic arms for new energy vehicles, and detection of hand gripping in the trunk of new energy vehicles.

[0003] Currently, publicly available research methods for robot collision detection include model-free collision detection and collision detection based on precise dynamic models. Model-free collision detection relies on visual cameras or external sensors. By installing force sensors on the robot's wrist or base plate, or by covering the robot's surface with an electronic skin, this method responds to external force collision signals, detects changes in external force, and realizes collision event detection and subsequent processing logic. While existing collision detection methods show some performance in terms of collision event detection sensitivity, their implementation requires the introduction of external sensor components. This design significantly increases the overall hardware cost of the robot. Furthermore, due to limitations in the actual deployment process and coverage capacity of the electronic skin, it is difficult to completely cover the entire outer surface of the robot. These missing coverage areas create blind spots in collision detection, thus affecting the integrity and reliability of the robot's collision detection function under complex working conditions. In the field of collision detection technology, visual sensors or 3D camera equipment can also be used as the core sensing components of collision detection solutions. Specifically, the 3D camera equipment can include structured light cameras, TOF (Time-of-Flight) cameras, binocular cameras, etc. This type of collision detection scheme can break away from the collision judgment logic based on sudden torque changes in traditional technologies, and directly acquire 3D point cloud data of the detection scene through the aforementioned visual sensors or 3D camera equipment. The 3D point cloud data contains the X-axis, Y-axis, and Z-axis coordinate information of each spatial point in the scene, which can be used to reconstruct the three-dimensional morphological features of the target object in space.

[0004] In actual detection, by performing "point cloud processing" on the collected 3D point cloud data, and analyzing the spatial relationship between the processed environment / target point cloud and the pre-built "robot self-model," it is possible to determine whether there is a collision risk between the robot and objects in the environment. However, the aforementioned collision detection schemes based on visual sensors or 3D cameras are mostly applied to the dynamic obstacle avoidance and dynamic path planning of robots in existing technologies. Their core implementation logic is to pre-plan a collision-free path, a process that relies on a large amount of data. This not only places higher demands on the computing power of the robot's hardware system but also makes the detection effect susceptible to interference from external environmental factors. Specifically, in the daily outdoor operation scenarios of service robots, strong light environments can overwhelm the detection signal of 3D cameras, leading to a surge in noise in the 3D point cloud data. This prevents the equipment from accurately distinguishing the target object from the background environment, ultimately affecting the accuracy of collision detection. In industrial operating environments, 3D cameras are susceptible to environmental factors such as weak light, dust, oil, and electromagnetic interference. Such interference can also have a significant negative impact on the accuracy and stability of collision detection. Given the aforementioned technical shortcomings, using visual sensors or 3D camera equipment as auxiliary detection components is a more practical and suitable technical choice in the field of collision detection.

[0005] In the field of robot collision detection technology, collision detection schemes based on dynamic models are another important research branch. The core logic of this scheme is as follows: during robot operation, three types of operational parameters are collected: joint current, joint position, and joint velocity. Based on these collected parameters, the external forces acting on each joint of the robot are estimated using a pre-set dynamic calculation model. Subsequently, the external torque corresponding to the estimated joint external forces is compared with a pre-set collision judgment threshold. If the external torque exceeds the collision judgment threshold, it is used as the basis for determining that a collision event has occurred. In the current collision detection technology system, the above-mentioned collision detection scheme based on dynamic models is one of the most widely used techniques. However, in the actual operation of a robot, its joint acceleration is usually difficult to obtain directly. This technical bottleneck prevents the external force observer based on the dynamic model from fully utilizing its detection performance, resulting in a significant reduction in detection effectiveness. To solve this problem, additional joint accelerometers need to be added to the robot system, but this directly increases the hardware configuration cost of the robot, hindering the low-cost promotion and application of the technology. To address the shortcomings of the existing technology, the academic community has proposed an external force observer technology based on generalized momentum. This scheme, with its rigorous mathematical derivation and extended applicability in collision event isolation and collision object identification, effectively avoids the inversion of the inertia matrix required in traditional external force observers, while also eliminating the need to solve for joint acceleration terms. In the current field of collision detection technology based on dynamic models, the aforementioned external force observer based on generalized momentum has become one of the most widely used mainstream technologies. However, the detection performance of this collision detection scheme based on generalized momentum is highly dependent on the accuracy of the dynamic model's identification. Specifically, the dynamic model constructed using traditional dynamic parameter identification methods has two significant drawbacks: firstly, the model cannot accurately characterize the nonlinear characteristics of the friction term in the low-speed operating range; secondly, the model itself contains uncertainties that have not been precisely quantified. Affected by these drawbacks, when a robot joint is in the process of reversing motion, a reversal torque spike often occurs. This phenomenon forces traditional collision detection threshold-based judgment methods to set higher collision judgment thresholds to avoid false positives, and the increase in the threshold directly reduces the collision detection rate, affecting the sensitivity of the scheme to collision events. Furthermore, in practical applications, collision detection schemes based on generalized momentum observers require setting a collision sensitivity coefficient on the external force estimation results. While this coefficient improves the collision detection sensitivity, it also amplifies the torque noise in the system, which may increase the probability of false detections of collision events and affect the reliability of the detection results. Summary of the Invention

[0006] The main objective of this invention is to overcome the shortcomings and deficiencies of existing technologies and provide an improved MESO collision detection method based on real-time differential joint torques of collaborative robots. By employing the Stribeck friction model in the low-speed region, the improved extended state observer MESO optimizes external torque estimation noise, processes joint differential torques, and differential velocity state variables, combined with a collaborative robot collision detection method based on dynamic Bayesian formulas, the problems encountered in traditional collision detection are solved.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides an improved MESO collision detection method based on real-time differential analysis of joint torques in collaborative robots, comprising the following steps:

[0009] S1. Using single-joint friction identification, four parameters of the Stribeck friction model are obtained. The four parameters are used as inputs for the offline dataset sampling process and the online collision detection process state estimation;

[0010] S2. In the offline dataset sampling process, the improved Extended State Observer (MESO) is used to estimate the external torque, the torque and velocity state variables are differentially processed, and non-collision dataset samples are collected during the free space operation of the robotic arm. And dataset samples when robots collide. ;

[0011] S3. In the offline dataset sampling process, calculate the mean of the non-collision dataset and the collision dataset respectively. Covariance , as input to the online collision detection process;

[0012] S4. In the offline dataset sampling process, calculate the Mahalanobis distance between samples in the non-collision dataset and the collision dataset, respectively. Divide the domain into subdomains based on Mahalanobis distance and obtain the values ​​of the Mahalanobis distance subdomain boundaries. The coefficients of the exponential basis function are calculated using this numerical value. ;

[0013] S5. In the online collision detection process, acquire the measurement data for the current cycle in real time. The algorithm calculates the Mahalanobis distance and collision / non-collision basis function values ​​for the current period, updates the confidence level for the current period based on the current measurement and the confidence level of the previous period, and determines whether a collision has occurred by comparing the magnitudes of the collision probability and the non-collision probability.

[0014] As a preferred technical solution, in step S1, the Stribeck friction model used is a nonlinear friction model for the i-th joint, and its expression is:

[0015]

[0016] in, Let be the static friction term of the i-th joint. Coulomb friction term, For Stribeck speed, This is the term of viscous friction. It is an exponential factor.

[0017] As a preferred technical solution, in step S1, the objective function is minimized using the nonlinear least squares method to identify the parameters of the Stribeck friction model. The objective function is:

[0018] ;

[0019] in, Here, N represents the model parameters, and N represents the number of data points. and The torque and velocity measurements are for the Kth data point of joint i. The predicted frictional torque is calculated using the Stribeck model.

[0020] As a preferred technical solution, in step S2, when using MESO to estimate the external moment during a collision, the joint moment is... Converted into joint differential torque Furthermore, joint differential velocity is introduced. Collision detection is performed using state variables, and non-collision dataset samples are collected during robot motion in free space. and collision dataset samples .

[0021] As a preferred technical solution, the formula for estimating the external torque using MESO is:

[0022]

[0023] in, , For robot control torque, The Cocteau force matrix, For the gravity matrix, Let be the friction force matrix, and let State variables The estimated value, State variables The predicted value, and It is a constant.

[0024] As a preferred technical solution, step S4 specifically involves:

[0025] Solve for Mahalanobis distance between offline non-collision and collision dataset samples. and And solve for the boundary subdomain distance subdomain. and As a completely non-collision subfield and fully collided subfield The division; in the subdomain The edge, calculated area ,in In the subdomain Edge, computation ,in According to the design principles of exponential basis functions, based on and The Markov transition matrix is The form of the exponential basis function is: .

[0026] As a preferred technical solution, in step S5, the confidence update formula for each period is:

[0027] ;

[0028] in, Obtained through iteration. As the normalization factor, It is a Markov matrix.

[0029] Each cycle is compared and The size of the collision determines the collision state, among which Indicates the collision state. Indicates a non-collision state.

[0030] Secondly, the present invention provides an improved MESO collision detection system based on real-time differential joint torque of collaborative robots, which is applied to the improved MESO collision detection method based on real-time differential joint torque of collaborative robots, including a parameter acquisition module, a differential processing module, an offline data statistics module, a basis function coefficient solving module, and an online collision judgment module.

[0031] The parameter acquisition module uses single-joint friction force identification to obtain four parameters of the Stribeck friction force model. The four parameters are used as inputs for the offline dataset sampling process and the online collision detection process state estimation;

[0032] The differential processing module, in the offline dataset sampling process, uses an improved extended state observer MESO to estimate the external torque, performs differential processing on the torque and velocity state variables, and collects non-collision dataset samples of the robotic arm running in free space. And dataset samples when robots collide. ;

[0033] The offline data statistics module calculates the mean values ​​of the non-collision dataset and the collision dataset respectively during the offline dataset sampling process. Covariance , as input to the online collision detection process;

[0034] The basis function coefficient solving module calculates the Mahalanobis distance between samples in the non-collision dataset and the collision dataset respectively during the offline dataset sampling process. Divide the domain into subdomains based on Mahalanobis distance and obtain the values ​​of the Mahalanobis distance subdomain boundaries. The coefficients of the exponential basis function are calculated using this numerical value. ;

[0035] The online collision detection module acquires the measurement value of the current period in real time during the online collision detection process. The algorithm calculates the Mahalanobis distance and collision / non-collision basis function values ​​for the current period, updates the confidence level for the current period based on the current measurement and the confidence level of the previous period, and determines whether a collision has occurred by comparing the magnitudes of the collision probability and the non-collision probability.

[0036] Thirdly, the present invention provides an electronic device, the electronic device comprising:

[0037] At least one processor; and,

[0038] A memory communicatively connected to the at least one processor; wherein,

[0039] The memory stores computer program instructions that can be executed by the at least one processor, which enables the at least one processor to perform the improved MESO collision detection based on real-time differential joint torques of collaborative robots.

[0040] Fourthly, the present invention provides a computer-readable storage medium storing a program, which, when executed by a processor, implements the improved MESO collision detection based on real-time differential joint torques of a collaborative robot.

[0041] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0042] This invention employs the Steibeck model, which better characterizes the nonlinear frictional forces in the low-speed region, and a differential torque approach. This avoids the external torque spikes caused by dynamic uncertainties due to inaccurate nonlinear frictional force and dynamic modeling during joint reversal, and avoids the problem of excessively high collision thresholds required by traditional threshold collision methods. Then, differential velocity measurements are introduced as a supplement to the collision detection state variables, improving the reliability and accuracy of collision detection. An improved extended state observer is introduced to avoid the potential false detections caused by the high noise of conventional generalized momentum observers. The traditional static threshold-based collision detection method is transformed into a dynamic Bayesian probabilistic judgment method. The non-collision exponential basis functions and collision exponential basis functions fitted with a large amount of data are used as priors for the Bayesian update formula, making the collision detection rate more reliable. Moreover, the state variables can change rapidly when a collision occurs, making collision detection more sensitive and effectively improving collision detection performance. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart illustrating the improved MESO collision detection method based on real-time differential joint torques of collaborative robots according to the present invention.

[0045] Figure 2 This is a diagram of the Stribeck friction model of the present invention;

[0046] Figure 3 This is a comparison diagram of the torque and differential torque of the present invention;

[0047] Figure 4 This is a comparison chart of the effects of MESO and GMO on the estimation of external forces in this invention;

[0048] Figure 5 This is a diagram of the dynamic Bayesian collision detection process during collisions in this invention. Part (a) shows the variation of differential torque during collisions, part (b) shows the variation of differential velocity during collisions, part (c) shows the variation of Mahalanobis distance during collisions, part (d) shows the variation of basis functions during collisions, part (e) shows the variation of Bayesian probability during collisions, and part (f) shows the subdomain transition during collisions.

[0049] Figure 6 The diagram shows the collision detection experiment of the present invention, wherein (a) is a schematic diagram of the initial position; and (b) is a schematic diagram of the collision detection and emergency stop response. Detailed Implementation

[0050] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0051] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0052] The equations of motion for the joint space of an n-degree-of-freedom robotic arm can be written as:

[0053]

[0054] In the formula, , For the machine inertia matrix, The Cocteau force matrix, For the gravity matrix, Here is the friction force matrix. For joint control torque, This refers to the external torque on the joint.

[0055] The brakes of each joint of a robot typically consist of servo motors and a transmission system. The compliance of the transmission system, the inertia of the motor, and friction are represented as follows:

[0056] , ;

[0057] in These are the diagonal elements, damping coefficient, and joint stiffness coefficient of the motor rotor inertia matrix, respectively. This represents the electromagnetic torque of the motor and is considered the input to the system. The motor position is obtained from the motor-side encoder. This is the frictional torque.

[0058] See Figure 1 This embodiment presents an improved MESO collision detection method based on real-time differential joint torques of collaborative robots, comprising the following steps:

[0059] S1. In process 1, single-joint friction identification is used to obtain the four parameters of the Stribeck friction model. This serves as the input for offline dataset sampling (process 2) and online collision detection process (process 3) state acquisition;

[0060] Understandably, employing a nonlinear friction model can improve the accuracy of friction in the low-speed region. For the i-th joint, a friction torque identification experiment is conducted using a single joint design to establish the relationship between joint angular velocity and friction torque. The robot joint trajectory is then planned according to a trapezoidal velocity. The Stirbeck friction model can be expressed as:

[0061]

[0062] in, Let be the static friction term of the i-th joint. Coulomb friction term, For Stribeck speed, This is the term of viscous friction. It is an exponential factor.

[0063] Taking advantage of the fact that the joint angular acceleration is zero when the joint moves at a constant velocity, the measured torque is not affected by the inertial torque. Furthermore, since the joint speed is low during movement, the Coriolis force is approximately negligible. Therefore, the robot's dynamic equation is equivalent to:

[0064]

[0065] In the experiment, the joint angle was gradually increased, and the torque values ​​sampled during positive uniform motion and negative uniform motion were measured respectively. The experimental data for both sets were taken during the uniform acceleration segment. Frictional torque and uniform deceleration phase By subtracting the frictional torques at corresponding time points and then averaging them, the influence of gravity can be eliminated, yielding the frictional torque at that uniform velocity. The established velocity-torque relationship is as follows: Figure 2 As shown.

[0066] Furthermore, the frictional torque at various speeds was statistically analyzed, and the Stribeck model was obtained by fitting the data using the nonlinear least squares method.

[0067] ;

[0068] in Here, N represents the model parameters, and N represents the number of data points. and The torque and velocity measurements are for the Kth data point of joint i. The predicted frictional torque is calculated using the Stribeck model.

[0069] In step S1, a complex model of a flexible joint robot is constructed by combining the dynamic equations with the motor transmission coefficient equations. The motor transmission coefficient equation model is regarded as a perturbation factor acting on the dynamic model of the main rigid robot.

[0070] S2. In the offline dataset sampling process (process 2), the improved extended state observer MESO is used to estimate the external torque, the torque and velocity methods are differentially processed, and non-collision dataset samples are collected during the operation of the robotic arm in free space. And data samples collected when robots collide. .

[0071] The core idea of ​​traditional extended state observers is to achieve nonlinear perturbation estimation by enhancing the state vector. Based on a general model of a second-order MIMO system:

[0072] ;

[0073] in It is a state vector. It is system input. It is an unknown external input. This represents the total disturbance, including internal dynamics and external disturbances.

[0074] In step S2 of this embodiment, the improved extended state observer is used as the external force estimate during collision detection, and then...

[0075] ;

[0076] Based on the joint space dynamics equations of an n-degree-of-freedom robotic arm, the acceleration can be expressed as:

[0077]

[0078] The above acceleration is obtained by multiplying both sides of the equation on the left. available:

[0079]

[0080] Define momentum After differentiation, we get:

[0081]

[0082] So there is. ;

[0083] intermediate variables Expressed as:

[0084]

[0085] Based on consideration The equations of motion for the joint space of a multi-degree-of-freedom robotic arm can be obtained as follows:

[0086]

[0087] Therefore, based on the improved extended second-order state observer, it can be expressed as:

[0088] ;

[0089] The observation state is designed as follows:

[0090] ;

[0091] make State variables The estimated value, State variables The predicted value, and It is a constant.

[0092] Compared to traditional GMO, MESO has the same input parameters, but the MESO method provides a better estimated state for external torque through higher-order design. Comparison experimental figures are shown below. Figure 3 As shown, this design can effectively improve observation performance when there is torque disturbance in the robot system.

[0093] Model-based external force observers are highly dependent on the accuracy of the model, and uncertainties in the dynamic model can significantly impact the estimation of external forces. In the field of collision detection, since collision is a binary state consisting of two mutually exclusive states, collision and non-collision, the collision problem can be transformed into a 0-1 problem by performing differential processing on the external forces. This method can effectively address the estimation error of external torque caused by inaccurate dynamic models, because the torque value processed by differential processing fluctuates around 0, reducing the dependence on model accuracy. There are comparative experiments with and without differential torque, such as... Figure 4 As shown.

[0094] S3. In the offline dataset sampling process (process 2), calculate the mean of the non-collision dataset and the collision dataset respectively. Covariance This serves as the input for the online collision detection process.

[0095] In this embodiment, a dynamic Bayesian network is used to transform traditional threshold-based collision detection into a probabilistic model that propagates along a time series direction. An exponential basis function, designed from a large amount of process data in an offline process, is used as a priori formula for collision detection and then incorporated into the collision process for real-time collision probability updates.

[0096] In this embodiment, the acquisition of offline sampling datasets also incorporates joint differential velocity state variables for collision detection, by collecting non-collision dataset samples of the robot's movement in free space. and collision dataset samples Calculate the mean for both non-collision and collision datasets. and covariance and .

[0097] S4. In the offline dataset sampling process (process 2), calculate the Mahalanobis distance between the samples in the non-collision dataset and the collision dataset, respectively. Divide the domain into subdomains based on Mahalanobis distance and obtain the values ​​of the Mahalanobis distance subdomain boundaries. The coefficients of the exponential basis function are calculated using this numerical value. .

[0098] In this embodiment, the exponential basis function is designed by using Mahalanobis distance to divide the collision subdomains and obtaining the coefficients of the exponential basis function at the boundary.

[0099] Solve for Mahalanobis distance between offline non-collision and collision dataset samples. and And solve for the boundary subdomain distance subdomain. and As a completely non-collision subfield and fully collided subfield The division; in the subdomain The edge, calculated area ,in In the subdomain Edge, computation ,in .

[0100] In this embodiment, during the sampling process of the offline collision dataset, tennis balls are randomly struck at any part of the robotic arm operating in free space. A model-based, sensorless external force estimation method is employed. Since the last three axes of the 6-axis robot are relatively small, the collision force does not fluctuate significantly. Therefore, only the torque and velocity of the robot's first three axes are sampled and differentially processed in a real-time thread. The peak values ​​of the differential collision torque and velocity are selected, and values ​​with smaller collision torques are removed. Data with consistent peak values ​​are selected, and the fitted exponential prior basis function is dynamically updated in real-time during the online collision detection process to detect collisions promptly.

[0101] In this embodiment, the offline phase non-collision dataset sampling adopts the fifth-order Fourier series trajectory generated by the robotic arm in free space:

[0102] ;

[0103] in, for Joint position in the joint's trajectory Let be the order of the Fourier series. and Let be the amplitudes of the sine and cosine functions. The fundamental frequency of the Fourier series is used. To ensure the quality of the excitation trajectory while reducing the computational load, a 5th-order Fourier series is employed. .

[0104] Because the robot operates in a collision-free free space, the non-collision dataset samples... Since the data distribution is always near 0, it is approximately an unbiased sample. Therefore, in the design of non-collision basis functions, the sample mean is assumed to be... .

[0105] In this embodiment, Mahalanobis distance determines the basis for subdomain division in the projected observation domain. In the middle, the Mahalanobis distance satisfies The elliptical region is divided into As the entire non-collision sample The distribution area. Similarly, in the collision region, the Mahalanobis distance satisfies The elliptical region is divided into As the entire non-collision sample The distribution area. (By...) Extended Mahalanobis distance to and The tangent region, and from Extended Mahalanobis distance to and The tangent region is defined as the region where these two regions intersect. ;satisfy The region is defined as Subdomain; satisfying The region is defined as Subdomain; if the Mahalanobis distance satisfies These are defined as abnormal regions. After dividing the six regions according to Mahalanobis distance, the transition process of the collision state can be objectively observed as the Mahalanobis distance changes when a collision occurs in the robot system.

[0106] In this embodiment, the design principle of the exponential basis function is based on and The Markov transition matrix is The form of the exponential basis function is: .

[0107] S5. In the online collision detection process (process 3), the measurement quantity of the current cycle is acquired in real time. The algorithm calculates the Mahalanobis distance and collision / non-collision basis function values ​​for the current period, updates the confidence level for the current period based on the current measurement and the confidence level of the previous period, and determines whether a collision has occurred by comparing the magnitudes of the collision probability and the non-collision probability.

[0108] In this embodiment, the dynamic Bayesian formula combines the designed prior exponential basis function, Markov transition matrix, Bayesian probability of the previous week, and normalization factor. The collision Bayesian state is obtained by transitioning from the non-collision state and collision state of the previous week. Therefore, the Bayesian probability formula is as follows:

[0109] ;

[0110] In this embodiment, the confidence level is updated periodically online: ,in It is obtained through iteration, and each cycle is performed. The probability of collision is compared to that of non-collision in order to distinguish the collision state.

[0111] In this embodiment, the collision detection parameters for the robotic arm are set as follows. Taking the robot's two axes as an example, the mean value of the robot's non-collision samples is... The mean of the collision samples is The inverse of the covariance of non-collision samples is The inverse of the collision sample covariance is The coefficients of the non-collision basis functions are , The values ​​of the collision basis functions are , The Markov transition matrix is ​​set as follows: The first cycle of the robot's operation is considered a completely non-collision cycle, i.e. , The collision detection Bayesian probability is derived from the first cycle in the second cycle, thus solving the initial confidence problem.

[0112] Furthermore, such as Figure 6 As shown in parts (a) and (b) of the document, the experimental process of dynamic Bayesian collision detection based on the improved extended state observer in this embodiment is as follows:

[0113] The robot is programmed to run a pre-defined joint trajectory. During the real-time program execution and before the joints are powered on, the robot calculates the trajectory and writes it back to its original position. The robot remains stationary and then runs the pre-defined trajectory. First, the robot moves away from the experimenter, then moves closer to them. A soft foam object can be used as the collision target. The collision generates a sudden change in differential torque and differential velocity, such as... Figure 5 As shown in parts (a) and (b) of the diagram. When a collision occurs, the non-collision Mahalanobis distance increases, while the collision Mahalanobis distance decreases, as... Figure 5 As shown in (c) in the diagram. The Bayesian probability change upon collision is as follows: Figure 5 As shown in section (e). At this point, the robot will move from a completely non-collision domain to a collision domain. The subdomain transition upon collision is as follows: Figure 5 As shown in part (f) of the diagram. Due to the design of the exponential state basis functions, the value of the collision basis function will increase sharply upon a collision, as shown in part (f). Figure 5 As shown in section (d), the rapid change in the state basis function causes the Bayesian collision probability to quickly exceed the Bayesian non-collision probability, thus triggering the system to detect a collision and initiate alarm and emergency stop operations. It is important to note that since the collision detection device does not incorporate a vision camera, emergency stop is a relatively effective collision response measure when a collision with a person or object occurs. If the robot attempts to return to its original position or take other response measures after a collision, a secondary collision may occur, causing unnecessary secondary damage to the robot.

[0114] It should be noted that, for the sake of simplicity, the aforementioned method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously.

[0115] Based on the same idea as the improved MESO collision detection method based on real-time differential joint torques of collaborative robots in the above embodiments, this invention also provides an improved MESO collision detection system based on real-time differential joint torques of collaborative robots. This system can be used to execute the above-described improved MESO collision detection method based on real-time differential joint torques of collaborative robots. For ease of explanation, the structural schematic diagram of the embodiment of the improved MESO collision detection system based on real-time differential joint torques of collaborative robots only shows the parts related to the embodiments of this invention. Those skilled in the art will understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0116] In another embodiment of this application, an improved MESO collision detection system based on real-time differential joint torque of collaborative robots is provided. The system includes a parameter acquisition module, a differential processing module, an offline data statistics module, a basis function coefficient solving module, and an online collision judgment module.

[0117] The parameter acquisition module uses single-joint friction force identification to obtain four parameters of the Stribeck friction force model. The four parameters are used as inputs for the offline dataset sampling process and the online collision detection process state estimation;

[0118] The differential processing module, in the offline dataset sampling process, uses an improved Extended State Observer (MESO) to estimate external torques, performs differential processing on torque and velocity state variables, and collects non-collision dataset samples of the robotic arm operating in free space. And dataset samples when robots collide. ;

[0119] The offline data statistics module calculates the mean values ​​of the non-collision dataset and the collision dataset respectively during the offline dataset sampling process. Covariance , as input to the online collision detection process;

[0120] The basis function coefficient solving module calculates the Mahalanobis distance between samples in the non-collision dataset and the collision dataset respectively during the offline dataset sampling process. Divide the domain into subdomains based on Mahalanobis distance and obtain the values ​​of the Mahalanobis distance subdomain boundaries. The coefficients of the exponential basis function are calculated using this numerical value. ;

[0121] The online collision detection module acquires the measurement value of the current period in real time during the online collision detection process. The algorithm calculates the Mahalanobis distance and collision / non-collision basis function values ​​for the current period, updates the confidence level for the current period based on the current measurement and the confidence level of the previous period, and determines whether a collision has occurred by comparing the magnitudes of the collision probability and the non-collision probability.

[0122] It should be noted that the improved MESO collision detection system based on real-time differential joint torque of collaborative robots of the present invention corresponds one-to-one with the improved MESO collision detection method based on real-time differential joint torque of collaborative robots of the present invention. The technical features and beneficial effects described in the embodiments of the improved MESO collision detection method based on real-time differential joint torque of collaborative robots are applicable to the embodiments of the improved MESO collision detection based on real-time differential joint torque of collaborative robots. For details, please refer to the description in the embodiments of the method of the present invention, which will not be repeated here.

[0123] Furthermore, in the above embodiments of the improved MESO collision detection system based on real-time differential joint torque of collaborative robots, the logical division of each program module is only an example. In actual applications, the above functions can be assigned to different program modules as needed, for example, for the sake of corresponding hardware configuration requirements or software implementation convenience. That is, the internal structure of the improved MESO collision detection system based on real-time differential joint torque of collaborative robots is divided into different program modules to complete all or part of the functions described above.

[0124] In one embodiment, an electronic device is provided for implementing an improved MESO collision detection method based on real-time differential joint torques of collaborative robots. The electronic device may include a first processor, a first memory, and a bus, and may also include a computer program stored in the first memory and executable on the first processor, such as an improved MESO collision detection program based on real-time differential joint torques of collaborative robots.

[0125] The first memory includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the first memory can be an internal storage unit of an electronic device, such as a portable hard drive. In other embodiments, the first memory can be an external storage device of the electronic device, such as a plug-in portable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc. Furthermore, the first memory can include both internal and external storage units of the electronic device. The first memory can be used not only to store application software and various types of data installed on the electronic device, such as the code of the improved MESO collision detection program based on real-time differential joint torques of collaborative robots, but also to temporarily store data that has been output or will be output.

[0126] In some embodiments, the first processor may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The first processor is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the first memory and calls data stored in the first memory to perform various functions of the electronic device and process data.

[0127] The improved MESO collision detection program based on real-time differential joint torques of collaborative robots, stored in the first memory of the electronic device, is a combination of multiple instructions. When run in the first processor, it can achieve the following:

[0128] S1. Using single-joint friction identification, four parameters of the Stribeck friction model are obtained. The four parameters are used as inputs for the offline dataset sampling process and the online collision detection process state estimation;

[0129] S2. In the offline dataset sampling process, the improved Extended State Observer (MESO) is used to estimate the external torque, the torque and velocity state variables are differentially processed, and non-collision dataset samples are collected during the free space operation of the robotic arm. And dataset samples when robots collide. ;

[0130] S3. In the offline dataset sampling process, calculate the mean of the non-collision dataset and the collision dataset respectively. Covariance , as input to the online collision detection process;

[0131] S4. In the offline dataset sampling process, calculate the Mahalanobis distance between samples in the non-collision dataset and the collision dataset, respectively. Divide the domain into subdomains based on Mahalanobis distance and obtain the values ​​of the Mahalanobis distance subdomain boundaries. The coefficients of the exponential basis function are calculated using this numerical value. ;

[0132] S5. In the online collision detection process, acquire the measurement data for the current cycle in real time. The algorithm calculates the Mahalanobis distance and collision / non-collision basis function values ​​for the current period, updates the confidence level for the current period based on the current measurement and the confidence level of the previous period, and determines whether a collision has occurred by comparing the magnitudes of the collision probability and the non-collision probability.

[0133] Furthermore, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0134] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0135] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0136] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. An improved MESO collision detection method based on real-time differential joint torques of collaborative robots, characterized in that, Includes the following steps: S1. Using single-joint friction identification, four parameters of the Stribeck friction model are obtained. The four parameters are used as inputs for the offline dataset sampling process and the online collision detection process state estimation; S2. In the offline dataset sampling process, the improved Extended State Observer (MESO) is used to estimate the external torque, the torque and velocity state variables are differentially processed, and non-collision dataset samples are collected during the free space operation of the robotic arm. And dataset samples when robots collide. ; S3. In the offline dataset sampling process, calculate the mean of the non-collision dataset and the collision dataset respectively. Covariance Input for the online collision detection process; S4. In the offline dataset sampling process, calculate the Mahalanobis distance between samples in the non-collision dataset and the collision dataset, respectively. Divide the domain into subdomains based on Mahalanobis distance and obtain the values ​​of the Mahalanobis distance subdomain boundaries. The coefficients of the exponential basis function are calculated using this numerical value. ; S5. In the online collision detection process, acquire the measurement data for the current cycle in real time. The algorithm calculates the Mahalanobis distance and collision / non-collision basis function values ​​for the current period, updates the confidence level for the current period based on the current measurement and the confidence level of the previous period, and determines whether a collision has occurred by comparing the magnitudes of the collision probability and the non-collision probability.

2. The improved MESO collision detection method based on real-time differential joint torques of collaborative robots according to claim 1, characterized in that, In step S1, the Stribeck friction model used is a nonlinear friction model for the i-th joint, and its expression is: in, Let be the static friction term of the i-th joint. Coulomb friction term, For Stribeck speed, This is the term of viscous friction. It is an exponential factor.

3. The improved MESO collision detection method based on real-time differential joint torques of collaborative robots according to claim 2, characterized in that, In step S1, the nonlinear least squares method is used to minimize the objective function to identify the parameters of the Stribeck friction model. The objective function is: ； in, Here, N represents the model parameters, and N represents the number of data points. and The torque and velocity measurements are for the Kth data point of joint i. The predicted frictional torque is calculated using the Stribeck model.

4. The improved MESO collision detection method based on real-time differential joint torques of collaborative robots according to claim 1, characterized in that, In step S2, when using MESO to estimate the external moment during the collision, the joint moments are... Converted into joint differential torque Furthermore, joint differential velocity is introduced. Collision detection is performed using state variables, and non-collision dataset samples are collected during robot motion in free space. and collision dataset samples .

5. The improved MESO collision detection method based on real-time differential joint torques of collaborative robots according to claim 4, characterized in that, The MESO formula for estimating external torque is: in, , For robot control torque, The Cocteau force matrix, For the gravity matrix, Let be the friction force matrix, and let State variables The estimated value, State variables The predicted value, and It is a constant.

6. The improved MESO collision detection method based on real-time differential joint torques of collaborative robots according to claim 1, characterized in that, Step S4 is as follows: Solve for Mahalanobis distance between offline non-collision and collision dataset samples. and And solve for the boundary subdomain distance subdomain. and As a completely non-collision subfield and fully collided subfield The division; in the subdomain The edge, calculated area ,in In the subdomain Edge, computation ,in According to the design principles of exponential basis functions, based on and The Markov transition matrix is The form of the exponential basis function is: .

7. The improved MESO collision detection method based on real-time differential joint torques of collaborative robots according to claim 1, characterized in that, In step S5, the confidence update formula for each period is: ； in, Obtained through iteration. As the normalization factor, It is a Markov matrix. Each cycle is compared and The size of the collision determines the collision state, among which Indicates the collision state. Indicates a non-collision state.

8. An improved MESO collision detection system based on real-time differential joint torques of collaborative robots, characterized in that, The improved MESO collision detection method based on real-time differential joint torque of collaborative robots, applied to any one of claims 1-7, includes a parameter acquisition module, a differential processing module, an offline data statistics module, a basis function coefficient solving module, and an online collision judgment module; The parameter acquisition module uses single-joint friction force identification to obtain four parameters of the Stribeck friction force model. The four parameters are used as inputs for the offline dataset sampling process and the online collision detection process state estimation; The differential processing module, in the offline dataset sampling process, uses an improved Extended State Observer (MESO) to estimate external torques, performs differential processing on torque and velocity state variables, and collects non-collision dataset samples of the robotic arm operating in free space. And dataset samples when robots collide. ; The offline data statistics module calculates the mean values ​​of the non-collision dataset and the collision dataset respectively during the offline dataset sampling process. Covariance , as input to the online collision detection process; The basis function coefficient solving module calculates the Mahalanobis distance between samples in the non-collision dataset and the collision dataset respectively during the offline dataset sampling process. Divide the domain into subdomains based on Mahalanobis distance and obtain the values ​​of the Mahalanobis distance subdomain boundaries. The coefficients of the exponential basis function are calculated using this numerical value. ; The online collision detection module acquires the measurement value of the current period in real time during the online collision detection process. The algorithm calculates the Mahalanobis distance and collision / non-collision basis function values ​​for the current period, updates the confidence level for the current period based on the current measurement and the confidence level of the previous period, and determines whether a collision has occurred by comparing the magnitudes of the collision probability and the non-collision probability.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores computer program instructions that can be executed by the at least one processor, which enables the at least one processor to perform an improved MESO collision detection method based on real-time differential joint torques of collaborative robots as described in any one of claims 1-7.

10. A computer-readable storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the improved MESO collision detection method based on real-time differential joint torque of collaborative robots as described in any one of claims 1-7.

Images