Mechanical arm real-time control method and system based on edge computing

Abstract

The application provides an edge-computing-based real-time control method and system for a mechanical arm, relates to the technical field of mechanical arm control, and comprises the following steps: acquiring state information of the mechanical arm and a task instruction; performing semantic analysis at an edge node and converting the semantic analysis into a joint space trajectory sequence; performing time domain decomposition to generate an execution section and a cache section; generating a motion plan based on a time optimization criterion and calculating a control instruction; acquiring an actual state to calculate a deviation; when the deviation exceeds a threshold value, calculating a deviation influence quantity and generating a compensation quantity to correct the cache section and realize real-time control. The application realizes high-precision real-time control and intelligent correction of the mechanical arm.

Description

Technical Field

[0001] This invention relates to the field of robotic arm control technology, and in particular to a real-time control method and system for robotic arms based on edge computing. Background Technology

[0002] With the rapid development of Industry 4.0 and intelligent manufacturing, robotic arms, as important automated devices, are widely used in industrial production, medical services, and home assistance. Traditional robotic arm control systems typically employ a centralized architecture, concentrating control computation tasks on a central server or controller. However, as robotic arm tasks become more complex and precision requirements increase, the demand for real-time control is growing. Edge computing, as an emerging technology that deploys computing resources at the network edge, can effectively reduce data transmission latency and improve processing efficiency, providing a new technological path for real-time control of robotic arms.

[0003] In traditional robotic arm control systems, a semantic gap exists between task commands and low-level control, making it difficult to effectively translate from high-level task descriptions to specific joint controls. This results in slow system response and difficulty adapting to complex and changing operating environments. Most existing robotic arm control methods employ pre-planned trajectory execution patterns, lacking the ability to adjust in real-time based on actual conditions during operation. When external environmental changes or deviations occur during execution, timely and effective corrections are impossible, affecting the accuracy and robustness of task execution. Current robotic arm control systems lack in-depth research on multi-joint collaborative control, making it difficult to handle the coupling effects between joints. When deviations occur, independent compensation for each joint is often used, ignoring the interactions between joints, leading to poor compensation effects and even causing system oscillations or instability. Summary of the Invention

[0004] This invention provides a real-time control method and system for robotic arms based on edge computing, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a real-time control method for a robotic arm based on edge computing, comprising:

[0006] Obtain the current status information and target task instructions of the robotic arm;

[0007] The target task instructions are semantically parsed at the edge computing node, and based on the mapping relationship between the joint motion range constraints and the robot arm workspace constraints, the target task instructions are converted into a joint space trajectory sequence.

[0008] The joint space trajectory sequence is decomposed in the time domain according to the availability of computing resources and control cycle constraints to generate the current execution segment and the prediction cache segment.

[0009] Based on the current execution segment and the current state information, a segmented motion planning result is generated at the edge computing node according to the time-optimal criterion; based on the segmented motion planning result, control commands for each joint are calculated at the edge computing node;

[0010] Obtain the actual state information of the robotic arm after executing the control command, and calculate the deviation information between the actual state information and the joint target position of the corresponding time segment in the segmented motion planning result;

[0011] When the deviation information exceeds a preset deviation threshold, the deviation transmission impact is calculated at the edge computing node, and a multi-joint collaborative compensation amount is generated based on the deviation transmission impact. Based on the deviation information and the multi-joint collaborative compensation amount, the prediction buffer segment is corrected at the edge computing node, and the corrected prediction buffer segment is used to control the robotic arm to perform motion correction in real time.

[0012] At the edge computing node, semantic parsing of the target task instructions is performed. Based on the mapping relationship between joint motion range constraints and robotic arm workspace constraints, the target task instructions are converted into a joint space trajectory sequence, including:

[0013] In the edge computing node, the target position and target posture of the robotic arm in the Cartesian coordinate system are extracted from the target task instruction through syntax structure decomposition;

[0014] By calculating the reachable space boundary of the robotic arm end effector when each joint of the robotic arm takes the value of the joint motion range constraint boundary, a two-way mapping relationship is formed from the joint angle limit value to the reachable range of the Cartesian space.

[0015] Determine whether the target position and the target posture are within the reachable space boundary of the robotic arm end effector. If so, use the bidirectional mapping relationship to transform the robotic arm workspace constraint into boundary limiting conditions for joint angle configuration, and generate each candidate joint angle configuration.

[0016] The robot arm end position corresponding to each candidate joint angle configuration is obtained by forward kinematics calculation. The spatial deviation between the robot arm end position and the target position is calculated. The candidate joint angle configuration with the smallest spatial deviation is selected as the effective joint angle configuration.

[0017] Based on the effective joint angle configuration and the joint position information in the current state information, an interpolation path connecting the current joint position and the effective joint angle configuration is generated based on the joint motion continuity constraint, and the interpolation path is arranged in a time sequence to form the joint spatial trajectory sequence.

[0018] The joint space trajectory sequence is decomposed in the time domain according to the constraints of computing resource availability and control cycle to generate the current execution segment and the prediction cache segment, including:

[0019] The processor utilization and available memory capacity of the edge computing node are obtained from the quantitative indicators of computing resource availability; the control cycle duration of the robotic arm is obtained as the time base for the control cycle constraint.

[0020] Calculate the joint angle change gradient of each trajectory point in the joint space trajectory sequence. If the joint angle change gradient exceeds a preset gradient threshold, it is marked as a high dynamic trajectory interval; otherwise, it is marked as a low dynamic trajectory interval.

[0021] The reduction coefficient is determined based on the ratio of the maximum value of the joint angle change gradient to the preset gradient threshold, and the expansion coefficient is determined based on the ratio of the minimum value of the joint angle change gradient to the preset gradient threshold.

[0022] The number of reference trajectory points that the edge computing node can process within a single control cycle is calculated based on the processor utilization rate. The number of trajectory points in the high dynamic trajectory interval is determined according to the number of reference trajectory points and the reduction coefficient. The number of trajectory points in the low dynamic trajectory interval is determined according to the number of reference trajectory points and the expansion coefficient.

[0023] When the available memory capacity of the edge computing node is sufficient to store the current execution segment and the prediction cache segment, starting from the start time point of the joint space trajectory sequence, the joint space trajectory sequence is non-uniformly segmented according to the number of trajectory points in the high dynamic trajectory interval and the low dynamic trajectory interval. The trajectory content of the non-uniform segmentation covers at least one complete control cycle duration corresponding to the time base. The segmented trajectory content is used as the current execution segment, and the remaining trajectory content is used as the prediction cache segment.

[0024] Based on the current execution segment and the current state information, the segmented motion planning results generated at the edge computing node according to the time-optimal criterion include:

[0025] Extract the current joint position and current joint speed of each joint of the robotic arm from the current state information, and extract the position of each target joint from the current execution segment;

[0026] Calculate the joint displacement required for each joint to move from the current joint position to the target joint position. Calculate the shortest motion time required for each joint to complete the corresponding joint displacement based on the upper limit of the velocity constraint of each joint. Select the maximum value of the shortest motion time corresponding to each joint as the global motion time benchmark of the current execution segment.

[0027] The motion of each joint is time-synchronized and scheduled according to the global motion time reference, and the joint velocity and joint acceleration values ​​of each joint during the motion are recorded.

[0028] The global motion time reference is discretized according to the control cycle duration to generate a discrete time sequence within the current execution segment.

[0029] The time interval between discrete moments is integrated based on the joint velocity value corresponding to each discrete moment to obtain the joint position increment of each joint relative to the current joint position. The joint position increment is then added to the current joint position to obtain the joint position value of each joint at each discrete moment.

[0030] The joint position values, joint velocity values, and joint acceleration values ​​at each discrete moment are organized in chronological order according to the discrete moment sequence to form the segmented motion planning result.

[0031] Based on the segmented motion planning results, the control commands for each joint are calculated at the edge computing node, including:

[0032] Based on the changing trends of the position and velocity deviations of each joint at the current discrete moment, it is determined whether each joint is in an oscillation state. The damping adjustment factor is calculated based on the oscillation frequency, and the control response gain of the joint in the oscillation state is reduced using the damping adjustment factor.

[0033] The load current value of each joint is used to determine whether there is a sudden change in load in each joint. For joints with sudden load changes, the load disturbance compensation amount is calculated based on the change in the load current value and the mass distribution parameters of each joint of the robotic arm.

[0034] Based on the damping adjustment factor and the load disturbance compensation amount, the error compensation amount of each joint is calculated, and the joint acceleration value of each joint at the current discrete moment is superimposed with the error compensation amount to obtain the expected joint acceleration of each joint.

[0035] Based on the desired joint acceleration of each joint and the coupling relationship between the joints of the robotic arm, the joint torque value required by each joint to achieve the desired joint acceleration is solved by inverse dynamics calculation; the joint torque value is then converted into the control command for each joint according to the torque constant of each joint actuator.

[0036] When the deviation information exceeds a preset deviation threshold, the deviation propagation impact is calculated at the edge computing node, and a multi-joint collaborative compensation amount is generated based on the deviation propagation impact, including:

[0037] Based on the deviation information, joints whose position deviation value exceeds a preset position deviation threshold or whose velocity deviation value exceeds a preset velocity deviation threshold are determined to be abnormal joints.

[0038] Based on the current configuration of the robotic arm, the Jacobian mapping relationship between each abnormal joint and the end effector is calculated, and the position deviation value and the speed deviation value of each abnormal joint are converted into the deviation transmission influence amount of the end effector through the Jacobian mapping relationship.

[0039] The deviation transmission effects of all abnormal joints are vectorized to obtain the total position offset vector and the total velocity offset vector. The degree and direction of deviation of the end effector are determined based on the magnitude and direction of the total position offset vector and the total velocity offset vector. The compensation priority weight is determined based on the degree and direction of deviation.

[0040] By performing inverse operations on the Jacobian matrix, the total position offset vector and the total velocity offset vector are mapped inversely to the joint space to obtain the joint compensation displacement and joint compensation velocity required by each joint to offset the total position offset vector and the total velocity offset vector of the end effector. The joint compensation displacement and joint compensation velocity are weighted and combined according to the compensation priority weight to form the multi-joint collaborative compensation amount.

[0041] Based on the deviation information and the multi-joint collaborative compensation amount, the prediction buffer segment is corrected at the edge computing node, and the corrected prediction buffer segment is used to control the robotic arm to perform motion correction in real time, including:

[0042] Extract the predicted joint position and predicted joint velocity of each joint at multiple discrete moments in the future from the prediction cache segment;

[0043] The deviation spatial gradient of each joint is calculated based on the deviation information. The cross-coupling deviation component of each joint's deviation on other joints at multiple discrete moments in the future is calculated based on the deviation spatial gradient. The position deviation value and velocity deviation value of each joint are superimposed with the cross-coupling deviation component to obtain the global coupling deviation amount of each joint.

[0044] The compensation bearing capacity coefficient of each joint is calculated based on the predicted joint position and the predicted joint velocity. The multi-joint collaborative compensation amount is distributed according to the compensation bearing capacity coefficient of each joint to obtain the distributed compensation component of each joint at multiple discrete moments in the future.

[0045] Calculate the compensation phase consistency factor based on the phase relationship between the distributed compensation component and the global coupling deviation, and use the compensation phase consistency factor to perform phase calibration on the distributed compensation component.

[0046] Using the global coupling deviation and the phase-calibrated distributed compensation component, motion correction is performed on the predicted joint position and the predicted joint velocity, and motion correction commands for each joint are generated. The robotic arm is then controlled to perform motion correction in real time according to the motion correction commands.

[0047] A second aspect of the present invention provides a real-time control system for a robotic arm based on edge computing, comprising:

[0048] The first unit is used to acquire the current status information of the robotic arm and the target task instructions;

[0049] The second unit is used to perform semantic parsing of the target task instructions at the edge computing node, and convert the target task instructions into a joint space trajectory sequence based on the mapping relationship between joint motion range constraints and robotic arm workspace constraints.

[0050] The third unit is used to decompose the joint space trajectory sequence in the time domain according to the availability of computing resources and control cycle constraints, and generate the current execution segment and the prediction cache segment.

[0051] The fourth unit is used to generate segmented motion planning results at the edge computing node based on the current execution segment and the current state information and the time-optimal criterion; and to calculate control commands for each joint at the edge computing node based on the segmented motion planning results.

[0052] The fifth unit is used to acquire the actual state information of the robotic arm after executing the control command, and to calculate the deviation information between the actual state information and the joint target position of the corresponding time segment in the segmented motion planning result;

[0053] The sixth unit is used to calculate the deviation transmission impact at the edge computing node when the deviation information exceeds a preset deviation threshold, generate a multi-joint collaborative compensation amount based on the deviation transmission impact amount, and correct the prediction buffer segment at the edge computing node based on the deviation information and the multi-joint collaborative compensation amount, and use the corrected prediction buffer segment to control the robotic arm to perform motion correction in real time.

[0054] A third aspect of the present invention,

[0055] An electronic device is provided, comprising:

[0056] processor;

[0057] Memory used to store processor-executable instructions;

[0058] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0059] Fourth aspect of the present invention,

[0060] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0061] The beneficial effects of this application are as follows:

[0062] By semantically parsing target task instructions at edge computing nodes and converting them into joint space trajectory sequences based on the mapping relationship between joint motion range constraints and workspace constraints, the parsing accuracy and execution efficiency of complex task instructions for the robotic arm are improved, while reducing reliance on cloud communication. A temporal decomposition strategy is employed to divide the joint space trajectory sequences into a current execution segment and a prediction cache segment, making rational use of edge computing resources. This approach ensures both real-time performance and computational efficiency, adapting to different computing resource constraints. Based on the time-optimal criterion, segmented motion planning results are generated at the edge, enabling the robotic arm to complete tasks at the fastest speed while satisfying kinematic and dynamic constraints. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating the real-time control method for a robotic arm based on edge computing, according to an embodiment of the present invention.

[0064] Figure 2 This is a schematic diagram of the segmented motion planning algorithm. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0067] Figure 1 This is a flowchart illustrating the real-time control method for a robotic arm based on edge computing, as described in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0068] Obtain the current status information and target task instructions of the robotic arm;

[0069] The target task instructions are semantically parsed at the edge computing node, and based on the mapping relationship between the joint motion range constraints and the robot arm workspace constraints, the target task instructions are converted into a joint space trajectory sequence.

[0070] The joint space trajectory sequence is decomposed in the time domain according to the availability of computing resources and control cycle constraints to generate the current execution segment and the prediction cache segment.

[0071] Based on the current execution segment and the current state information, a segmented motion planning result is generated at the edge computing node according to the time-optimal criterion; based on the segmented motion planning result, control commands for each joint are calculated at the edge computing node;

[0072] Obtain the actual state information of the robotic arm after executing the control command, and calculate the deviation information between the actual state information and the joint target position of the corresponding time segment in the segmented motion planning result;

[0073] When the deviation information exceeds a preset deviation threshold, the deviation transmission impact is calculated at the edge computing node, and a multi-joint collaborative compensation amount is generated based on the deviation transmission impact. Based on the deviation information and the multi-joint collaborative compensation amount, the prediction buffer segment is corrected at the edge computing node, and the corrected prediction buffer segment is used to control the robotic arm to perform motion correction in real time.

[0074] In one optional implementation, semantic parsing of the target task instructions is performed at the edge computing node. Based on the mapping relationship between joint motion range constraints and robotic arm workspace constraints, the target task instructions are converted into a joint space trajectory sequence, including:

[0075] In the edge computing node, the target position and target posture of the robotic arm in the Cartesian coordinate system are extracted from the target task instruction through syntax structure decomposition;

[0076] By calculating the reachable space boundary of the robotic arm end effector when each joint of the robotic arm takes the value of the joint motion range constraint boundary, a two-way mapping relationship is formed from the joint angle limit value to the reachable range of the Cartesian space.

[0077] Determine whether the target position and the target posture are within the reachable space boundary of the robotic arm end effector. If so, use the bidirectional mapping relationship to transform the robotic arm workspace constraint into boundary limiting conditions for joint angle configuration, and generate each candidate joint angle configuration.

[0078] The robot arm end position corresponding to each candidate joint angle configuration is obtained by forward kinematics calculation. The spatial deviation between the robot arm end position and the target position is calculated. The candidate joint angle configuration with the smallest spatial deviation is selected as the effective joint angle configuration.

[0079] Based on the effective joint angle configuration and the joint position information in the current state information, an interpolation path connecting the current joint position and the effective joint angle configuration is generated based on the joint motion continuity constraint, and the interpolation path is arranged in a time sequence to form the joint spatial trajectory sequence.

[0080] In edge computing nodes, a syntactic structure decomposition method is used to extract the target position and orientation of the robotic arm in the Cartesian coordinate system from the target task instructions. For example, when receiving the instruction "Move the object to position (0.5, 0.3, 0.2) with an orientation (roll=30°, pitch=45°, yaw=60°)," the edge computing node will identify the target position coordinates (0.5, 0.3, 0.2) and the target orientation (30°, 45°, 60°). Natural language processing technology is used in the syntactic structure decomposition process to segment the instruction into words, identify key action words such as "move," "grasp," and "place," and extract position-related numerical information. For complex instructions, such as "Move the red square on the table to 10 centimeters to the right of the blue cylinder," semantic analysis is needed to extract relative position information, and the actual target position coordinates are obtained in combination with a visual perception system.

[0081] After extracting the target position and orientation, it is determined whether these targets are within the reachable space of the robotic arm. Through forward kinematics calculations, the workspace boundaries of each joint of the robotic arm under range-of-motion constraints are determined. For a six-DOF robotic arm, each joint has maximum and minimum angle constraints, such as the first joint's angle range being [-180°, 180°], the second joint's range being [-90°, 90°], and so on. These joint angle constraints are mapped to the reachable range of the robotic arm's end effector in Cartesian space using forward kinematics equations, forming a discrete set of points. For example, for a six-DOF robotic arm, the range of motion of each joint can be divided into ten equal parts, forming one million joint angle combinations. Each combination is used to calculate the corresponding end effector position and orientation through forward kinematics, generating a point cloud in the workspace. Using this point cloud data, a geometric model of the workspace boundary is constructed, such as a convex hull or voxel model, to quickly determine whether the target point is reachable.

[0082] A spatial geometry detection algorithm is used to determine whether the target position and orientation are within the reachable space boundary of the robotic arm's end effector. If the target point is within the reachable space, a bidirectional mapping relationship is used to transform the workspace constraints into boundary conditions for joint angles. An inverse kinematics algorithm is then used to generate candidate joint angle configurations that meet the conditions, such as analytical or numerical iterative methods. Analytical methods are suitable for robotic arms with specific structures, while numerical iterative methods have wider applicability. To improve solution efficiency, a hierarchical search strategy can be adopted: coarse sampling is performed in the joint space, such as dividing each joint into five equal parts to obtain an initial candidate set; forward kinematic calculations are performed on these candidate points to evaluate the deviation between the end effector position and the target position; a subset with smaller deviations is selected, and a more detailed search is performed in its surrounding area until a solution that meets the accuracy requirements is found.

[0083] For redundant degree-of-freedom robotic arms, there may be situations where multiple joint angle configurations can reach the same end-effector position. In such cases, additional optimization objectives are needed, such as minimizing energy consumption or joint angle variations. For example, the Euclidean distance between each candidate configuration and the current joint position can be calculated, and the configuration with the smallest distance can be selected to reduce large-amplitude movements of the robotic arm.

[0084] After calculating the end-effector position corresponding to each candidate joint angle configuration using forward kinematics, the spatial deviation between the end-effector position and the target position is calculated. The spatial deviation is typically measured using Euclidean distance: Spatial Deviation = Euclidean distance between the end-effector position coordinates and the target position coordinates + Attitude Deviation Weighting Coefficient × Attitude Deviation. The attitude deviation can be calculated using the difference in rotation matrices or the difference in Euler angles. The candidate joint angle configuration with the smallest spatial deviation is selected as the effective joint angle configuration.

[0085] Based on the effective joint angle configuration and the joint position information in the current state information, an interpolation path is generated based on joint motion continuity constraints. The interpolation algorithm typically employs a polynomial interpolation method, such as fifth-order polynomial interpolation, to ensure the continuity of joint position, velocity, and acceleration at the start and end points of the path. For example, for interpolation from the current joint position q1 to the target joint position q2, a path function can be designed so that the velocity and acceleration of the joint at the start and end points are zero, ensuring smooth motion.

[0086] The generated interpolation paths are arranged according to a time series, forming a joint space trajectory sequence. The planning of the time series needs to consider the dynamic characteristics of the robotic arm, such as maximum joint velocity and acceleration limits. A typical trajectory sequence contains uniformly sampled joint position points, with sampling intervals typically on the order of milliseconds, to meet real-time control requirements.

[0087] In one optional implementation, the joint space trajectory sequence is decomposed in the time domain according to the availability of computing resources and control cycle constraints to generate the current execution segment and the prediction cache segment, including:

[0088] The processor utilization and available memory capacity of the edge computing node are obtained from the quantitative indicators of computing resource availability; the control cycle duration of the robotic arm is obtained as the time base for the control cycle constraint.

[0089] Calculate the joint angle change gradient of each trajectory point in the joint space trajectory sequence. If the joint angle change gradient exceeds a preset gradient threshold, it is marked as a high dynamic trajectory interval; otherwise, it is marked as a low dynamic trajectory interval.

[0090] The reduction coefficient is determined based on the ratio of the maximum value of the joint angle change gradient to the preset gradient threshold, and the expansion coefficient is determined based on the ratio of the minimum value of the joint angle change gradient to the preset gradient threshold.

[0091] The number of reference trajectory points that the edge computing node can process within a single control cycle is calculated based on the processor utilization rate. The number of trajectory points in the high dynamic trajectory interval is determined according to the number of reference trajectory points and the reduction coefficient. The number of trajectory points in the low dynamic trajectory interval is determined according to the number of reference trajectory points and the expansion coefficient.

[0092] When the available memory capacity of the edge computing node is sufficient to store the current execution segment and the prediction cache segment, starting from the start time point of the joint space trajectory sequence, the joint space trajectory sequence is non-uniformly segmented according to the number of trajectory points in the high dynamic trajectory interval and the low dynamic trajectory interval. The trajectory content of the non-uniform segmentation covers at least one complete control cycle duration corresponding to the time base. The segmented trajectory content is used as the current execution segment, and the remaining trajectory content is used as the prediction cache segment.

[0093] Edge computing nodes periodically collect processor utilization data through the system monitoring module. This data is expressed as a percentage; for example, a current processor utilization of 60% means that 40% of the processing power is still available. The available storage capacity is queried through the memory management interface, recording the currently available storage space in bytes, such as 32MB of storage space remaining.

[0094] The control cycle duration of the robotic arm is obtained as the time reference for the control cycle constraint. The robotic arm control system usually has a fixed control cycle, such as the common industrial robotic arm which uses a control cycle of 8ms or 10ms. The time reference determines the time granularity of trajectory planning and execution.

[0095] For each point in the trajectory sequence, the angular difference between adjacent trajectory points is calculated and divided by the time interval to obtain the angular velocity value as the gradient of the joint angle change. A preset gradient threshold is set, such as 5 degrees / second. Trajectory points exceeding this threshold are marked as high-dynamic trajectory intervals, otherwise they are marked as low-dynamic trajectory intervals. High-dynamic intervals typically correspond to the stages of rapid movement or drastic changes in direction of the robotic arm, requiring denser trajectory points to ensure control accuracy.

[0096] The reduction coefficient is determined based on the ratio of the maximum value of the joint angle change gradient to a preset gradient threshold. Assuming the maximum value of the joint angle change gradient is 10 degrees / second and the preset gradient threshold is 5 degrees / second, the ratio is 2. Therefore, the reduction coefficient can be set to 0.5, meaning the trajectory point spacing in high-dynamic regions will be reduced to half the baseline value, increasing sampling density. Similarly, the expansion coefficient is determined based on the ratio of the minimum value of the joint angle change gradient to a preset gradient threshold. If the minimum value is 1 degree / second, the ratio is 0.2, and the expansion coefficient can be set to 2. This means that in low-dynamic regions, the trajectory point spacing can be doubled to the baseline, reducing computational load.

[0097] The number of baseline trajectory points that an edge computing node can process within a single control cycle is calculated based on processor utilization. For example, if a control cycle is 10ms, and the edge computing node can process 5 trajectory points per millisecond under the current processor load, then the number of baseline trajectory points is 50. Multiplying the number of baseline trajectory points by a reduction factor yields the number of trajectory points in the high-dynamic trajectory range, such as 50 × 0.5 = 25. Multiplying the number of baseline trajectory points by an expansion factor yields the number of trajectory points in the low-dynamic trajectory range, such as 50 × 2 = 100.

[0098] When the available memory capacity of the edge computing node meets the storage requirements, starting from the initial time point of the joint space trajectory sequence, the sequence is non-uniformly segmented based on the number of trajectory points in the high-dynamic and low-dynamic trajectory intervals. Specifically, in the high-dynamic region, every 25 trajectory points constitute a segment; in the low-dynamic region, every 100 trajectory points constitute a segment. It is ensured that the non-uniformly segmented trajectory content covers at least one complete control cycle corresponding to the time base. For example, if the control cycle is 10ms, it is necessary to ensure that the segmented trajectory covers at least 10ms of time. The segmented trajectory content is used as the current execution segment for the immediate control of the robotic arm; the remaining trajectory content is stored in the edge computing node as a prediction cache segment, awaiting subsequent calls.

[0099] In practical applications, such as when a robotic arm performs material sorting tasks, there are two action modes: grasping and placing. Grasping actions typically require precise positioning and fall within a high-dynamic trajectory range, requiring more computational resources and a higher trajectory point density. During material transport, the robotic arm's movement is relatively smooth and is marked as a low-dynamic trajectory range, allowing for a lower trajectory point density and saving computational resources. This time-domain decomposition method ensures both the control accuracy of key actions and optimizes the system's resource utilization efficiency.

[0100] In one optional implementation, generating segmented motion planning results at the edge computing node based on the current execution segment and the current state information, according to the time-optimal criterion, includes:

[0101] Extract the current joint position and current joint speed of each joint of the robotic arm from the current state information, and extract the position of each target joint from the current execution segment;

[0102] Calculate the joint displacement required for each joint to move from the current joint position to the target joint position. Calculate the shortest motion time required for each joint to complete the corresponding joint displacement based on the upper limit of the velocity constraint of each joint. Select the maximum value of the shortest motion time corresponding to each joint as the global motion time benchmark of the current execution segment.

[0103] The motion of each joint is time-synchronized and scheduled according to the global motion time reference, and the joint velocity and joint acceleration values ​​of each joint during the motion are recorded.

[0104] The global motion time reference is discretized according to the control cycle duration to generate a discrete time sequence within the current execution segment.

[0105] The time interval between discrete moments is integrated based on the joint velocity value corresponding to each discrete moment to obtain the joint position increment of each joint relative to the current joint position. The joint position increment is then added to the current joint position to obtain the joint position value of each joint at each discrete moment.

[0106] The joint position values, joint velocity values, and joint acceleration values ​​at each discrete moment are organized in chronological order according to the discrete moment sequence to form the segmented motion planning result.

[0107] like Figure 2 As shown, the method includes:

[0108] The segmented motion planning process of the current execution segment extracts the current joint position and current joint velocity of each joint of the robotic arm from the current state information received from the edge computing node. Taking a six-axis robotic arm as an example, the current joint positions extracted from the current state information are [30°, 45°, 60°, -30°, 20°, 15°], and the current joint velocities are [5° / s, 8° / s, -3° / s, 0° / s, 2° / s, -4° / s]. Simultaneously, the target joint positions extracted from the current execution segment are [60°, 90°, 30°, 0°, 45°, 30°].

[0109] Calculate the required joint displacement for each joint. Subtract the current joint position from the target joint position to obtain the displacements for the six joints as [30°, 45°, -30°, 30°, 25°, 15°]. Based on the physical characteristics of the robotic arm, assume the upper limits of velocity constraints for each joint are [60° / s, 50° / s, 55° / s, 60° / s, 50° / s, 65° / s], and the upper limit of acceleration constraints is [120° / s]. 2 100° / s 2 110° / s 2 120° / s 2 100° / s 2 130° / s 2 ].

[0110] A trapezoidal velocity curve model is used to calculate the shortest motion time required for each joint to complete its corresponding displacement. The joint motion is divided into three stages: acceleration, constant velocity, and deceleration. For each joint, the displacement required to accelerate from zero velocity to the upper limit of the velocity constraint is calculated. If this displacement is less than half of the total displacement, the maximum velocity can be reached; otherwise, it can only accelerate to a certain peak velocity and then immediately decelerate.

[0111] Taking the first joint as an example, its displacement is 30°, its maximum velocity is 60° / s, and its maximum acceleration is 120° / s. 2 The time required to accelerate to maximum speed is 60° / s / 120° / s. 2 = 0.5s, and the displacement completed during this process is 0.5 × 60° / s × 0.5s = 15°, which is less than half of the total displacement. Therefore, the joint can reach the maximum speed, and the total motion time is the acceleration time (0.5s) plus the constant speed time ((30°-2×15°) / 60° / s = 0s) plus the deceleration time (0.5s), totaling 1s.

[0112] Through similar calculations, the shortest times required for the six joints to complete their respective displacements were obtained as [1s, 1.34s, 0.9s, 1s, 0.87s, 0.6s]. The maximum value of 1.34s was selected as the global motion time reference for the current execution segment to ensure that all joints can reach the target position synchronously.

[0113] Based on a global motion time reference, time synchronization scheduling of the motion of each joint requires recalculating the actual velocity and acceleration values ​​of each joint to ensure that they complete their respective displacements within the same 1.34s. For example, if the first joint originally needed to complete a 30° displacement in 1 second, it now needs to complete it in 1.34s, therefore its velocity and acceleration values ​​need to be reduced.

[0114] For each joint, the maximum velocity and acceleration after synchronization are recalculated using a symmetrical trapezoidal velocity curve, with acceleration and deceleration times equal. Taking the first joint as an example, substituting the displacement of 30° into the motion equation and combining it with a global motion time of 1.34s, the new maximum velocity is approximately 44.8° / s, and the acceleration is approximately 66.9° / s. 2 Similarly, calculate the synchronization parameters of other joints and record the velocity and acceleration values ​​of each joint at different times.

[0115] The global motion time base of 1.34s is discretized according to the control cycle length. Assuming the control cycle is 0.01s, a discrete time sequence [0, 0.01s, 0.02s, ..., 1.33s, 1.34s] can be generated, with a total of 135 discrete time points.

[0116] For each joint, its velocity value is calculated at each discrete time step. During the acceleration phase, the velocity increases linearly with time; during the constant velocity phase, the velocity remains constant; and during the deceleration phase, the velocity decreases linearly. Then, the position increment between adjacent discrete time steps is calculated using numerical integration. Taking the first joint as an example, at t=0.01s, its velocity is 0.669° / s, and the position increment is 0.00334° through integration; at t=0.02s, the velocity is 1.338° / s, and the position increment is 0.01003°. By accumulating these position increments and superimposing them on the initial position of 30°, the position value of the joint at each discrete time step can be obtained.

[0117] The joint position, velocity, and acceleration values ​​at each discrete moment are organized in chronological order according to the discrete time sequence to form a structured data table, which is the result of the piecewise motion planning. This result will be used for the actual control of the robotic arm, guiding each joint to move smoothly and synchronously along the planned trajectory to reach the target position in the shortest time.

[0118] In practical applications, this motion planning method can adjust parameters according to the physical characteristics and task requirements of different robotic arms, and is applicable to various scenarios such as industrial automation, medical surgery assistance, and logistics handling, effectively improving the time efficiency and path smoothness of robotic arm movements.

[0119] In one optional implementation, based on the segmented motion planning results, calculating the control commands for each joint at the edge computing node includes:

[0120] Based on the changing trends of the position and velocity deviations of each joint at the current discrete moment, it is determined whether each joint is in an oscillation state. The damping adjustment factor is calculated based on the oscillation frequency, and the control response gain of the joint in the oscillation state is reduced using the damping adjustment factor.

[0121] The load current value of each joint is used to determine whether there is a sudden change in load in each joint. For joints with sudden load changes, the load disturbance compensation amount is calculated based on the change in the load current value and the mass distribution parameters of each joint of the robotic arm.

[0122] Based on the damping adjustment factor and the load disturbance compensation amount, the error compensation amount of each joint is calculated, and the joint acceleration value of each joint at the current discrete moment is superimposed with the error compensation amount to obtain the expected joint acceleration of each joint.

[0123] Based on the desired joint acceleration of each joint and the coupling relationship between the joints of the robotic arm, the joint torque value required by each joint to achieve the desired joint acceleration is solved by inverse dynamics calculation; the joint torque value is then converted into the control command for each joint according to the torque constant of each joint actuator.

[0124] By analyzing the changing trends of the position and velocity deviations of each joint, it can be determined whether the joint is in an oscillating state. A position deviation sequence E_p(kn) to E_p(k) and a velocity deviation sequence E_v(kn) to E_v(k) are taken over multiple consecutive sampling periods, where k represents the current sampling time and n is the number of selected historical data points. When the position and velocity deviations exhibit periodic changes over multiple consecutive sampling periods, and the amplitude exceeds a preset threshold, the joint can be determined to be in an oscillating state. The oscillation frequency can be calculated by analyzing the number of zero-point crossings in the position deviation sequence.

[0125] For joints determined to be in an oscillating state, the damping adjustment factor is calculated as follows: determine the ratio of the oscillation frequency to the standard frequency, and then use a nonlinear mapping function to determine the damping adjustment factor based on this ratio. When the oscillation frequency is high, it indicates that the response is too fast, requiring a larger damping adjustment factor to reduce the control response gain; when the oscillation frequency is close to the standard frequency, the damping adjustment factor approaches 1, i.e., maintaining the original control gain.

[0126] By comparing the load current value of the current sampling period with the average value of the previous few sampling periods, a load mutation can be identified in the joint when the current change exceeds a preset threshold. For joints with load mutations, the load disturbance compensation amount needs to be calculated based on the change in load current value and the mass distribution parameters of each joint of the robotic arm. The change in load current value is converted into an equivalent torque change amount through the motor's torque constant. This equivalent torque change amount reflects the direct disturbance torque generated by the external load change on the joint. Based on the mass distribution parameters of each joint of the robotic arm, including the mass value of each joint, the moment of inertia tensor of each joint link, and the center of gravity position vector of each joint link, the mass inertia characteristic parameters of each joint are constructed. The equivalent torque change amount is coupled with the mass inertia characteristic parameters of the joint to obtain the local disturbance compensation torque of the joint. Since there is kinematic and dynamic coupling between the joints of the robotic arm, the load change of one joint will affect other joints through the links. Therefore, it is necessary to calculate the coupled disturbance torque component generated by the load mutation of the joint on other joints based on the Jacobian coupling relationship matrix between the joints. The local disturbance compensation torque is vector-superimposed with the received coupled disturbance torque components from other joints to obtain the global disturbance compensation torque for that joint. Finally, the global disturbance compensation torque is converted into a compensation amount in the acceleration domain based on the torque constants of each joint actuator, which is the load disturbance compensation amount.

[0127] For each joint's position and velocity deviations at the current discrete moment, a basic feedback compensation acceleration is calculated based on the proportional-derivative control principle. This basic feedback compensation acceleration reflects the normal response to position and velocity deviations. For joints determined to be in an oscillating state, the basic feedback compensation acceleration is multiplied by the corresponding damping adjustment factor to obtain the damped feedback compensation acceleration. This damping adjustment suppresses oscillations by reducing the control response gain. For joints not in an oscillating state, the damped feedback compensation acceleration is equal to the basic feedback compensation acceleration. Then, for joints with sudden load changes, the load disturbance compensation is used as a feedforward compensation component and superimposed with the damped feedback compensation acceleration to obtain the overall compensation acceleration of that joint. For joints without sudden load changes, the overall compensation acceleration is equal to the damped feedback compensation acceleration. The overall compensation acceleration of each joint is used as the error compensation for that joint. The error compensation is superimposed with the joint acceleration value to obtain the desired joint acceleration. The desired joint acceleration reflects the response to position and velocity deviations, as well as the compensation for load disturbances.

[0128] In a practical application scenario, when a robotic arm performs a grasping task, if the end effector suddenly comes into contact with an object, it can cause a sudden change in the load current of the relevant joints. By calculating the load disturbance compensation, the robotic arm can smoothly adapt to this load change, avoiding grasping failure or robotic arm oscillation due to a delayed response from the control system.

[0129] Considering the coupling relationships between the joints of the robotic arm, based on the desired joint acceleration and the mass-inertia characteristic parameters of each joint, the inertial torque that each joint needs to overcome to achieve the desired joint acceleration is calculated. The calculation of this inertial torque requires consideration of the robotic arm's mass matrix, which is a function of the joint positions and reflects the inertial coupling relationships during joint movement. Based on the joint velocity and position values ​​of each joint at the current discrete moment, the Coriolis torque and centrifugal torque components between the joints are calculated. The Coriolis torque reflects the coupling effect between the joint velocities, while the centrifugal torque reflects the radial force effect generated by the joint's own velocity. The calculation of these two torques requires the use of the Christopher sign matrix, which is composed of the partial derivatives of the robotic arm's mass matrix with respect to the joint positions. Based on the joint position values ​​of each joint at the current discrete moment and the gravitational acceleration vectors of each link in the robotic arm, the gravitational torque that each joint needs to overcome is calculated. The calculation of the gravitational torque requires consideration of the mass of each link, the position of the center of gravity, and the projection of gravity into the joint space under the joint position configuration. By vector summing the inertial torque, Coriolis torque, centrifugal torque, and gravitational torque, the joint torque values ​​required for each joint to achieve the desired joint acceleration are obtained. The dynamic model of the robotic arm can be represented as the relationship between joint torque and joint position, velocity, and acceleration, including the influence of factors such as joint mass, Coriolis force, centrifugal force, and gravity. Through inverse dynamics calculations, the joint torque required to achieve a given joint acceleration can be calculated.

[0130] The calculated joint torque values ​​are converted into control commands based on the torque constants of each joint actuator. The torque constant is an inherent parameter of the motor, representing the torque generated per unit current. By dividing by the torque constant, the required torque can be converted into the current command needed for motor control. In practical systems, the control commands need to be further converted into voltage or PWM signals to adapt to specific driver hardware interfaces.

[0131] Through the above steps, control commands for each joint can be calculated at the edge computing node, enabling precise control of the robotic arm. This also effectively addresses issues such as oscillation and sudden load changes, improving the motion stability and accuracy of the robotic arm.

[0132] In one optional implementation, when the deviation information exceeds a preset deviation threshold, the deviation propagation impact is calculated at the edge computing node, and the multi-joint collaborative compensation amount is generated based on the deviation propagation impact, including:

[0133] Based on the deviation information, joints whose position deviation value exceeds a preset position deviation threshold or whose velocity deviation value exceeds a preset velocity deviation threshold are determined to be abnormal joints.

[0134] Based on the current configuration of the robotic arm, the Jacobian mapping relationship between each abnormal joint and the end effector is calculated, and the position deviation value and the speed deviation value of each abnormal joint are converted into the deviation transmission influence amount of the end effector through the Jacobian mapping relationship.

[0135] The deviation transmission effects of all abnormal joints are vectorized to obtain the total position offset vector and the total velocity offset vector. The degree and direction of deviation of the end effector are determined based on the magnitude and direction of the total position offset vector and the total velocity offset vector. The compensation priority weight is determined based on the degree and direction of deviation.

[0136] By performing inverse operations on the Jacobian matrix, the total position offset vector and the total velocity offset vector are mapped inversely to the joint space to obtain the joint compensation displacement and joint compensation velocity required by each joint to offset the total position offset vector and the total velocity offset vector of the end effector. The joint compensation displacement and joint compensation velocity are weighted and combined according to the compensation priority weight to form the multi-joint collaborative compensation amount.

[0137] When the detected deviation exceeds the preset deviation threshold, compensation calculation is started at the edge computing node. Specifically, the position deviation value is obtained by collecting the difference between the actual position and the target position of each joint, and the speed deviation value is obtained by collecting the difference between the actual speed and the target speed.

[0138] Based on the obtained deviation information, joints whose position deviation value exceeds a preset position deviation threshold or whose speed deviation value exceeds a preset speed deviation threshold are determined to be abnormal joints. For example, if the position deviation threshold is set to ±0.5 degrees and the speed deviation threshold is set to ±0.3 degrees / second, if a joint has a position deviation of 0.7 degrees or a speed deviation of 0.4 degrees / second, then the joint is determined to be an abnormal joint.

[0139] Based on the current configuration of the robotic arm, the Jacobian mapping relationship between each abnormal joint and the end effector is calculated. The Jacobian matrix represents the velocity mapping relationship between joint space and Cartesian space. This matrix can be used to convert deviations in joint space into offsets in the end effector. Specifically, for a six-axis robotic arm, its Jacobian matrix has a dimension of 6×6 and includes the influence coefficients of each joint rotation axis on the position and attitude of the end effector.

[0140] The position and velocity deviations of each abnormal joint are converted into the deviation transmission impact on the end effector using the Jacobian mapping relationship. Assuming the position deviation of the second joint is 0.7 degrees, the position offset on the end effector caused by this deviation can be calculated as [2.1 mm, 1.5 mm, 0.8 mm] and attitude offset as [0.3 degrees, 0.2 degrees, 0.1 degrees] using the Jacobian matrix. Similarly, the velocity deviation of the fourth joint is 0.4 degrees / second, and the calculated velocity offset on the end effector is [1.2 mm / s, 0.9 mm / s, 0.5 mm / s] and angular velocity offset as [0.2 degrees / s, 0.1 degrees / s, 0.05 degrees / s].

[0141] The deviation transmission effects of all abnormal joints are vectorized to obtain the total position offset vector and the total velocity offset vector. For example, if the second and fourth joints are both abnormal joints, their position effects on the end effector are vectorized to obtain the total position offset vector as [3.0mm, 2.1mm, 1.2mm] and the total attitude offset as [0.5 degrees, 0.3 degrees, 0.15 degrees]. Similarly, the total velocity offset vector is [1.8mm / s, 1.3mm / s, 0.8mm / s] and the total angular velocity offset is [0.3 degrees / s, 0.2 degrees / s, 0.1 degrees / s].

[0142] The degree and direction of deviation of the end effector are determined by the magnitude and direction of the total position offset vector and the total velocity offset vector. The magnitude of the position offset vector is calculated to be 3.8 mm, and the magnitude of the velocity offset vector is 2.3 mm / s. The offset direction is mainly concentrated in the xy plane.

[0143] The compensation priority weight is determined based on the degree and direction of deviation. When the position deviation is large, the position compensation weight is higher, for example, set to 0.7; when the velocity deviation is large, the velocity compensation weight is higher, for example, set to 0.6. In the deviation direction, different weights can be set for different axes, for example, the weight in the x-direction is 0.8, the weight in the y-direction is 0.7, and the weight in the z-direction is 0.5.

[0144] By inversely operating the Jacobian matrix, the total position offset vector and the total velocity offset vector are mapped back to the joint space. For position offset, the required compensation displacement value for each joint is calculated, such as -0.3 degrees for the first joint, -0.5 degrees for the second joint, and so on. For velocity offset, the required compensation velocity value for each joint is calculated, such as -0.2 degrees / second for the first joint, -0.3 degrees / second for the second joint, and so on.

[0145] The joint compensation displacement and joint compensation velocity are weighted and combined according to the compensation priority to form the multi-joint collaborative compensation amount. For example, for the first joint, the final compensation amount is position compensation weight × compensation displacement + velocity compensation weight × compensation velocity × time step, i.e., 0.7 × (-0.3 degrees) + 0.6 × (-0.2 degrees / second) × 0.05 seconds = -0.216 degrees. For the second joint, the final compensation amount is calculated in the same way, which is -0.359 degrees. Other joints are calculated in the same way.

[0146] This multi-joint collaborative compensation method fully considers the comprehensive impact of joint deviations on the end effector. By performing spatial mapping and inverse mapping through the Jacobian matrix, it achieves precise compensation results. Especially in high-precision application scenarios, such as precision assembly and minimally invasive surgery assistance, it can effectively improve the positioning accuracy and trajectory smoothness of the robotic arm end effector.

[0147] In one optional implementation, based on the deviation information and the multi-joint collaborative compensation amount, the prediction buffer segment is corrected at the edge computing node, and the corrected prediction buffer segment is used to control the robotic arm to perform motion correction in real time, including:

[0148] Extract the predicted joint position and predicted joint velocity of each joint at multiple discrete moments in the future from the prediction cache segment;

[0149] The deviation spatial gradient of each joint is calculated based on the deviation information. The cross-coupling deviation component of each joint's deviation on other joints at multiple discrete moments in the future is calculated based on the deviation spatial gradient. The position deviation value and velocity deviation value of each joint are superimposed with the cross-coupling deviation component to obtain the global coupling deviation amount of each joint.

[0150] The compensation bearing capacity coefficient of each joint is calculated based on the predicted joint position and the predicted joint velocity. The multi-joint collaborative compensation amount is distributed according to the compensation bearing capacity coefficient of each joint to obtain the distributed compensation component of each joint at multiple discrete moments in the future.

[0151] Calculate the compensation phase consistency factor based on the phase relationship between the distributed compensation component and the global coupling deviation, and use the compensation phase consistency factor to perform phase calibration on the distributed compensation component.

[0152] Using the global coupling deviation and the phase-calibrated distributed compensation component, motion correction is performed on the predicted joint position and the predicted joint velocity, and motion correction commands for each joint are generated. The robotic arm is then controlled to perform motion correction in real time according to the motion correction commands.

[0153] The predicted joint positions and velocities for each joint at multiple discrete future time points are extracted from the prediction buffer. The prediction buffer typically contains motion prediction data for each joint of the robotic arm over a future period, calculated based on the robotic arm's kinematic and dynamic models. By accessing the data structures in the prediction buffer, the position and velocity values ​​of each joint at future time points t+1, t+2, ..., t+n (where n is the prediction duration) can be obtained. These values ​​form the basis for subsequent corrections.

[0154] After acquiring the predicted data, the spatial gradient of the deviation for each joint is calculated based on the deviation information, which typically originates from the difference between the actual position reported by the sensor and the expected position. For joint i, the spatial gradient of the deviation can be expressed as the rate of change of the deviation in the position and velocity dimensions. Specifically, this can be calculated by dividing the difference between the deviation values ​​at the current time t and the previous time t-1 by the time interval. This gradient information reflects the trend of deviation change and provides a basis for subsequent cross-coupling analysis.

[0155] Based on the calculated deviation spatial gradient, the cross-coupling deviation components of each joint's deviation on other joints at multiple discrete moments in the future are further calculated. In a multi-joint robotic arm system, the motion deviation of one joint is transmitted to other joints through the mechanical structure, forming a cross-coupling effect. Assuming the robotic arm has m joints, for joint i, the cross-coupling deviations from the other m-1 joints need to be calculated. These cross-coupling deviations can be described by the Jacobian matrix to describe the mutual influence between joints. The position deviation and velocity deviation values ​​of each joint are superimposed with the calculated cross-coupling deviation components to obtain the global coupling deviation of each joint. The global coupling deviation comprehensively reflects the overall deviation of the joint's own deviation and the influence from other joints.

[0156] After determining the deviation of each joint, the compensation load-bearing capacity coefficient of each joint needs to be calculated based on the predicted joint position and predicted joint speed. The compensation load-bearing capacity coefficient represents the extent of compensation that a joint can bear under different conditions. For example, when a joint is close to its speed or position limit, its compensation capacity decreases; conversely, when the joint is in the middle of its range of motion and the speed is moderate, its compensation capacity is stronger. The actual load torque of each joint is obtained by reading the load current value of the driver. The load occupancy rate is calculated by comparing the actual load torque with the rated maximum torque of the joint. A nonlinear mapping function is used to convert the load occupancy rate into a load-bearing capacity component. When the load occupancy rate is low, this component is close to 1; when the load occupancy rate is high, this component decays rapidly. The minimum distance from each joint position to the maximum and minimum position limits is calculated and compared with the total range of motion of the joint to obtain the position margin ratio, which is also converted into a position load-bearing capacity component through nonlinear mapping. The speed occupancy rate is obtained by comparing the absolute value of the current speed of each joint with the maximum speed limit, and is converted into a speed load-bearing capacity component. The load-bearing capacity component, position-bearing capacity component, and velocity-bearing capacity component are weighted and multiplied to obtain the comprehensive compensation bearing capacity coefficient of each joint. The coefficient value is between 0 and 1.

[0157] The multi-joint collaborative compensation amount is allocated according to the compensation bearing capacity coefficient of each joint, resulting in the distributed compensation components of each joint at multiple discrete moments in the future. The multi-joint collaborative compensation amount is the amount that the entire system needs to compensate for. Allocation through the bearing capacity coefficient ensures that the compensation task is reasonably distributed to each joint, avoiding overload or exceeding limits caused by some joints bearing too much compensation. The allocation process can adopt a weighted allocation method, making the compensation amount proportional to the bearing capacity.

[0158] To ensure the effectiveness of compensation, the global coupling deviation of each joint is decomposed into position deviation vector components and velocity deviation vector components. Similarly, the distributed compensation components of each joint are decomposed into position compensation vector components and velocity compensation vector components. The vector angle between the position and velocity domains is calculated separately. The cosine similarity in the position domain is obtained by calculating the inner product of the position deviation vector component and the position compensation vector component and dividing by the product of the magnitudes of the two vectors. The cosine similarity in the velocity domain is calculated in the same way. When the cosine similarity is close to 1, it indicates that the compensation direction is consistent with the deviation direction; close to -1, it indicates that the direction is opposite; and close to 0, it indicates that they are orthogonal. The weighted average of the cosine similarities in the position and velocity domains is used to obtain a comprehensive phase consistency metric. Based on this metric, a compensation phase consistency factor is calculated using a nonlinear mapping function. When the phase consistency metric is positive and large, the factor close to 1 indicates that no calibration is needed. When the phase consistency metric is small or negative, the factor is reduced to suppress inconsistent compensation components. The compensation effect is best when the compensation direction is consistent with the deviation direction; conversely, if the direction is opposite, the compensation effect is poor or even the deviation is aggravated. By calculating the compensation phase consistency factor, this degree of consistency can be quantified, and the compensation component can be phase-calibrated accordingly to ensure that the compensation is consistent with the direction of the deviation.

[0159] The predicted joint position and velocity are corrected using a globally coupled bias and a phase-calibrated distributed compensation component. The correction process employs a weighted fusion method, fusing the original predicted value, bias, and compensation component in a specific ratio to obtain the corrected predicted value. This corrected data is then used to generate motion correction commands for each joint. These commands contain corrected position, velocity, and other control parameters, and are sent to the robotic arm's control system via a communication interface for real-time motion correction.

[0160] In practical applications, taking a 6-axis industrial robotic arm as an example, when performing a complex trajectory task, a significant positional deviation occurs in the third joint due to load variations. After detecting this deviation, the edge computing node calculates a large and increasing spatial gradient of the deviation for this joint. Further analysis shows that this deviation has a significant cross-coupling effect on the fourth and fifth joints through the mechanical structure. After evaluating the compensation load-bearing capacity of each joint, it was found that the second and sixth joints currently have a lighter load and higher compensation load-bearing capacity. Therefore, part of the compensation task is assigned to these two joints, while ensuring that the compensation direction is consistent with the deviation direction. Ultimately, the motion trajectory is successfully corrected through multi-joint collaborative compensation, ensuring the precise execution of the robotic arm's task.

[0161] This invention relates to a real-time control system for a robotic arm based on edge computing, the system comprising:

[0162] The first unit is used to acquire the current status information of the robotic arm and the target task instructions;

[0163] The second unit is used to perform semantic parsing of the target task instructions at the edge computing node, and convert the target task instructions into a joint space trajectory sequence based on the mapping relationship between joint motion range constraints and robotic arm workspace constraints.

[0164] The third unit is used to decompose the joint space trajectory sequence in the time domain according to the availability of computing resources and control cycle constraints, and generate the current execution segment and the prediction cache segment.

[0165] The fourth unit is used to generate segmented motion planning results at the edge computing node based on the current execution segment and the current state information and the time-optimal criterion; and to calculate control commands for each joint at the edge computing node based on the segmented motion planning results.

[0166] The fifth unit is used to acquire the actual state information of the robotic arm after executing the control command, and to calculate the deviation information between the actual state information and the joint target position of the corresponding time segment in the segmented motion planning result;

[0167] The sixth unit is used to calculate the deviation transmission impact at the edge computing node when the deviation information exceeds a preset deviation threshold, generate a multi-joint collaborative compensation amount based on the deviation transmission impact amount, and correct the prediction buffer segment at the edge computing node based on the deviation information and the multi-joint collaborative compensation amount, and use the corrected prediction buffer segment to control the robotic arm to perform motion correction in real time.

[0168] A third aspect of the present invention provides an electronic device, comprising:

[0169] processor;

[0170] Memory used to store processor-executable instructions;

[0171] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0172] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0173] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0174] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A real-time control method for a robotic arm based on edge computing, characterized in that, include: Obtain the current status information and target task instructions of the robotic arm; The target task instructions are semantically parsed at the edge computing node, and based on the mapping relationship between the joint motion range constraints and the robot arm workspace constraints, the target task instructions are converted into a joint space trajectory sequence. The joint space trajectory sequence is decomposed in the time domain according to the availability of computing resources and control cycle constraints to generate the current execution segment and the prediction cache segment. Based on the current execution segment and the current state information, a segmented motion planning result is generated at the edge computing node according to the time-optimal criterion; based on the segmented motion planning result, control commands for each joint are calculated at the edge computing node; Obtain the actual state information of the robotic arm after executing the control command, and calculate the deviation information between the actual state information and the joint target position of the corresponding time segment in the segmented motion planning result; When the deviation information exceeds the preset deviation threshold, the deviation transmission impact is calculated at the edge computing node, and a multi-joint collaborative compensation amount is generated based on the deviation transmission impact. Based on the deviation information and the multi-joint collaborative compensation amount, the prediction cache segment is corrected at the edge computing node, and the corrected prediction cache segment is used to control the robotic arm to perform motion correction in real time.

2. The method according to claim 1, characterized in that, At the edge computing node, semantic parsing of the target task instructions is performed. Based on the mapping relationship between joint motion range constraints and robotic arm workspace constraints, the target task instructions are converted into a joint space trajectory sequence, including: In the edge computing node, the target position and target posture of the robotic arm in the Cartesian coordinate system are extracted from the target task instruction through syntax structure decomposition; By calculating the reachable space boundary of the robotic arm end effector when each joint of the robotic arm takes the value of the joint motion range constraint boundary, a two-way mapping relationship is formed from the joint angle limit value to the reachable range of the Cartesian space. Determine whether the target position and the target posture are within the reachable space boundary of the robotic arm end effector. If so, use the bidirectional mapping relationship to transform the robotic arm workspace constraint into boundary limiting conditions for joint angle configuration, and generate each candidate joint angle configuration. The robot arm end position corresponding to each candidate joint angle configuration is obtained by forward kinematics calculation. The spatial deviation between the robot arm end position and the target position is calculated. The candidate joint angle configuration with the smallest spatial deviation is selected as the effective joint angle configuration. Based on the effective joint angle configuration and the joint position information in the current state information, an interpolation path connecting the current joint position and the effective joint angle configuration is generated based on the joint motion continuity constraint, and the interpolation path is arranged in a time sequence to form the joint spatial trajectory sequence.

3. The method according to claim 1, characterized in that, The joint space trajectory sequence is decomposed in the time domain according to the constraints of computing resource availability and control cycle to generate the current execution segment and the prediction cache segment, including: The processor utilization and available memory capacity of the edge computing node are obtained from the quantitative indicators of computing resource availability; the control cycle duration of the robotic arm is obtained as the time base for the control cycle constraint. Calculate the joint angle change gradient of each trajectory point in the joint space trajectory sequence. If the joint angle change gradient exceeds a preset gradient threshold, it is marked as a high dynamic trajectory interval; otherwise, it is marked as a low dynamic trajectory interval. The reduction coefficient is determined based on the ratio of the maximum value of the joint angle change gradient to the preset gradient threshold, and the expansion coefficient is determined based on the ratio of the minimum value of the joint angle change gradient to the preset gradient threshold. The number of reference trajectory points that the edge computing node can process within a single control cycle is calculated based on the processor utilization rate. The number of trajectory points in the high dynamic trajectory interval is determined according to the number of reference trajectory points and the reduction coefficient. The number of trajectory points in the low dynamic trajectory interval is determined according to the number of reference trajectory points and the expansion coefficient. When the available memory capacity of the edge computing node is sufficient to store the current execution segment and the prediction cache segment, starting from the start time point of the joint space trajectory sequence, the joint space trajectory sequence is non-uniformly segmented according to the number of trajectory points in the high dynamic trajectory interval and the low dynamic trajectory interval. The trajectory content of the non-uniform segmentation covers at least one complete control cycle duration corresponding to the time base. The segmented trajectory content is used as the current execution segment, and the remaining trajectory content is used as the prediction cache segment.

4. The method according to claim 1, characterized in that, Based on the current execution segment and the current state information, the segmented motion planning results generated at the edge computing node according to the time-optimal criterion include: Extract the current joint position and current joint speed of each joint of the robotic arm from the current state information, and extract the position of each target joint from the current execution segment; Calculate the joint displacement required for each joint to move from the current joint position to the target joint position. Calculate the shortest motion time required for each joint to complete the corresponding joint displacement based on the upper limit of the velocity constraint of each joint. Select the maximum value of the shortest motion time corresponding to each joint as the global motion time benchmark of the current execution segment. The motion of each joint is time-synchronized and scheduled according to the global motion time reference, and the joint velocity and joint acceleration values ​​of each joint during the motion are recorded. The global motion time reference is discretized according to the control cycle duration to generate a discrete time sequence within the current execution segment. The time interval between discrete moments is integrated based on the joint velocity value corresponding to each discrete moment to obtain the joint position increment of each joint relative to the current joint position. The joint position increment is then added to the current joint position to obtain the joint position value of each joint at each discrete moment. The joint position values, joint velocity values, and joint acceleration values ​​at each discrete moment are organized in chronological order according to the discrete moment sequence to form the segmented motion planning result.

5. The method according to claim 4, characterized in that, Based on the segmented motion planning results, the control commands for each joint are calculated at the edge computing node, including: Based on the changing trends of the position and velocity deviations of each joint at the current discrete moment, it is determined whether each joint is in an oscillation state. The damping adjustment factor is calculated based on the oscillation frequency, and the control response gain of the joint in the oscillation state is reduced using the damping adjustment factor. The load current value of each joint is used to determine whether there is a sudden change in load in each joint. For joints with sudden load changes, the load disturbance compensation amount is calculated based on the change in the load current value and the mass distribution parameters of each joint of the robotic arm. Based on the damping adjustment factor and the load disturbance compensation amount, the error compensation amount of each joint is calculated, and the joint acceleration value of each joint at the current discrete moment is superimposed with the error compensation amount to obtain the expected joint acceleration of each joint. Based on the desired joint acceleration of each joint and the coupling relationship between the joints of the robotic arm, the joint torque value required by each joint to achieve the desired joint acceleration is solved by inverse dynamics calculation; the joint torque value is then converted into the control command for each joint according to the torque constant of each joint actuator.

6. The method according to claim 1, characterized in that, When the deviation information exceeds a preset deviation threshold, the deviation propagation impact is calculated at the edge computing node, and a multi-joint collaborative compensation amount is generated based on the deviation propagation impact, including: Based on the deviation information, joints whose position deviation value exceeds a preset position deviation threshold or whose velocity deviation value exceeds a preset velocity deviation threshold are determined to be abnormal joints. Based on the current configuration of the robotic arm, the Jacobian mapping relationship between each abnormal joint and the end effector is calculated, and the position deviation value and the speed deviation value of each abnormal joint are converted into the deviation transmission influence amount of the end effector through the Jacobian mapping relationship. The deviation transmission effects of all abnormal joints are vectorized to obtain the total position offset vector and the total velocity offset vector. The degree and direction of deviation of the end effector are determined based on the magnitude and direction of the total position offset vector and the total velocity offset vector. The compensation priority weight is determined based on the degree and direction of deviation. By performing inverse operations on the Jacobian matrix, the total position offset vector and the total velocity offset vector are mapped inversely to the joint space to obtain the joint compensation displacement and joint compensation velocity required by each joint to offset the total position offset vector and the total velocity offset vector of the end effector. The joint compensation displacement and joint compensation velocity are weighted and combined according to the compensation priority weight to form the multi-joint collaborative compensation amount.

7. The method according to claim 1, characterized in that, Based on the deviation information and the multi-joint collaborative compensation amount, the prediction buffer segment is corrected at the edge computing node, and the corrected prediction buffer segment is used to control the robotic arm to perform motion correction in real time, including: Extract the predicted joint position and predicted joint velocity of each joint at multiple discrete moments in the future from the prediction cache segment; The deviation spatial gradient of each joint is calculated based on the deviation information. The cross-coupling deviation component of each joint's deviation on other joints at multiple discrete moments in the future is calculated based on the deviation spatial gradient. The position deviation value and velocity deviation value of each joint are superimposed with the cross-coupling deviation component to obtain the global coupling deviation amount of each joint. The compensation bearing capacity coefficient of each joint is calculated based on the predicted joint position and the predicted joint velocity. The multi-joint collaborative compensation amount is distributed according to the compensation bearing capacity coefficient of each joint to obtain the distributed compensation component of each joint at multiple discrete moments in the future. Calculate the compensation phase consistency factor based on the phase relationship between the distributed compensation component and the global coupling deviation, and use the compensation phase consistency factor to perform phase calibration on the distributed compensation component. Using the global coupling deviation and the phase-calibrated distributed compensation component, motion correction is performed on the predicted joint position and the predicted joint velocity, and motion correction commands for each joint are generated. The robotic arm is then controlled to perform motion correction in real time according to the motion correction commands.

8. A real-time control system for a robotic arm based on edge computing, used to implement the method as described in any one of claims 1-7, characterized in that, include: The first unit is used to acquire the current status information of the robotic arm and the target task instructions; The second unit is used to perform semantic parsing of the target task instructions at the edge computing node, and convert the target task instructions into a joint space trajectory sequence based on the mapping relationship between joint motion range constraints and robotic arm workspace constraints. The third unit is used to decompose the joint space trajectory sequence in the time domain according to the availability of computing resources and control cycle constraints, and generate the current execution segment and the prediction cache segment. The fourth unit is used to generate segmented motion planning results at the edge computing node based on the current execution segment and the current state information and the time-optimal criterion; and to calculate control commands for each joint at the edge computing node based on the segmented motion planning results. The fifth unit is used to acquire the actual state information of the robotic arm after executing the control command, and to calculate the deviation information between the actual state information and the joint target position of the corresponding time segment in the segmented motion planning result; The sixth unit is used to calculate the deviation transmission impact at the edge computing node when the deviation information exceeds a preset deviation threshold, and generate a multi-joint collaborative compensation amount based on the deviation transmission impact. Based on the deviation information and the multi-joint collaborative compensation amount, the prediction cache segment is corrected at the edge computing node, and the corrected prediction cache segment is used to control the robotic arm to perform motion correction in real time.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images