A redundancy manipulator acceleration layer hierarchical control method considering priority switching

Abstract

The application provides a redundancy manipulator acceleration layer hierarchical control method considering priority switching and joint constraints, and belongs to the technical field of robot control. The method comprises the following steps: S1: establishing an acceleration layer kinematics model, and obtaining a Jacobian matrix of a task; S2: constructing a hierarchical null space projection matrix based on a task priority, so as to realize spatial decoupling of the task; S3: constructing a multi-level physical constraint model containing a state constraint operator, so as to realize constraints on joint speed and acceleration; S4: constructing an error-driven priority switching hierarchical recurrent dynamic controller, and eliminating signal mutation during priority switching by using a state maintaining and compensating mechanism; S5: inputting parameters to the controller for iterative solution; and S6: controlling the redundancy manipulator according to the solution result obtained in step S5 to execute multiple tasks. The control method is suitable for a redundancy manipulator system which executes multiple tasks and needs to adjust the task priority in real time.

Description

Technical Field

[0001] This invention relates to the field of robot control technology, specifically to a hierarchical control method for the acceleration layer of a redundant robotic arm that considers priority switching. It is particularly suitable for redundant robotic arm systems that need to perform multiple tasks in complex dynamic environments and whose task priorities need to be adjusted in real time. Background Technology

[0002] A redundant robotic arm is a robotic arm with more degrees of freedom than the minimum number of degrees of freedom required to perform a specific task. Leveraging its kinematic redundancy, a redundant robotic arm can not only perform primary tasks such as end-effector trajectory tracking, but also utilize the remaining degrees of freedom to simultaneously perform secondary tasks such as obstacle avoidance, singularity avoidance, and joint control. To coordinate conflicts between multiple tasks, hierarchical control frameworks are widely used, with zero-space projection-based methods being the mainstream approach.

[0003] However, existing hierarchical control methods for redundant robotic arms have the following significant drawbacks in practical applications: First, traditional null-space projection methods typically require the total dimension of the task to be less than the degrees of freedom of the robotic arm. When the total dimension of the task exceeds the degrees of freedom (i.e., in "over-constraint" or "task overload" scenarios), the calculation of the projection matrix becomes numerically unstable or even fails due to the involvement of inverting singular matrices, making it unable to effectively handle task sets of arbitrary dimensions. Second, during multi-task execution, task priorities often need to be dynamically adjusted according to environmental changes (such as sudden obstacle avoidance requirements). Existing methods lack effective transition mechanisms; when task priorities switch, the projection matrix undergoes abrupt changes, causing drastic discontinuous jumps in control signals (such as joint velocity or acceleration). These jumps can cause severe shaking of the robotic arm, resulting in mechanical damage or even instability of the control system. Third, existing methods struggle to simultaneously accommodate multi-level physical constraints. The robotic arm is limited by joint velocity and acceleration during operation. Therefore, a redundant robotic arm control method that can handle task overload, ensure smooth priority switching, and simultaneously satisfy multi-level physical constraints is urgently needed. Summary of the Invention

[0004] In view of this, the present invention proposes a layered control method for the acceleration layer of a redundant robotic arm that considers priority switching, aiming to solve the problems of numerical instability under task overload, sudden changes in control signals during priority switching, and lack of multi-level physical constraint protection in the existing technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A hierarchical control method for the acceleration layer of a redundant robotic arm considering priority switching includes the following steps:

[0007] S1: Obtain the task set of the redundant robotic arm and set the initial priority sequence; establish the kinematic model of the redundant robotic arm in the acceleration layer, and calculate the Jacobian matrix and task space input vector corresponding to each task;

[0008] S2: For scenarios where the task dimension may exceed the degrees of freedom of the robotic arm, a hierarchical null projection matrix is ​​constructed based on the priority sequence; the hierarchical null projection matrix realizes task space decoupling and strictly guarantees task priority;

[0009] S3: Construct a multi-level physical constraint model and define state constraint operators. It is used to handle the joint velocity constraints and joint acceleration constraints of the robotic arm;

[0010] S4: Construct an error-driven priority switching hierarchical recurrent neurodynamic controller;

[0011] S5: Input the parameters calculated in steps S1 to S4 into the priority switching hierarchical recurrent neurodynamic controller for iterative solution to obtain a smooth and continuous joint acceleration control signal that satisfies all physical constraints.

[0012] S6: Control the redundant robotic arm to perform multiple tasks based on the solution obtained in step S5.

[0013] Furthermore, in step S2, regarding the first... Level 1 priority task ( Construct the null space projection matrix The specific calculation formula is as follows: ,in, It is the identity matrix. For the first The residual null space projection matrix obtained from the task calculation. For the first The effective subspace component occupied by the task level;

[0014] The effective subspace components The calculation is based on The numerical state adopts a segmented strategy:

[0015] (1) Normal mode: when When the matrix is ​​non-singular, the subspace components are calculated using pseudo-inverse operations: .

[0016] (2) Singular processing mode: when When the matrix is ​​singular or zero, the transpose operation is used to approximate the pseudo-inverse operation for calculating the subspace components: .

[0017] in, For the first Jacobian matrix for level-one tasks, This indicates the Moore Penrose pseudo-inverse operation. This represents the matrix transpose operation.

[0018] Furthermore, the constraint operator in step S3 is defined as follows: The state constraint operator The amplitude limiting function is used to limit the motion variables in joint space; it is defined as a piecewise function with saturation characteristics. For the... Input variables of each joint , The expression is:

[0019]

[0020] in, These are the upper and lower bounds of the joint's dynamic range, respectively. These two boundary values ​​are jointly determined by the joint's velocity and acceleration constraints, and the specific calculation formula is as follows: In the formula, For the current moment Joint velocity of each joint; The first The physical lower and upper limits of the joint velocity of each joint; The first The physical lower and upper limits of joint acceleration for each joint; The scaling factor is used to convert the velocity margin into an equivalent acceleration constraint. and These represent the functions for finding the maximum and minimum values, respectively.

[0021] Furthermore, in step S4, the priority switching hierarchical recurrent neurodynamic controller is updated using the following state evolution equation:

[0022] ,

[0023] in, Given an inertial time constant, the state evolution equation is configured to solve a generalized multi-task programming problem, defined as: given that... While constraining, minimize the joint acceleration energy objective function. ; For dynamic compensation of bias; reference control quantity Defined as ,in, This is a horizontally concatenated matrix of null projection matrices for each task level; For each level of task-driven terms, there is a vertically stacked vector; where, Sort the index according to the current priority. These are the Lagrange multiplier parameters for the corresponding task.

[0024] Furthermore, the dynamic compensation bias amount Updates follow the following state preservation logic: During initialization, let... At the instant a change in the task priority sequence is detected. (For example (seconds), execute the following strategy:

[0025] (1) State Locking: Lock the control output at the end of the previous stage. And lock the Lagrange multipliers corresponding to each task at that moment. ;

[0026] (2) Constructing a virtual reference quantity: at the time after the switch Using the current robotic arm configuration (i.e., the current Jacobian matrix) and projection matrix ), in conjunction with locked old Lagrange multipliers Calculate a virtual control quantity It is defined as the null projection matrix updated at the current time. and the task-driven items updated at the current moment With the Lagrange multipliers locked in step (1) The control quantity obtained by matrix multiplication;

[0027] (3) Real-time dynamic compensation: Calculate the dynamic compensation amount ; By using this time-varying Inject controller, so that input items Mathematically, this is equivalent to a smooth transition at the moment of switching, and correction of deviations caused by changes in the projected space as the robotic arm moves.

[0028] Furthermore, the acceleration layer kinematic relationship model in step S1 is defined as follows: , among which, the Task space input vector for each task It includes position error feedback items and velocity error feedback items.

[0029] Further, step S6 specifically includes: iteratively solving the output joint acceleration using the priority-switching hierarchical recurrent neurodynamic controller from step S5. Then, a smooth and continuous joint acceleration control signal that satisfies all physical constraints is obtained, thereby controlling the redundant robotic arm to achieve multi-task operation.

[0030] The beneficial effects of this invention are as follows: By constructing a null space projection matrix, this invention effectively solves the numerical stability problem under task overload scenarios. Combined with a recurrent neurodynamic controller with a state-preservation compensation mechanism, it eliminates potential control signal abrupt changes and mechanical shocks during dynamic task priority switching, ensuring smooth motion. Simultaneously, by utilizing multi-level physical constraint operators to strictly limit joint velocity and acceleration within a unified framework, it comprehensively improves the operational safety and stability of the robotic arm system in complex multi-task environments. Attached Figure Description

[0031] To further illustrate the objectives and technical solutions of this invention, the following figures are provided for illustrative purposes:

[0032] Figure 1 This is a flowchart of a redundancy robotic arm acceleration layer hierarchical control method considering priority switching according to the present invention;

[0033] Figure 2 This is a diagram showing the spatial motion trajectory and actual tracking effect of the redundant robotic arm when performing end-effector trajectory tracking in an embodiment of the present invention.

[0034] Figure 3 This is a tracking error curve diagram of the main tasks (task 1: end position control; task 2: end attitude maintenance) before and after priority switching in the embodiments of the present invention;

[0035] Figure 4 This is a tracking error curve of the secondary task (task 3: elbow joint trajectory tracking) in the insertion and removal process in an embodiment of the present invention.

[0036] Figure 5 This is a tracking error curve for the optimization tasks (task 4: acceleration optimization of the first joint; task 5: acceleration optimization of the seventh joint) in the embodiments of the present invention;

[0037] Figure 6 The graph shows the joint acceleration variation of the redundant robotic arm in this embodiment of the invention, verifying the smoothness of the control signal and its satisfaction with physical constraints. Detailed Implementation

[0038] To make the objectives and technical solutions of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0039] Example: Assume a redundant robotic arm with 7 degrees of freedom, whose joint variables are denoted as... The robotic arm is configured to perform the following 5 tasks, and the total dimensions of the tasks are... This exceeds the degrees of freedom of the robotic arm. This is a typical task overload scenario.

[0040] The task is defined as follows: Task 1: End effector trajectory tracking (dimensions) Task 2: End-effector attitude maintenance (dimensions) Task 3: Elbow joint (5th joint) trajectory tracking (dimensions) Task 4: Optimization of acceleration at joint 1 (dimensions) Task 5: Optimization of acceleration at the 7th joint (dimensions) ).

[0041] In this embodiment, the simulation time is set to 30 seconds, and the priority sequence is defined as follows: the earlier a task appears in the sequence, the higher its priority. For example, the sequence... Two dynamic priority switches occur during operation: 0-10 seconds: priority sequence is as follows (Task 3 not activated); 10-20 seconds: Priority sequence switches to (Task 3 is inserted and its priority is increased, while Task 1's priority is decreased); 20-30 seconds: The priority sequence switches to... (Task 3 removed, Task 1 priority restored).

[0042] Combination Figure 1 The control method of the present invention includes the following steps:

[0043] Step S1: Establish the kinematic model of the acceleration layer: Obtain the Jacobian matrix for each task. And calculate the task space input vector based on the desired trajectory. The specific form is: In the formula, , For actual task coordinates and velocity, , represents the desired task trajectory parameters. The current joint velocity, , This is the feedback gain coefficient, where the feedback gain is set to... .

[0044] Step S2: Construct the hierarchical null space projection matrix: Recursively calculate the projection matrix based on the current priority sequence. For example, in the 10-20 second phase, the priority order is... First, calculate the projection matrix of the highest priority task 2: Next, calculate the projection matrix for secondary task 3: If during this process, it is discovered If the rank is insufficient (close to zero), then the transpose operation is used. Instead of pseudo-inverse operations, this prevents computational divergence. In this example, the total task dimension is 10, exceeding the 7 degrees of freedom. The projection matrix of low-priority tasks (such as the adjusted task 1) degenerates to zero. In this case, this strategy ensures that the algorithm does not crash and remains stable.

[0045] Step S3: Construct the physical constraint model: Set joint velocity limits and acceleration limits. Construct state constraint operators. For example, state constraint operators By using a saturation function, the joint acceleration is strictly limited to Within a certain range, the joint speed is strictly limited to Within the range.

[0046] Step S4: Solving the priority-switching hierarchical recurrent neurodynamic controller: Real-time iteration is performed using an error-driven priority-switching hierarchical recurrent neurodynamic controller. Set the inertial time constant .

[0047] Step S5: In At the second, the priority is from Mutation At this point, due to the fundamental change in the null space structure of the projection matrix, without compensation, the reference control quantity... This will cause a significant jump. The controller's internal compensation logic takes effect immediately: locking the system's total output at the instant preceding the last 10 seconds. And the Lagrange multiplier parameters for each task; during the run 10 seconds later, the virtual reference quantity is calculated in real time. This quantity is calculated using the current projection matrix structure and the locked old multiplier parameters; the dynamic compensation quantity is calculated. The amount of compensation This precisely cancels out the algebraic jump caused by the abrupt change in the projection matrix structure, making the input to the constraint operator... Instructions in Maintain numerical continuity. Similarly, in The same compensation operation will also be performed at the second mark. The compensation increment at each second is added to the total compensation to ensure smooth and stable operation of the system during multiple switching operations.

[0048] Step S6: Execution control: Use the joint acceleration signals obtained from the solution to drive the redundant robotic arm and perform multi-task operations.

[0049] This embodiment utilizes MATLAB software for simulation experiments. To verify the effectiveness of the control method proposed in this invention, simulation experiments were conducted for the aforementioned five tasks. The experimental results are as follows: Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 As shown.

[0050] further, Figure 2 The comparison between the actual motion trajectory and the expected trajectory of the redundant robotic arm's end effector is shown. It can be seen that even when the task priority dynamically switches between the 10th and 20th seconds, the robotic arm's end effector can still smoothly and accurately track the expected trajectory without significant deviation, verifying the algorithm's control capability in Cartesian space.

[0051] Figure 3 The tracking error curves for the main tasks (Task 1 and Task 2) are shown. For Task 1 (end-effector position control), it has the highest priority from 0 to 10 seconds, with minimal error. From 10 to 20 seconds, due to its reduced priority and insufficient system degrees of freedom, the error increases as expected, but remains within a controllable range under the controller's adjustment. After 20 seconds, as the priority recovers, the error rapidly converges to zero. For Task 2 (end-effector attitude maintenance), it maintains a high priority throughout the process, thus its tracking error remains close to zero and is not significantly affected by the switching of other tasks.

[0052] Figure 4 The figure illustrates the error variation of the secondary task (Task 3: Elbow Joint Trajectory Tracking). The graph clearly reflects the dynamic adjustment process of task priority: during the periods of 0 to 10 seconds (task inactive) and 20 to 30 seconds (task removed), the error is large and fluctuates because the task is not executed; however, during the period of 10 to 20 seconds (task inserted and executed), the error converges rapidly and remains stable. This demonstrates that the method of this invention can flexibly handle dynamic changes in task priority.

[0053] Figure 5 The execution errors of the joint-level optimization tasks (tasks 4 and 5) are illustrated. Task 4 requires the first joint to track a sinusoidally changing acceleration signal, and task 5 requires the seventh joint to maintain zero acceleration. As can be seen from the figure, despite these two tasks being of lower priority and facing the challenge of limited system degrees of freedom, the controller proposed in this invention can still control their tracking errors to an extremely small order of magnitude (e.g., (within radians per square second), enabling refined management of the movement of the underlying joints.

[0054] Figure 6 The graph shows the acceleration curves of each joint of the redundant robotic arm as a function of time. As can be seen from the graph, throughout the entire 30-second motion, including the two priority switching moments, the acceleration of all joints remained strictly within the preset physical limits. The curve is 2 radians per second squared, and the curve is smooth and continuous without any peaks or jitters. This fully demonstrates that the error-driven priority switching hierarchical recurrent neurodynamic controller proposed in this invention effectively achieves smooth control of the robotic arm while ensuring numerical stability, thus avoiding impact on the mechanical body.

[0055] In summary, the method of the present invention has significant advantages in handling task overload, multi-level physical constraints, and dynamic priority switching, and can simultaneously ensure the accurate execution and smooth transition of five different levels of tasks.

[0056] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A redundancy-based robotic arm acceleration layer hierarchical control method considering priority switching, characterized in that, The method includes the following steps: S1: Obtain the task set of the redundant robotic arm and set the initial priority sequence; establish the kinematic model of the redundant robotic arm in the acceleration layer, and calculate the Jacobian matrix and task space input vector corresponding to each task; for tasks with... One degree of freedom, needs to be executed For a redundant robotic arm performing a task, establish its kinematic model at the acceleration layer: in, Index representing task priority ( (Highest priority) Indicates the first Jacobian matrix of each priority task For the dimensions of this task; Represents the joint acceleration vector; Indicates the first The task space reference input vector for each priority task; S2: For scenarios where the task dimension may exceed the degrees of freedom of the robotic arm, a hierarchical null projection matrix is ​​constructed based on task priority; the hierarchical null projection matrix achieves task space decoupling and strictly guarantees task priority; S3: Construct a multi-level physical constraint model and define state constraint operators. It is used to handle the joint velocity constraints and joint acceleration constraints of the robotic arm; S4: Construct an error-driven priority switching hierarchical recurrent neurodynamic controller; S5: Input the parameters calculated in steps S1 to S4 into the priority switching hierarchical recurrent neurodynamic controller for iterative solution to obtain a smooth and continuous joint acceleration control signal that satisfies all physical constraints. S6: Control the redundant robotic arm to perform multiple tasks based on the solution obtained in step S5.

2. The redundancy-based acceleration layer hierarchical control method for a robotic arm considering priority switching as described in claim 1, characterized in that, In step S2, for the first Level 1 priority task ( Construct the null space projection matrix The specific calculation formula is as follows: ,in, It is the identity matrix. For the first The residual null space projection matrix obtained from the task calculation. For the first The effective subspace component occupied by the level task; the effective subspace component The calculation is based on The numerical state adopts a segmented strategy: (1) Normal mode: when When the matrix is ​​non-singular, the subspace components are calculated using pseudo-inverse operations: ； (2) Singular processing mode: when When the matrix is ​​singular or zero, the transpose operation is used to approximate the pseudo-inverse operation for calculating the subspace components: ； in, For the first Jacobian matrix for level-one tasks, This indicates the Moore Penrose pseudo-inverse operation. This represents the matrix transpose operation.

3. The redundancy-based acceleration layer hierarchical control method for a robotic arm considering priority switching as described in claim 1, characterized in that, The state constraint operator Used to constrain motion variables in joint space, it is defined as a piecewise function with saturation properties; for the ... Input variables of each joint , The expression is: in, These are the upper and lower bounds of the dynamic limit allowed for the joint, respectively. These two boundary values ​​are jointly determined by the joint's velocity and acceleration constraints, and the specific calculation formula is as follows: ； in, For the current moment Joint velocity of each joint; The first The physical lower and upper limits of the joint velocity of each joint; The first The physical lower and upper limits of joint acceleration for each joint; The scaling factor is used to convert the velocity margin into an equivalent acceleration constraint. and These represent the functions for finding the maximum and minimum values, respectively.

4. The redundancy-based acceleration layer hierarchical control method for a robotic arm considering priority switching as described in claim 1, characterized in that, In step S4, the priority switching hierarchical recurrent neurodynamic controller is updated using the following state evolution equation: ， in, Given an inertial time constant, the state evolution equation is configured to solve a generalized multi-task programming problem, defined as: given that... While constraining, minimize the joint acceleration energy objective function. ; For dynamic compensation of bias; reference control quantity Defined as ,in This is a horizontally concatenated matrix of null projection matrices for each task level; For each level of task-driven terms, there is a vertically stacked vector; where, Sort the index according to the current priority. For the corresponding task's Lagrange multiplier.

5. A layered control method for acceleration layer of a redundant robotic arm considering priority switching, as described in claim 4, is characterized in that... The dynamic compensation bias The update follows the following state-preserving logic: Set the initial dynamic compensation bias to a zero vector, i.e. During the control cycle, the task priority sequence was detected to have changed from the first... Phase switch to the first The moment of a phase : (1) Lock the control output at the end of the previous stage. and the Lagrange multiplier vectors corresponding to each level of task ; (2) At the current moment after the switch Construct virtual reference control quantity It is defined as the null projection matrix updated at the current time. and the task-driven items updated at the current moment With the Lagrange multipliers locked in step (1) The control quantity obtained through calculation; (3) Calculate the current dynamic compensation bias in real time. : ， in, This is the residual value of the compensation accumulated from all previous switching phases; through the aforementioned dynamic update logic, utilizing... and The real-time difference eliminates joint acceleration jumps caused by abrupt changes in the projection matrix structure.

6. A layered control method for acceleration layer of a redundant robotic arm considering priority switching, as described in claim 1, is characterized in that... The acceleration layer kinematic relationship model in step S1 is defined as follows: , among which, the Task space input vector for each task It includes position error feedback items and velocity error feedback items.

7. A layered control method for acceleration layer of a redundant robotic arm considering priority switching, as described in claim 1, is characterized in that... In step S5, after iterative solution, a smooth and continuous joint acceleration control signal that satisfies all physical constraints is obtained, thereby controlling the redundant robotic arm to achieve multi-task operation.

Images