Robot adaptive control method based on human-in-the-loop and asymmetric potential barrier

Abstract

The invention relates to the field of robot control, in particular to a robot adaptive control method based on a human-in-the-loop and an asymmetric barrier, and the method comprises the following steps: obtaining a reconstructed decoupling constraint based on a U-K principle; asymmetric logarithmic barrier transformation based on a dynamic skew factor is constructed, and bounded constraints are mapped to an unbounded virtual space; an ideal constraint force without a Lagrange multiplier is constructed based on decoupling constraint and a U-K principle, and a self-adaptive robust controller and an optimal decision model based on a fuzzy theory are constructed; and constructing a dynamic adjustment mechanism of a basic weight based on a constraint following error norm and an internal force safety index, and combining a smooth regional transition mechanism with a human decision based on a fuzzy theory to synthesize final control input. By orthogonally decomposing the task space, it is ensured that the robot can independently and strictly maintain the clamping internal force between the two arms while responding to a human random motion instruction, and objects are effectively prevented from sliding off in the cooperation process.

Description

Technical Field

[0001] This invention relates to the field of robot control, and more particularly to an adaptive control method for robots based on human presence in the loop and asymmetric potential barriers. Background Technology

[0002] Dual-arm collaborative robots, with their humanoid manipulator structure and high flexibility, have shown great application potential in scenarios such as industrial precision assembly, collaborative material handling, medical surgical assistance, and rehabilitation services. Unlike single-arm robots, dual-arm collaborative robots, when performing tasks, not only need to control the position and trajectory of the two end effectors, but also need to precisely control the relative position and force between the two end effectors to prevent objects from slipping or being crushed and damaged. Existing dual-arm robot control methods typically employ HITL control technology, but this technology has the following shortcomings when applied:

[0003] Existing impedance control or admittance control methods typically treat the two arms as two independent entities or a single rigid unit, lacking independent mathematical descriptions of "internal force maintenance" and "external motion." During human-machine collaboration, when the operator applies an unstable dragging force, it can easily interfere with the clamping force between the two arms, leading to a decrease in collaborative safety.

[0004] Existing constraint-following control techniques typically employ symmetric transformations, such as the tangent function, to map bounded constraints into unbounded space. However, real-world working environments are often asymmetric (e.g., one side is a wall requiring strict avoidance, while the other side is open space allowing for more relaxed constraints). Using symmetric transformations can lead to unnecessarily strong constraints being imposed on the side furthest from the obstacle, reducing the utilization of the operating space and the naturalness of human-computer interaction. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides the following technical solution:

[0006] The robot adaptive control method based on human-in-the-loop and asymmetric potential barrier includes the following steps:

[0007] S10: Decompose the task space of the dual-arm robot into an absolute motion subspace and a relative internal force subspace, and obtain the decoupling constraints for reconstruction based on the UK principle.

[0008] S20: Based on asymmetric environmental constraints, construct an asymmetric logarithmic barrier transformation based on dynamic skew factor to map bounded constraints to an unbounded virtual space.

[0009] S30: Constructing ideal constraint forces without Lagrange multipliers based on decoupling constraints and UK principle.

[0010] S40: Based on a collaborative control architecture and ideal constraints, construct an adaptive robust controller on the machine side and an optimal decision-making model based on fuzzy theory on the human side.

[0011] S50: A dynamic adjustment mechanism for basic weights is constructed based on the constraint following error norm and internal force safety index. It combines a smooth regional transition mechanism with human decision-making based on fuzzy theory to synthesize the final control input.

[0012] As an improvement to the above technical solution, before step S10 is executed, a Lagrange dynamics model needs to be constructed based on parameter uncertainties and environmental disturbances.

[0013] As an improvement to the above technical solution, step S10 includes the following steps:

[0014] S11: Define absolute motion coordinates and relative internal force coordinates based on the end positions of the two arms, and obtain the Jacobian relationship based on these coordinates.

[0015] S12: Construct decoupling constraints based on the Jacobian relation, which include the absolute motion Jacobian matrix and the relative internal force Jacobian matrix.

[0016] S13: Based on the UK principle, obtain the decoupling constraints after differential reconstruction.

[0017] As an improvement to the above technical solution, the asymmetric logarithmic barrier transformation includes the following formula:

[0018]

[0019] in, This is the dynamic skew factor. Used to describe the spatial pose of the clamped object. and for Spatial constraint boundary.

[0020] As an improvement to the above technical solution, the ideal constraint force includes the following formula:

[0021]

[0022] in, Here is the system inertia matrix. Represents the Moore-Penrose generalized inverse matrix; Indicates system time. Represents coordinates, Indicates speed, Indicates acceleration. Let the system's free acceleration be in an unconstrained state. For the decoupling constraint matrix, The vector is in the second-order differential constraint form of the decoupling constraint matrix.

[0023] As an improvement to the above technical solution, the machine-side adaptive robust controller uses an adaptive law to update parameters online with the constraints of the error to eliminate the influence of system parameter drift, and constructs a robust feedback term to compensate for parameter uncertainty and environmental disturbances, so as to ensure that the actual trajectory of the system converges to the ideal trajectory calculated by UK theory.

[0024] As an improvement to the above technical solution, the construction of the optimal decision-making model based on fuzzy theory on the human side includes the following steps:

[0025] S31: Construct a cost functional that integrates UK-constrained energy cost and human-machine adversarial power.

[0026] S32: Solve the Euler-Lagrange equations using the variational method and obtain the closed-form analytical expression of the optimal membership function.

[0027] As an improvement to the above technical solution, step S50 includes the following steps:

[0028] S51: Obtain the norm of the dynamic constraint following error and use a modified Sigmoid function to obtain the basic weights.

[0029] S52: Define the internal force safety index. When the internal force safety index exceeds the preset safety threshold, the basic weight is forcibly set to 0, and the system enters the internal force priority maintenance mode. Only the control instructions of the relative internal force subspace on the machine side are executed. When the internal force safety index exceeds the preset safety threshold, no action is triggered.

[0030] As an improvement to the above technical solution, the final control input includes the following formula:

[0031]

[0032] in, Machine-side adaptive robust UK control torque, fully machine-dominated. It is the human torque directly measured by the sensor. This is the basic weight control parameter, with a value of 0 or 1. Based on the weights, This is the optimal membership function.

[0033] The beneficial effects of this invention are:

[0034] By orthogonally decomposing the task space, the robot can independently and strictly maintain the gripping force between its two arms while responding to random human movement commands, effectively preventing objects from slipping during the collaborative process and achieving higher safety and decoupling. Furthermore, the proposed asymmetric logarithmic barrier transformation can dynamically adjust the shape of the virtual potential field according to the actual distribution of environmental obstacles, ensuring both the safety of unilateral obstacle avoidance and preserving the operational flexibility of the safe side. Moreover, based on UK theory, an analytical closed-form solution for the constraint force is obtained, avoiding complex iterative calculations and significantly improving the real-time response capability of the control system. Attached Figure Description

[0035] Figure 1 This is a block diagram illustrating the principle logic of the present invention;

[0036] Figure 2 This is a schematic diagram of the structure of a dual-arm robot in the prior art. Detailed Implementation

[0037] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0038] Existing dual-arm robot control methods typically employ HITL control technology, but this technology has the following shortcomings when applied:

[0039] Existing impedance control or admittance control methods typically treat the two arms as two independent entities or a single rigid unit, lacking independent mathematical descriptions of "internal force maintenance" and "external motion." During human-machine collaboration, when the operator applies an unstable dragging force, it can easily interfere with the clamping force between the two arms, leading to a decrease in collaborative safety.

[0040] Existing constraint-following control techniques typically employ symmetric transformations, such as the tangent function, to map bounded constraints into unbounded space. However, real-world working environments are often asymmetric (e.g., one side is a wall requiring strict avoidance, while the other side is open space allowing for more relaxed constraints). Using symmetric transformations can lead to unnecessarily strong constraints being imposed on the side furthest from the obstacle, reducing the utilization of the operating space and the naturalness of human-computer interaction.

[0041] To resolve the above issues, please refer to Figure 1 and Figure 2 This paper provides an adaptive control method for robots based on human-in-the-loop and asymmetric potential barriers, comprising the following steps:

[0042] First, for the dual-arm collaborative robot system, a dynamic model incorporating parameter uncertainties and environmental disturbances is constructed.

[0043]

[0044] in, Indicates system time. Represents coordinates, Indicates speed, Indicates acceleration. Indicates an uncertain parameter. This indicates an unknown environmental disturbance. This is the control input (whose size may depend on human decision-making). Set and It is compact, representing respectively and The possible locations are all compact. Here is the system inertia matrix. Represents Coriolis force / centrifugal force. Represents gravity. Indicates the input matrix, matrix / vector , , , They all have appropriate dimensions. , , , They are all continuous.

[0045] To address the issue of maintaining internal forces during dual-arm collaboration, this embodiment proposes a task space decoupling strategy, including the following steps:

[0046] S10: Decompose the task space of the dual-arm robot into an absolute motion subspace and a relative internal force subspace, and obtain the decoupling constraints for reconstruction based on the UK principle.

[0047] Specifically, step S10 includes the following steps:

[0048] S11: Define absolute motion coordinates and relative internal force coordinates based on the end positions of the two arms, and obtain the Jacobian relationship based on these coordinates.

[0049] Let the positions of the ends of the two arms be respectively .definition:

[0050] Absolute Motion : Describes the spatial pose of the clamped object.

[0051]

[0052] Relative Internal Coordinates : Describes the relative distance between the two arms, which is directly related to the clamping force.

[0053]

[0054] Differentiating the above coordinates yields the Jacobian relation. .

[0055] S12: Construct decoupling constraints based on the Jacobian relationship, including the absolute motion Jacobian matrix and the relative internal force Jacobian matrix;

[0056] The decoupling constraints include the following:

[0057]

[0058] in, For absolute motion Jacobian matrix, It is the Jacobian matrix of relative internal forces.

[0059] S13: Based on the UK principle, obtain the decoupling constraints after differential reconstruction.

[0060] To apply UK theory, the control objective is uniformly written in the form of second-order differential constraints. The specific formula is as follows:

[0061]

[0062] in, The reference acceleration representing the absolute motion subspace. This is used to set the desired clamping acceleration between the two arms to maintain constant internal force constraints. Through this construction, the matrix... It contains complete topological constraint information for the dual-arm system.

[0063] In real-world collaborative scenarios, constraints are often asymmetric (e.g.: However, the obstacle is only (One side). To avoid the space wastage caused by traditional symmetric transformations, this embodiment provides the following solution:

[0064] S20: Based on asymmetric environmental constraints, construct an asymmetric logarithmic barrier transformation based on dynamic skew factor to map bounded constraints to an unbounded virtual space.

[0065] The asymmetric logarithmic barrier transformation includes the following equation:

[0066]

[0067] in, This is the dynamic skew factor. Used to describe the spatial pose of the clamped object. and for Spatial constraint boundary.

[0068] When detected When there are high-risk obstacles on the side, increase This causes the barrier function to rise sharply on that side, forming a "hard wall".

[0069] when When side safety is involved, reduce This makes the barrier on that side gentler, forming a "soft wall".

[0070] After transformation, the original bounded coordinates Mapped to unbounded virtual coordinates In subsequent control, only... By performing tracking control, the original coordinates can be automatically preserved. It remains within a safe range. At this point, in step S13... It needs to be replaced with based on The reference acceleration.

[0071] S30: Constructing ideal constraint forces without Lagrange multipliers based on decoupling constraints and UK principle.

[0072] The Udwadia-Kalaba (UK) theory provides an analytical method for solving the dynamics of constrained systems without introducing Lagrange multipliers, and can use the UK fundamental equations to calculate ideal constraint forces.

[0073] The basic equation of the UK is shown below:

[0074]

[0075] (Also written in the implementation formula) ) represents the ideal constraint torque of Udwadia-Kalaba.

[0076] Based on the UK fundamental equations, the above decoupling constraints are satisfied. Required ideal constraint for:

[0077]

[0078] in, Here is the system inertia matrix. Represents the Moore-Penrose generalized inverse matrix; Let the system's free acceleration be in an unconstrained state. For the decoupling constraint matrix, This is a vector in the second-order differential constraint form of the decoupling constraint matrix. The formula directly gives the condition that the system must not let go of either arm. (Constraints) and do not hit the wall ( The minimum norm compensation force required for the constraint.

[0079] S40: Based on a collaborative control architecture and ideal constraints, construct an adaptive robust controller on the machine side and an optimal decision-making model based on fuzzy theory on the human side.

[0080] Due to the uncertainty of model parameters in actual systems ( Unknown) and external interference ( ), relying solely on the nominal model Accuracy cannot be guaranteed. Therefore, an adaptive robust controller is designed on the machine side, including the following formula:

[0081]

[0082] in, For the nominal UK term, the practical parameter estimate is... Substitute these values ​​into the formula for calculating ideal constraint forces for calculation. As an adaptive compensation term, the parameters are updated online according to the error using an adaptive law to eliminate the influence of system parameter drift. This is a robust feedback term used to compensate for parameter uncertainties and environmental disturbances, ensuring that the actual trajectory of the system converges to the ideal trajectory calculated by UK theory.

[0083] The adaptive law includes the following formula:

[0084]

[0085] Where Z is the constraint following error. Here is the gain matrix. For the regression matrix

[0086] Robust feedback terms include the following:

[0087]

[0088] in, It is a positive definite gain matrix. This represents the robust gain boundary.

[0089] To enable the robot to understand and optimize fuzzy human intentions, this embodiment models decision fusion as a functional optimization problem to obtain the optimal decision model based on fuzzy theory from the human side. Specifically, the steps include:

[0090] S31: Construct a cost functional that integrates UK-constrained energy cost and human-machine adversarial power, aiming to minimize the path length of decision changes and human-machine adversarial energy.

[0091] The cost functional is shown below:

[0092]

[0093] in, For human biosignal input, For membership function, These are the weighting coefficients. (Also written in the implementation formula) ) represents the ideal constraint torque of Udwadia-Kalaba. This represents the physical cost (i.e., UK constraint energy) that the system incurs to satisfy the current constraints.

[0094] S32: Solve the Euler-Lagrange equations using the variational method and obtain the closed-form analytical expression of the optimal membership function.

[0095] Solving the Euler-Lagrange equations using the variational method includes the following equations:

[0096]

[0097] Solving for the optimal membership function yields the optimal membership function. The closed-form solution is in hyperbolic cosine form as shown in the following equation:

[0098]

[0099] in , The shape width constant, The center offset constant is determined by the boundary conditions of the variational method. For robustness cost parameters, The normalization coefficient is... To control the gain for humans.

[0100] This function can automatically calculate the optimal human intention weight based on the intensity of human input and the constraint energy of the current system, thus achieving a smooth and optimal fusion of human and machine intentions.

[0101] To further ensure physical safety, especially to prevent slippage during clamping, a dynamic weight adjustment mechanism is designed. This includes the following steps:

[0102] S50: A dynamic adjustment mechanism for basic weights is constructed based on the constraint following error norm and internal force safety index. It combines a smooth regional transition mechanism with human decision-making based on fuzzy theory to synthesize the final control input.

[0103] Specifically, step S50 includes the following steps:

[0104] S51: Obtain the norm of the dynamic constraint following error and use a modified Sigmoid function to obtain the basic weights.

[0105] The constrained following error norm includes the following formula:

[0106]

[0107] The base weights obtained from the modified Sigmoid function are shown in the following formula:

[0108]

[0109] in, The center point is usually taken as the center point. This indicates the point where control is equally divided.

[0110] When the error is small (system compliant), Allowing human intervention; when the error is large (such as when about to hit a wall), Machine-driven.

[0111] S52: Define the internal force safety index. When the internal force safety index exceeds the preset safety threshold, the basic weight is forcibly set to 0, and the system enters the internal force priority maintenance mode. Only the control instructions of the relative internal force subspace on the machine side are executed. When the internal force safety index exceeds the preset safety threshold, no action is triggered.

[0112] The internal force safety index is defined as follows:

[0113]

[0114] in, It is an external tangential force. For internal normal force, is the coefficient of friction.

[0115] The final control input includes the following formula:

[0116]

[0117] in, Machine-side adaptive robust UK control torque, fully machine-dominated. It is the human torque directly measured by the sensor, with a value of 0 or 1. Based on the weights, The optimal membership function. As the basic weight control parameter, for The judgment logic is as follows:

[0118]

[0119] Through the above steps, this embodiment achieves high safety, high precision, and natural human-machine collaborative control of the dual-arm collaborative robot in complex, uncertain, and asymmetric constraint environments.

[0120] The above embodiments are merely illustrative of the technical solutions of the present invention and are not intended to limit it. Anyone skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A robot adaptive control method based on human-in-the-loop and asymmetric barrier, characterized in that, Includes the following steps: S10: Decompose the task space of the dual-arm robot into an absolute motion subspace and a relative internal force subspace, and obtain the decoupling constraints for reconstruction based on the UK principle; S20: Based on asymmetric environmental constraints, construct an asymmetric logarithmic barrier transformation based on dynamic skew factor to map bounded constraints to unbounded virtual space; S30: Constructing ideal constraint forces without Lagrange multipliers based on decoupling constraints and the UK principle; S40: Based on a collaborative control architecture and ideal constraints, construct an adaptive robust controller on the machine side and an optimal decision-making model based on fuzzy theory on the human side. S50: A dynamic adjustment mechanism for basic weights is constructed based on the constraint following error norm and internal force safety index. It combines a smooth regional transition mechanism with human decision-making based on fuzzy theory to synthesize the final control input.

2. The human-in-the-loop and asymmetric barrier based robot adaptive control method of claim 1, wherein: Before performing step S10, a Lagrange dynamics model needs to be constructed based on parameter uncertainties and environmental disturbances.

3. The robot adaptive control method based on human-in-the-loop and asymmetric potential barrier according to claim 1, characterized in that: Step S10 includes the following steps: S11: Define absolute motion coordinates and relative internal force coordinates based on the end positions of the two arms, and obtain the Jacobian relationship based on these coordinates; S12: Construct decoupling constraints based on the Jacobian relationship, including the absolute motion Jacobian matrix and the relative internal force Jacobian matrix; S13: Based on the UK principle, obtain the decoupling constraints after differential reconstruction.

4. The robot adaptive control method based on human-in-the-loop and asymmetric potential barrier according to claim 1, characterized in that: The asymmetric logarithmic barrier transformation includes the following equation: in, This is the dynamic skew factor. Used to describe the spatial pose of the clamped object. and for Spatial constraint boundary.

5. The robot adaptive control method based on human-in-the-loop and asymmetric potential barrier according to claim 1, characterized in that: The ideal constraint force includes the following formula: in, Here is the system inertia matrix. Represents the Moore-Penrose generalized inverse matrix; Let the system's free acceleration be in an unconstrained state. For the decoupling constraint matrix, The vector is in the second-order differential constraint form of the decoupling constraint matrix.

6. The robot adaptive control method based on human-in-the-loop and asymmetric potential barrier according to claim 1, characterized in that: The machine-side adaptive robust controller uses an adaptive law to update parameters online with the constraints of the error to eliminate the influence of system parameter drift, and constructs a robust feedback term to compensate for parameter uncertainty and environmental disturbances, so as to ensure that the actual trajectory of the system converges to the ideal trajectory calculated by UK theory.

7. The adaptive control method for robots based on human-in-the-loop and asymmetric potential barrier according to claim 1, characterized in that: The construction of the optimal decision-making model based on fuzzy theory on the human side includes the following steps: S31: Construct a cost functional that integrates UK-constrained energy cost and human-machine adversarial power; S32: Solve the Euler-Lagrange equations using the variational method and obtain the closed-form analytical expression of the optimal membership function.

8. The robot adaptive control method based on human-in-the-loop and asymmetric potential barrier according to claim 1, characterized in that: Step S50 includes the following steps: S51: Obtain the norm of the dynamic constraint following error and use a modified Sigmoid function to obtain the basic weights; S52: Define the internal force safety index. When the internal force safety index exceeds the preset safety threshold, the basic weight is forcibly set to 0, and the system enters the internal force priority maintenance mode. Only the control instructions of the relative internal force subspace on the machine side are executed. When the internal force safety index exceeds the preset safety threshold, no action is triggered.

9. The adaptive control method for robots based on human-in-the-loop and asymmetric potential barrier according to any one of claims 1-8, characterized in that: The final control input includes the following formula: in Machine-side adaptive robust UK control torque, fully machine-dominated. It is the human torque directly measured by the sensor. This is the basic weight control parameter, with a value of 0 or 1. Based on the weights, This is the optimal membership function.

Images