Exoskeleton teleoperation-oriented manipulator compliant control system and method

Abstract

The application provides an exoskeleton teleoperation-oriented manipulator compliance control system and method. Under the premise that an exoskeleton or a master device gives a reference pose, the method converts external contact force measured at the end of a manipulator into compliant displacement relative to the reference pose through a mobility model, and the compliant displacement and the reference pose are superimposed in a pose space and then sent to a motion solver for execution, so that the slave manipulator can track the operator's intention while maintaining compliance and safety to the contact environment. The compliance control loop of the application is decoupled from the specific manipulator body and the specific motion solver backend, and can be reused between different manipulators and different solvers.

Description

Technical Field

[0001] This invention relates to the fields of robot control, compliant control of robotic arms, human-machine collaborative teleoperation, and human-machine interaction safety. Specifically, it relates to a compliant control system and method for exoskeleton teleoperation of robotic arms. Background Technology

[0002] In exoskeleton teleoperation and human-machine collaborative operations, the slave robotic arm needs to both follow the operator's movement intentions issued through the exoskeleton or master device, and remain compliant when in contact with the environment, workpiece or human body to avoid damage or injury caused by rigid collisions.

[0003] Existing processing methods typically include: 1. Pure position-based master-slave teleoperation. This method directly issues the operator's pose command as the end-effector's target pose, with the robotic arm rigidly tracking it. Once the end-effector comes into contact with the environment, because it is unaware of external forces, the robotic arm will continuously press down, which can easily cause excessive contact force and pose a safety risk in human-machine interaction.

[0004] 2. Force control or impedance control implemented directly at the joint torque layer. This method typically relies on the torque interface of a specific robotic arm and the vendor's software development kit. The control algorithm is strongly coupled with the specific robotic arm body, making it difficult to port to other robotic arms or other solution backends. At the same time, it lacks a reference trajectory superposition mechanism and workspace safety constraints for teleoperation.

[0005] 3. Force feedback is only superimposed on the end target, but the difference between "operator-expected reference motion" and "compliant motion caused by environmental contact" is not distinguished. This method is prone to introducing sensor noise, gravity components, or disturbances caused by solution failures into the target during teleoperation, resulting in end-point jitter, drift, or being pushed out of the reachable workspace.

[0006] 4. No safety envelope or anti-integral saturation treatment was set for compliant displacement. Under continuous pressure or strong contact, the internal state of the admittance model continuously integrates, and the end effector is driven to the workspace boundary and stops. It cannot return smoothly after the operator releases the hand.

[0007] It should be noted that admittance-based compliant control is already quite common in the field of robotics. The difficulty of this invention does not lie in the admittance model itself, but in how to apply compliant control to the continuously changing upper-level / teleoperation commands, so that the reference pose of the slave robotic arm is continuously given and changed by the exoskeleton teleoperation (the operator's joint state is mapped to the end-effector TCP pose through positive kinematics) or the upper-level motion trajectory, so as to avoid conflict between compliant compliance and the constantly updated reference commands, and to ensure safe compliance during contact and stable following of commands when disengaging from contact.

[0008] Therefore, a compliant control method is needed that can track the continuously changing reference pose given by exoskeleton teleoperation or upper-level trajectory, be compliant with contact forces, be decoupled from the specific robotic arm body and the specific motion solution backend, and have human-machine interaction safety constraints.

[0009] Patent application CN115157252A discloses a sensor-based compliant control method for a robotic arm, comprising: determining redundant joints based on the number of degrees of freedom of the robotic arm and the sensor measurement dimensions; acquiring the external forces acting on the robotic arm through sensors; calculating the desired pose of the robotic arm under the control of the external forces; obtaining the control quantities of the non-redundant joints through inverse kinematics; and controlling the movement of the corresponding joints accordingly. However, this patent cannot completely solve the existing technical problems, nor can it meet the needs of this invention.

[0010] Terminology Explanation: Compliant control refers to a control method where the end effector of a robotic arm, when subjected to external contact forces, produces compliant displacement according to a preset dynamic relationship of mass, damping, and stiffness, rather than rigidly rejecting the external force. This invention employs admittance-type compliant control, i.e., force is the input and displacement is the output.

[0011] Admittance model: refers to a second-order mass-spring-damped dynamic relationship composed of mass parameters, damping parameters, and stiffness parameters. Its input is the external force acting on the end effector of the robotic arm, and its output is the three-dimensional translational compliance displacement or six-dimensional compliance pose of the end effector relative to the reference pose.

[0012] Exoskeleton teleoperation: refers to a human-machine collaborative operation method in which the operator wears an exoskeleton or operates a master device, the exoskeleton collects the operator's joint state, maps it to the target pose of the slave robotic arm through positive kinematics, and then sends the operator's intention to the slave robotic arm.

[0013] Reference pose: refers to the target pose that the robotic arm's end effector should track when it is not subjected to external forces, denoted as In this invention, the reference pose is mainly obtained by calculating the TCP pose of the robot's end effector through positive kinematics of the joint states operated by the exoskeleton, or it can be given by the upper-level motion planning trajectory.

[0014] Forward kinematics (FK): refers to the forward kinematic mapping of the end effector pose calculated from the joint states of a robotic arm or exoskeleton. This invention uses it to convert the joint states of an exoskeleton into a reference TCP pose for the slave robotic arm.

[0015] End-effector coordinate system (TCP): refers to the coordinate system in which the center point of the end-effector of the robotic arm is located. In this invention, the pose of the end-effector of the robotic arm is expressed in a seven-dimensional form of position plus attitude quaternion.

[0016] Force-torque sensor: refers to a six-dimensional force-torque sensor installed at the end of a robotic arm and connected to the end tool. It is used to measure external contact forces and torques, and its readings are expressed in the sensor coordinate system. The raw readings need to be transformed by coordinates, gravity compensation and dead zone processing before they are used as input to the admittance model.

[0017] Motion solver: refers to the solution module that solves the pose of the end-point target in Cartesian space into a target in joint space. This invention uses a motion solver based on optimization (such as Model Predictive Control, MPC) to achieve this.

[0018] Robot client interface and motion solver interface: These refer to the two minimal abstract interfaces defined in this invention. The compliant control loop interacts with the specific robotic arm body and the specific solver backend only through these two interfaces; the loop itself does not incorporate any robot manufacturer's software development kit or specific solver backend code.

[0019] Compliant displacement: The end-effector translation of the guide model relative to the reference pose under the action of external forces, denoted as .

[0020] Safety envelope: refers to the upper limit constraint applied to the compliant displacement amplitude, so that the end effector of the robotic arm will not be driven outside the reachable workspace under strong contact or continuous pushing. Summary of the Invention

[0021] To address the shortcomings of existing technologies, the purpose of this invention is to provide a compliant control system and method for exoskeleton teleoperation of robotic arms.

[0022] The compliant control system for exoskeleton teleoperation provided by the present invention includes: The reference pose acquisition module acquires the reference pose given by the exoskeleton teleoperation. An external force processing module has its input end connected to a force-torque sensor installed at the end of the robotic arm. It receives the external force measured by the sensor and performs gravity compensation, coordinate transformation, and force dead zone processing on the external force. The admittance model module has its input end connected to the output end of the external force processing module. It performs integral calculation based on the processed external force and outputs the compliant displacement relative to the reference pose. A safety constraint module, whose input is connected to the output of the admittance model module, applies a safety envelope constraint to the compliant displacement, limiting the upper limit of the amplitude of the compliant displacement; The pose overlay module has a first input terminal connected to the output terminal of the reference pose acquisition module and a second input terminal connected to the output terminal of the safety constraint module. It overlays the compliant displacement and the reference pose in the pose space to generate the end-effector target pose. The motion solver interface module has its input end connected to the output end of the pose superposition module, and outputs the end-effector target pose to the motion solver based on optimization solution to obtain the joint target of the robotic arm; The robot client interface module has its input end connected to the output end of the motion solver interface module, and its output end connected to the robot arm body, so as to send the joint target to the robot arm body for execution. The admittance model module and the pose superposition module constitute a compliant control loop. The compliant control loop interacts with the robotic arm body through the robot client interface module and with the motion solver through the motion solver interface module.

[0023] Preferably, when the external force processing module performs the gravity compensation, it uses the weight of the end tool as input; when the external force processing module performs the coordinate transformation, it supports both base coordinate system mode and tool coordinate system mode. In the base coordinate system mode, the external force processing module estimates the residual component of the tool gravity at the end of the robotic arm based on the change of the current posture of the end of the robotic arm relative to the initial posture, subtracts it, and then rotates the force after subtracting the residual component to the base coordinate system. In the tool coordinate system mode, the external force processing module rotates the sensor readings to the end tool coordinate system according to the fixed rotation relationship between the sensor and the end tool.

[0024] Preferably, the external force processing module applies the force dead zone processing in the following ways: Calculate the magnitude of the external contact force; When the magnitude of the external contact force is less than or equal to the force dead zone threshold, the external contact force is set to zero. When the magnitude of the external contact force is greater than the force dead zone threshold, the external contact force is kept at its original value and input into the admittance model module.

[0025] Preferably, the admittance model module adopts a second-order mass-spring-damped admittance model, and for each translational degree of freedom, the dynamic relationship of the admittance model module satisfies:

[0026] Where m is the virtual mass parameter, c is the virtual damping parameter, k is the virtual stiffness parameter, s is the compliant displacement relative to the reference pose, v is the compliant velocity, a is the compliant acceleration, and f is the external force component acting on this degree of freedom. The admittance model module performs discrete integration on the dynamic relationship in each control cycle, and expands it into a six-dimensional form when attitude compliance is required. The three translational degrees of freedom and the three rotational degrees of freedom are integrated according to the same mass-spring-damping relationship to obtain the six-dimensional compliant pose.

[0027] Preferably, the safety constraint module applies a safety envelope constraint to the compliant displacement in the following ways: Calculate the modulus of the compliant displacement; When the magnitude of the compliant displacement is less than or equal to the upper limit of the safety envelope, the compliant displacement remains unchanged; When the magnitude of the compliant displacement is greater than the upper limit of the safety envelope, the compliant displacement is scaled by a scaling factor so that the magnitude of the compliant displacement is equal to the upper limit of the safety envelope. When scaling is triggered, the safety constraint module also synchronously resets the internal state of the admittance model module, including setting the internal compliance displacement state to the scaled compliance displacement and attenuating the internal compliance velocity state by the same scaling factor.

[0028] Preferably, the virtual mass parameters, virtual damping parameters, and virtual stiffness parameters of the admittance model module are given according to a unified tuning principle: The virtual stiffness parameters are determined according to the desired static compliance. The virtual quality parameters are selected according to the desired response characteristics; The virtual damping parameters are determined according to the critical damping relationship, satisfying:

[0029] Where c is the virtual damping parameter, Where is the damping ratio, m is the virtual mass parameter, and k is the virtual stiffness parameter; The upper limit of the safety envelope is determined in conjunction with the virtual stiffness parameter and the maximum external contact force allowed by the design, satisfying:

[0030] in, Indicates the maximum external contact force allowed by the design or expected by the task. This indicates the maximum allowable compliance displacement.

[0031] Preferably, the pose overlay module overlays the compliant displacement and the reference pose in the pose space in the following manner: A pure translation transformation is constructed using the compliant displacement. This pure translation transformation is then superimposed on the reference pose to obtain the end-effector target pose, satisfying the following:

[0032] in, The target pose is executed for the end effector. Indicates to adapt to displacement Constructed pure translation transformation, For reference pose; The pose overlay module also converts the end-effector target pose into a seven-dimensional end-effector target pose consisting of a position plus an attitude quaternion, which is then used as input to the motion solver.

[0033] Preferably, it further includes a joint signal processing module, which performs low-pass filtering on the read joint states and uses the filtered joint states as joint seeds for the motion solver; The low-pass filter employs a first-order or second-order low-pass filter based on bilinear transformation. The control system also includes a solution failure handling module. When the motion solver returns an empty solution or an invalid solution containing non-finite values, the solution failure handling module determines that the solution for this cycle is invalid and maintains the current pose of the robotic arm. The control system also includes an anomaly rollback module. When an anomaly is detected in the joint target sent to the robotic arm body, the anomaly rollback module rolls back to the current joint state of the robotic arm body.

[0034] Preferably, the reference pose acquired by the reference pose acquisition module is obtained by calculating the TCP pose of the robotic arm end effector through positive kinematics of the joint states of the exoskeleton teleoperation, or by the upper-level motion planning trajectory. The motion solver interface module is connected to a model predictive control solver.

[0035] According to the compliant control method for exoskeleton teleoperation provided by the present invention, the compliant control system for exoskeleton teleoperation is employed, and the following steps are executed cyclically with a fixed control cycle: Step 1: Read the joint status of the robotic arm and perform low-pass filtering; Step 2: Read the end effector TCP pose of the robotic arm; Step 3: Read the external force measured by the force-torque sensor installed at the end of the robotic arm; Step 4: Perform gravity compensation, coordinate transformation, and force dead zone processing on the external force; Step 5: Input the processed external force into the second-order mass-spring-damped admittance model for integration to obtain the compliant displacement relative to the reference pose given by the exoskeleton teleoperation; Step 6: Apply a safety envelope constraint to the compliant displacement; Step 7: Superimpose the compliant displacement with the reference pose in the pose space to obtain the end effector target pose; Step 8: Output the end-effector target pose to the optimization-based motion solver via the motion solver interface to obtain the joint target; Step 9: Send the joint target to the robotic arm body for execution via the robot client interface.

[0036] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention decouples the “reference motion expected by the operator” from the “compliant motion caused by environmental contact” and superimposes them in the pose space, so that the robotic arm can remain compliant with the contact environment while tracking the teleoperation intention of the exoskeleton, avoiding the rigid pressure of pure position teleoperation when in contact. (2) The present invention converts the external contact force into a bounded compliant displacement through the admittance model and sets a force dead zone so that the end can stably track the reference trajectory when there is no contact or small disturbance, and only compliant when the contact force exceeds the threshold. (3) The present invention limits the maximum compliance amplitude by adapting to the displacement safety envelope and anti-integral saturation processing, so that strong push or continuous contact will not drive the end to outside the reachable work space, and the end will return smoothly after the operator releases the hand, thus improving the safety of human-machine interaction. (4) The compliant control loop of the present invention is decoupled from the specific robotic arm body and the specific motion solution backend, and works only through two minimal abstract interfaces, so that the same compliant control algorithm can be reused between different robotic arms and different motion solvers based on optimization, reducing the cost of cross-platform porting; (5) The stiffness of this invention is determined according to static compliance, the virtual mass is selected according to response characteristics, and the damping is determined according to critical damping relationship (damping ratio of about 0.7 to 1.0), avoiding blind trial and error for each model, and making the compliant motion without obvious overshoot and smooth return to position; (6) After the present invention is set once on the same robotic arm, the same set of parameters can be used in different tasks and deployments without repeated readjustment; when changing models, the same principle can be used to quickly obtain a similar smooth feel, which is significantly simplified compared to the joint torque layer force control which requires remodeling and readjustment for each model. (7) This invention improves the smoothness of compliant motion and the robustness of the control loop by processing joint low-pass filtering, solver seed initialization, solution failure holding and abnormal rollback; (8) The gravity compensation and configurable force reference coordinate system of the present invention make the estimation of external contact force more accurate and the compliance direction more consistent with the actual contact. Attached Figure Description

[0037] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A general block diagram of compliant control for a robotic arm designed for teleoperation of exoskeletons; Figure 2 A flowchart for compliant control single-cycle operation; Figure 3 Block diagram of external force handling and admittance model; Figure 4 (a) and (b) in the figure are schematic diagrams of the compliant displacement safety envelope and the anti-integral saturation, respectively; Figure 5 (a) and (b) in the figure are schematic diagrams of the superposition of the reference pose and the compliant displacement pose, respectively; Figure 6 Diagram of the decoupled architecture between the robot body and the solution backend; Figure 7 A schematic diagram illustrating the unified tuning of admittance parameters and the adoption of tuning principles across different models. Detailed Implementation

[0038] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0039] Example 1 This invention proposes a compliant control system and method for exoskeleton teleoperation of robotic arms. Its core is as follows: using the reference pose provided by the exoskeleton teleoperation as a benchmark, the external force measured at the end effector is input into the admittance model after gravity compensation, coordinate transformation, and dead zone processing to obtain the compliant displacement relative to the reference pose; the compliant displacement and the reference pose are superimposed in the pose space, and after being constrained by a safety envelope, the joint target is solved through an abstract motion solver interface to drive the robotic arm; the entire compliant control system operates only through an abstract robot client interface and a motion solver interface, without relying on a specific robotic arm body or a specific solver backend.

[0040] The system executes the following steps in a fixed control cycle: read the joint state of the robotic arm and perform low-pass filtering; read the TCP pose of the end effector; read the external force at the end effector; transform the external force to a unified reference coordinate system through gravity compensation and coordinate transformation and apply a force dead zone; input the processed external force into the admittance model for integration to obtain the compliant displacement; apply a safety envelope constraint to the compliant displacement; superimpose the compliant displacement with the reference pose given by the exoskeleton teleoperation to obtain the end effector target pose; solve the joint target through the motion solver interface; send the joint target to the robotic arm; and wait for one control cycle before entering the next cycle.

[0041] The compliant control loop has two types of inputs: one is the reference pose. The first type of input represents the operator's intention or higher-level command. In this invention, the robot's end-effector TCP pose is mainly obtained by calculating the joint states of the exoskeleton through forward kinematics, or it can be given by the upper-level motion planning trajectory. The second type is the external contact force measured by the end-effector force-torque sensor, representing environmental constraints. After the two types of inputs are processed by the "reference trajectory channel" and the "compliance channel" respectively, they are superimposed in the pose space, thereby decoupling "where the operator / upper-level wants to go" from "how the robotic arm should comply with the contact," so that compliance does not contaminate or conflict with the continuously changing reference command.

[0042] This invention employs a second-order mass-spring-damped admittance model to describe the end effector's compliance. For each translational degree of freedom, the admittance dynamics satisfy:

[0043] Where m is the virtual mass parameter, c is the virtual damping parameter, k is the virtual stiffness parameter, s is the compliant displacement relative to the reference pose, v is its first derivative (compliant velocity), a is its second derivative (compliant acceleration), and f is the external force component acting on this degree of freedom.

[0044] The system performs discrete integration on the above equation in each control cycle. Let the current adaptive displacement be s and the adaptive velocity be v, then the adaptive acceleration is:

[0045] Then according to the control cycle Perform a status update:

[0046]

[0047] By integrating the three translational degrees of freedom in a decoupled manner, the three-dimensional compliant displacement vector of the end effector relative to the reference pose is obtained.

[0048] In scenarios requiring compliant posture, this invention extends the aforementioned admittance dynamics into a six-dimensional form. By integrating the three translational degrees of freedom and the three rotational degrees of freedom according to the same mass-spring-damping relationship, a six-dimensional compliant posture is obtained, wherein the rotational part is represented by a small amount of rotation of the end posture.

[0049] The end-effector force-torque sensor is a six-dimensional force-torque sensor mounted on the end of a robotic arm and connected to an end-effector tool. Its output force and torque are expressed in a sensor coordinate system, with the mounting direction perpendicular to the end-effector tool direction. The raw readings include the sensor's own mounting orientation, the end-effector tool's gravity component, and the actual external contact force. This invention performs the following processing before inputting the force into the admittance model: gravity compensation uses the end-effector tool's weight as input.

[0050] The system supports two force reference coordinate system configurations: 1. Base coordinate system mode. The system first estimates the residual component of the tool's gravity at the end of the device based on the change in the current attitude of the end effector relative to the initial attitude. This residual gravity is then subtracted from the sensor readings. Finally, the force is rotated to the base coordinate system to obtain the external contact force. This mode is suitable for scenarios where contact forces need to be uniformly expressed in a world coordinate system.

[0051] 2. Tool Coordinate System Mode. The system rotates the sensor readings to the end-tool coordinate system based on a fixed rotational relationship between the sensor and the end-tool to obtain the external contact force. This mode is suitable for scenarios where the force is primarily applied along the direction of the tool.

[0052] External contact force obtained after the above treatment This represents the external contact force vector; in the admittance model, the input force f for each translational degree of freedom is uniformly defined as the component of this vector in the corresponding degree of freedom (corresponding to the x, y, or z direction). The tool mass required for gravity compensation, the rotational relationship between the sensor and the tool, and the type of force reference coordinate system are all given by the configuration items.

[0053] To suppress sensor noise and slow end-effector drift caused by zero drift in the absence of contact, this invention applies a dead zone to the external contact force. Let the magnitude of the external contact force be |||. The force dead zone threshold is ,but: when‖ ||≤ At that time, the external contact force is set to zero; when‖ ‖> At that time, the external contact force remains at its original value and is input into the admittance model.

[0054] Therefore, the robotic arm only produces compliant motion when the contact force exceeds the dead zone threshold, thus maintaining the stability of the reference trajectory tracking when there is no contact or only minor disturbance.

[0055] To ensure safety in human-computer interaction, this invention applies a safety envelope constraint to the compliance displacement output by the admittance model. Let the compliance displacement be... Its modulus is || The upper limit of the safety envelope is... ,but: when‖ ||≤ At that time, the compliant displacement remains unchanged; when‖ ‖> At that time, scaling factor Scale the compliant displacement so that its magnitude equals the upper limit:

[0056]

[0057] To prevent the end effector from being "pinned" to the workspace boundary due to continuous integration of the internal state of the admittance model during sustained strong contact, this invention simultaneously resets the internal state of the admittance model upon triggering the aforementioned scaling: the internal compliance displacement state is set to the scaled compliance displacement, and the internal compliance velocity state is attenuated by the same scaling factor σ. This anti-integral saturation treatment allows the end effector to smoothly return to the vicinity of the reference pose after the operator stops pushing, rather than stagnating at the boundary.

[0058] With a secure envelope and anti-integral saturation, strong pushes or continuous contact will not drive the end beyond the reachable workspace, thereby limiting the maximum compliance amplitude and improving the safety of human-computer interaction.

[0059] This invention utilizes the reference pose provided by exoskeleton teleoperation. Compliance displacement with admittance model output Superimposed within the pose space, the pose of the end effector target is obtained. :

[0060] in, Indicates to adapt to displacement The constructed pure translation transformation. This superposition method ensures that: in the absence of external force, the compliant displacement is zero, and the end effector strictly tracks the operator's reference pose; when there is contact force, the end effector generates a compliant offset along the contact compliance direction based on the reference pose.

[0061] The system will then The position is converted into a seven-dimensional end-target pose as a position plus attitude quaternion, which is then used as input to the motion solver.

[0062] The compliant control loop of this invention operates through only two minimal abstract interfaces, thereby decoupling it from the specific robotic arm body and the specific motion solution backend: 1. Robot Client Interface. This defines the minimum capabilities the robotic arm should provide, including enabling, stopping, force data clearing, reading joint states, reading end-effector TCP pose, reading external forces, and sending joint targets. Any robotic arm adapter that meets this interface specification can be injected into the control loop.

[0063] 2. Motion Solver Interface. This interface defines the ability to solve for joint targets from Cartesian end-effector poses, including solving joint targets based on target pose and joint seeds, resetting the solver's internal state with the current joint state, and releasing solution resources. Any optimization-based motion solver adapter (e.g., a model predictive control solver) that satisfies this interface definition can be injected into a control loop.

[0064] The compliant control loop itself does not incorporate any robotic arm manufacturer's software development kit or specific solution backend code; the robotic arm manufacturer's implementation and the solution backend's implementation are isolated in a separate adapter module. Therefore, the same compliant control algorithm can be reused across different robotic arms and different solution backends, significantly reducing cross-platform porting costs.

[0065] To achieve smooth and stable compliant motion, the present invention introduces the following processing in the control loop: 1. Joint signal low-pass filtering. The system uses first- or second-order low-pass filtering based on bilinear transformation (Tustin discretization) to suppress high-frequency noise in joint feedback on the read joint states. The filtered joint states are then used as joint seeds for the solver. Several frames are accumulated before filtering is enabled at the beginning of loop startup to avoid initial value distortion.

[0066] 2. Solver Seed Initialization. When the loop starts, the solver seed is initialized using the current joint configuration of the robotic arm to prevent the first solution from jumping to the solver's default pose.

[0067] 3. Solution Failure Hold. When the motion solver returns an empty solution or an invalid solution containing non-finite values, the system determines that the solution for this cycle is invalid, maintains the current pose of the robotic arm, and does not issue abnormal targets.

[0068] 4. Abnormal rollback. When an abnormality occurs when an abnormality is sent to the robotic arm, the system rolls back to the current joint state to prevent the robotic arm from executing abnormal commands.

[0069] The above-mentioned treatments together ensure the smoothness, stability, and safety of compliant control during teleoperation.

[0070] One of the key technical details of this invention is that the admittance parameters (mass m, damping c, stiffness k) are given according to a unified tuning principle, rather than being blindly adjusted for each model. The tuning principle is based on several physical relationships in the admittance model.

[0071] 1. Stiffness and Static Compliance. Under constant contact force, when the admittance model reaches steady state, both the compliance acceleration and compliance velocity are zero. From the dynamic relationship, the steady-state compliance displacement satisfies the following conditions with the external force:

[0072] in, This represents the compliant displacement relative to the reference pose in the corresponding degree of freedom when the admittance model reaches steady state under constant external force, where the subscript... "Stabilized" is an abbreviation for "steady state"; "f" represents the treated external contact force. The component in that degree of freedom; k represents the virtual stiffness parameter of that degree of freedom. Therefore, the above formula means that the steady-state compliant displacement is equal to the ratio of the external force component to the virtual stiffness, that is, the static compliance is 1 / k, which represents the steady-state compliant displacement caused by a unit contact force, directly corresponding to the "softness or hardness" felt by the operator, and is determined by the stiffness parameter k. The stiffness k is determined according to the desired contact softness or hardness (or the amount of compliance under the desired maximum contact force).

[0073] 2. Virtual Mass and Response Characteristics. The virtual mass *m* determines the speed of the end effector's response to changes in contact force and its "virtual inertia," as well as the undamped natural frequency of the admittance model. for:

[0074] Choose an appropriate virtual mass m according to the desired response speed. If m is too small, the response will be sensitive but easily affected by noise. If m is too large, the motion will be smoother but less responsive.

[0075] 3. Damping is tuned according to critical damping. After determining m and k, the damping c is tuned according to the critical damping relationship. Critical damping for:

[0076] Let the damping ratio ζ represent the ratio of actual damping to critical damping, then we have:

[0077] The present invention uses a damping ratio close to the critical damping (approximately 0.7 to 1.0): at this point, the compliant motion has no obvious overshoot and can return to position relatively quickly and smoothly, which is a compromise between smoothness and responsiveness.

[0078] This forms a unified tuning process: first, determine the stiffness k based on the desired contact hardness; then, select a suitable virtual mass m; finally, determine the damping c according to the critical damping relationship (taking a damping ratio of approximately 0.7 to 1.0). The upper limit of the safety envelope can be determined consistent with the stiffness and the desired maximum contact force.

[0079] To make the upper limit of the compliant displacement compatible with the stiffness setting.

[0080] in, Indicates the maximum external contact force allowed by the design or expected by the task; represents the maximum permissible compliant displacement (i.e., the upper limit of the safety envelope); k represents the virtual stiffness parameter for the corresponding degree of freedom. Therefore, the above formula indicates that the upper limit of the compliant displacement can be approximately determined based on the ratio of the maximum external contact force to the virtual stiffness.

[0081] Based on the above tuning results, you can optionally search for fine-tuning within a small range using offline evaluation indicators (contact force overshoot, return time, steady-state contact force, end jitter) to further improve the smoothness of the hand under a given task.

[0082] Another key technical detail of this invention is that different robotic arms can achieve a compliant feel using the same tuning principle, without needing to rebuild the body dynamics model for each model. The mechanism is as follows: this invention achieves compliance in a Cartesian task space, and treats the robotic arm as an end-effector pose tracking device through a motion solver based on optimization solutions; the compliant feel is determined by the virtual mass-damping-stiffness of the task space, and model-specific kinematic and execution differences are isolated in the adapter and solver backend and absorbed by the pose tracking loop.

[0083] This contrasts with impedance / force control implemented at the joint torque level: the latter requires online compensation of the robotic arm's own mass matrix, Coriolis terms, and friction-related quantities, and the control parameters are strongly tied to the specific model, necessitating remodeling and retuning when changing models. In contrast, this invention… 1. Compliance dynamics is defined in the Cartesian space at the end point. Stiffness is determined according to static compliance, virtual mass is selected according to response characteristics, and damping is tuned according to critical damping relationship. The entire tuning principle does not contain model-related dynamic terms. 2. Once the parameters are adjusted according to the above principle on the same robotic arm, the same set of admittance parameters can be used across different tasks and deployments of the robotic arm without having to readjust them one by one. 3. When switching to different robotic arms, the same tuning principle (selecting a suitable virtual mass, determining damping according to the critical damping relationship, and determining stiffness according to static compliance) can be used to quickly obtain a similar smooth feel, which is significantly simpler than remodeling and readjusting control parameters for each model.

[0084] Therefore, this invention reduces the tuning work for cross-model deployment from "re-establishing dynamic models and readjusting control parameters for each model" to "using a unified principle, only replacing the adapter and tuning according to a small number of physical quantities", significantly reducing the cost of cross-platform deployment.

[0085] Example 2 The operator wears an exoskeleton or operates the master end device. The exoskeleton collects the operator's joint states, and the reference pose of the slave end robotic arm is obtained through positive kinematics calculations. The control is then sent to the slave robotic arm (e.g., the RealMan six-DOF robotic arm) according to a control cycle. A six-dimensional force-torque sensor connected to the end effector tool is installed at the end of the slave robotic arm, and the motion solver uses a model predictive control solver. The robotic arm strictly tracks the operator's reference pose in free space; when the end effector contacts the workpiece or environment, the external force measured by the sensor is input into the admittance model after gravity compensation, coordinate transformation, and force dead zone processing, generating a compliant displacement relative to the reference pose. This displacement is then superimposed on the reference pose under the constraint of a safety envelope (e.g., the upper limit of the maximum compliant displacement). The operator thus perceives compliant contact, with the contact force limited within a safe range, and the compliant action does not interfere with the continuously sent reference motion from the exoskeleton teleoperation.

[0086] The operator guides the robotic arm's end effector to continuously press against a fixed obstacle. Once the compliant displacement output by the admittance model reaches the upper limit of the safe envelope, it is scaled proportionally. Simultaneously, the internal displacement and velocity states of the admittance model are reset to prevent continuous integration of the internal states. After the operator stops pressing, due to the reset internal states and the presence of the stiffness term, the end effector smoothly returns to the vicinity of the reference pose, rather than stagnating at the workspace boundary.

[0087] The same compliant control loop is deployed on different models of robotic arms. Only an adapter satisfying the robot client interface needs to be implemented for each robotic arm, and an adapter satisfying the motion solver interface needs to be implemented for each solution backend; the compliant control algorithm itself does not need modification. The solution backend can switch between different optimization-based motion solvers while maintaining consistent compliant control performance.

[0088] For a specific contact operation, parameters are first set according to a unified tuning principle: stiffness k is determined based on the desired contact hardness, a suitable virtual mass m is selected based on the desired response characteristics, and damping c is determined by a damping ratio of approximately 0.7 to 1.0 based on the critical damping relationship. The upper limit of the safety envelope is then determined by ensuring that the stiffness is consistent with the desired maximum contact force. After tuning once on a certain model of robotic arm according to this principle, the same set of parameters can be used across different tasks and deployments of the same robotic arm without repeated readjustment. When switching to another model of robotic arm, the same tuning principle (selecting a suitable virtual mass, determining damping based on critical damping, and determining stiffness based on static compliance) can be used to quickly obtain a similar compliant feel. Only the adapter needs to be replaced and the force sensor calibration needs to be checked; there is no need to rebuild the body dynamics model.

[0089] Preferably, the admittance model can be selected between 3D translational compliance and 6D pose compliance; the 6D form integrates translation and rotation according to the mass-spring-damping relationship, respectively. The force reference coordinate system can be switched between base coordinate system mode and tool coordinate system mode; gravity compensation can be configured or disabled according to the tool mass and mounting posture. The motion solver can be replaced between different optimization-based implementations (e.g., model predictive control solvers); the robotic arm body can be replaced between different manufacturers, requiring only the corresponding adapter to be replaced. Reference pose Input can be obtained via exoskeleton teleoperation, or from preset trajectories, visually guided trajectories, or upper-level planning. Force dead zone threshold, safety envelope upper limit, mass, damping, stiffness parameters, and control cycle can all be configured according to the task and the robotic arm; the safety envelope can be extended from a fixed upper limit to a dynamic upper limit that varies with the workspace or task stage. Joint filtering can be selected between first-order and second-order low-pass filters, and the cutoff frequency is configurable; similar filtering can also be applied to force signals.

[0090] like Figure 1 This is a general block diagram of compliant control for a robotic arm in exoskeleton teleoperation. It illustrates the relationship between the exoskeleton teleoperation reference pose channel, the end effector force compliance channel, pose overlay, motion solver, and robotic arm execution.

[0091] like Figure 2 This is a single-cycle flowchart for compliant control. It shows the cyclic steps of reading joints and filtering, reading TCP, reading and processing external forces, admittance integration, safety envelope, pose superposition, solving joints, and issuing execution.

[0092] like Figure 3 This is a block diagram of the external force handling and admittance model. It illustrates the process of sensor readings, gravity compensation, coordinate transformation, force dead zone, mass-spring-damping integration to adaptive displacement.

[0093] like Figure 4 (a) and (b) in the diagram are schematic diagrams of the compliant displacement safety envelope and anti-integral saturation, respectively. They illustrate the scaling, internal state reset, and return to position after release when the compliant displacement exceeds the upper limit.

[0094] like Figure 5 (a) and (b) in the figure are schematic diagrams of the spatial superposition of the reference pose and the compliant displacement pose, respectively. They show that the end effector tracks the reference pose when there is no external force, and the end effector generates a compliant displacement based on the reference pose when there is a contact force.

[0095] like Figure 6 This is a diagram illustrating the decoupled architecture between the robot body and the solution backend. It demonstrates how the compliant control loop interacts with specific robotic arm adapters and optimization-based motion solver (e.g., model predictive control) adapters through the robot client interface and motion solver interface.

[0096] like Figure 7 This diagram illustrates the principle of unified admittance parameter tuning and cross-model application. It demonstrates the mechanism of tuning admittance parameters according to a unified principle (stiffness based on static compliance, virtual mass based on response characteristics, and damping based on critical damping relationship), as well as the application of parameters on the same robotic arm and the rapid tuning on different robotic arms to achieve a similar compliant feel.

[0097] In this invention, the reference pose is mainly derived from the joint state of the exoskeleton teleoperation, which is calculated by positive kinematics to obtain the TCP pose of the robot end effector. It can also be given by the upper-level motion planning trajectory.

[0098] Force-torque sensor: A six-dimensional force-torque sensor is used, which is installed at the end of the robotic arm and connected to the end tool; force and torque are expressed in the sensor coordinate system; the installation direction is perpendicular to the end tool direction; gravity compensation uses the weight of the end tool as input.

[0099] Regarding the six-dimensional admittance for rotational degrees of freedom: Admittance-type compliant control is quite common in the field of robotics, therefore, six-dimensional admittance is not presented as an independent subordinate solution. The core of this invention lies in how to apply non-conflicting compliant control to the exoskeleton under the condition of continuous reference motion given by upper-level commands. Six-dimensional admittance is only considered as an optional implementation variation.

[0100] The values ​​for safety envelope upper limit, mass, damping, stiffness, force dead zone threshold, control cycle, etc. are all parameters of the implementation example and are adjusted according to the specific task and robotic arm.

[0101] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0102] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A compliant control system for a robotic arm used in teleoperation of exoskeletons, characterized in that, include: The reference pose acquisition module acquires the reference pose given by the exoskeleton teleoperation. An external force processing module has its input end connected to a force-torque sensor installed at the end of the robotic arm. It receives the external force measured by the sensor and performs gravity compensation, coordinate transformation, and force dead zone processing on the external force. The admittance model module has its input end connected to the output end of the external force processing module. It performs integral calculation based on the processed external force and outputs the compliant displacement relative to the reference pose. A safety constraint module, whose input is connected to the output of the admittance model module, applies a safety envelope constraint to the compliant displacement, limiting the upper limit of the amplitude of the compliant displacement; The pose overlay module has a first input terminal connected to the output terminal of the reference pose acquisition module and a second input terminal connected to the output terminal of the safety constraint module. It overlays the compliant displacement and the reference pose in the pose space to generate the end-effector target pose. The motion solver interface module has its input end connected to the output end of the pose superposition module, and outputs the end-effector target pose to the motion solver based on optimization solution to obtain the joint target of the robotic arm; The robot client interface module has its input end connected to the output end of the motion solver interface module, and its output end connected to the robot arm body, so as to send the joint target to the robot arm body for execution. The admittance model module and the pose superposition module constitute a compliant control loop. The compliant control loop interacts with the robotic arm body through the robot client interface module and with the motion solver through the motion solver interface module. The admittance model module adopts a second-order mass-spring-damped admittance model. For each translational degree of freedom, the dynamic relationship of the admittance model module satisfies: Where m is the virtual mass parameter, c is the virtual damping parameter, k is the virtual stiffness parameter, s is the compliant displacement relative to the reference pose, v is the compliant velocity, a is the compliant acceleration, and f is the external force component acting on this degree of freedom. The admittance model module performs discrete integration on the dynamic relationship in each control cycle, and expands it into a six-dimensional form when attitude compliance is required. The three translational degrees of freedom and the three rotational degrees of freedom are integrated according to the same mass-spring-damping relationship to obtain the six-dimensional compliant pose. The safety constraint module applies a safety envelope constraint to the compliant displacement in the following ways: Calculate the modulus of the compliant displacement; When the magnitude of the compliant displacement is less than or equal to the upper limit of the safety envelope, the compliant displacement remains unchanged; When the magnitude of the compliant displacement is greater than the upper limit of the safety envelope, the compliant displacement is scaled by a scaling factor so that the magnitude of the compliant displacement is equal to the upper limit of the safety envelope. When scaling is triggered, the safety constraint module also synchronously resets the internal state of the admittance model module, including setting the internal compliance displacement state to the scaled compliance displacement and attenuating the internal compliance velocity state by the same scaling factor.

2. The compliant control system for exoskeleton teleoperation according to claim 1, characterized in that, When the external force processing module performs the gravity compensation, the weight of the end tool is used as the input; when the external force processing module performs the coordinate transformation, it supports both base coordinate system mode and tool coordinate system mode. In the base coordinate system mode, the external force processing module estimates the residual component of the tool gravity at the end of the robotic arm based on the change of the current posture of the end relative to the initial posture, subtracts it, and then rotates the force after subtracting the residual component to the base coordinate system. In the tool coordinate system mode, the external force processing module rotates the sensor readings to the end tool coordinate system according to the fixed rotation relationship between the sensor and the end tool.

3. The compliant control system for exoskeleton teleoperation according to claim 1, characterized in that, The external force processing module applies the force dead zone processing in the following ways: Calculate the magnitude of the external contact force; When the magnitude of the external contact force is less than or equal to the force dead zone threshold, the external contact force is set to zero. When the magnitude of the external contact force is greater than the force dead zone threshold, the external contact force is kept at its original value and input into the admittance model module.

4. The compliant control system for exoskeleton teleoperation according to claim 1, characterized in that, The virtual mass parameters, virtual damping parameters, and virtual stiffness parameters of the admittance model module are given according to a unified tuning principle: The virtual stiffness parameters are determined according to the desired static compliance. The virtual quality parameters are selected according to the desired response characteristics; The virtual damping parameters are determined according to the critical damping relationship, satisfying: Where c is the virtual damping parameter, Where is the damping ratio, m is the virtual mass parameter, and k is the virtual stiffness parameter; The upper limit of the safety envelope is determined in conjunction with the virtual stiffness parameter and the maximum external contact force allowed by the design, satisfying: in, Indicates the maximum external contact force allowed by the design or expected by the task. This indicates the maximum allowable compliance displacement.

5. The compliant control system for exoskeleton teleoperation according to claim 1, characterized in that, In the pose overlay module, the adaptive displacement and the reference pose are overlaid in the pose space in the following way: A pure translation transformation is constructed using the compliant displacement. This pure translation transformation is then superimposed on the reference pose to obtain the end-effector target pose, satisfying the following: in, The target pose is executed for the end effector. Indicates to adapt to displacement Constructed pure translation transformation, For reference pose; The pose overlay module also converts the end-effector target pose into a seven-dimensional end-effector target pose consisting of a position plus an attitude quaternion, which is then used as input to the motion solver.

6. The compliant control system for exoskeleton teleoperation according to claim 1, characterized in that, It also includes a joint signal processing module, which performs low-pass filtering on the read joint states and uses the filtered joint states as the joint seeds of the motion solver; The low-pass filter employs a first-order or second-order low-pass filter based on bilinear transformation. The control system also includes a solution failure handling module. When the motion solver returns an empty solution or an invalid solution containing non-finite values, the solution failure handling module determines that the solution for this cycle is invalid and maintains the current pose of the robotic arm. The control system also includes an anomaly rollback module. When an anomaly is detected in the joint target sent to the robotic arm body, the anomaly rollback module rolls back to the current joint state of the robotic arm body.

7. The compliant control system for exoskeleton teleoperation according to claim 1, characterized in that, The reference pose obtained by the reference pose acquisition module is obtained by calculating the TCP pose of the end effector of the robotic arm through positive kinematics of the joint state of the exoskeleton teleoperation, or by the upper-level motion planning trajectory. The motion solver interface module is connected to a model predictive control solver.

8. A compliant control method for a robotic arm in exoskeleton teleoperation, characterized in that, The robotic arm compliant control system for exoskeleton teleoperation, as described in any one of claims 1 to 7, is used to perform the following steps in a fixed control cycle: Step 1: Read the joint status of the robotic arm and perform low-pass filtering; Step 2: Read the end effector TCP pose of the robotic arm; Step 3: Read the external force measured by the force-torque sensor installed at the end of the robotic arm; Step 4: Perform gravity compensation, coordinate transformation, and force dead zone processing on the external force; Step 5: Input the processed external force into the second-order mass-spring-damped admittance model for integration to obtain the compliant displacement relative to the reference pose given by the exoskeleton teleoperation; Step 6: Apply a safety envelope constraint to the compliant displacement; Step 7: Superimpose the compliant displacement with the reference pose in the pose space to obtain the end effector target pose; Step 8: Output the end-effector target pose to the optimization-based motion solver via the motion solver interface to obtain the joint target; Step 9: Send the joint target to the robotic arm body for execution via the robot client interface.

Images