Teleoperation control method and device of body-equipped robot and intelligent device

Abstract

This application relates to the field of robot teleoperation control, and in particular to a teleoperation control method, apparatus, and intelligent device for an embodied robot. It aims to solve the problems of lagging control data, asynchronous actions, and poor stability caused by communication delays, packet loss, or discontinuous state updates in existing master-slave teleoperation. This method is applied to a slave-end embodied robot. The slave receives control commands generated by the exoskeleton glove based on a fused state quantity formed by the actual state quantities of each joint and the predicted state quantities obtained under preset prediction conditions, combining the current motion state and historical motion states. The fused state quantities of each joint are parsed to obtain the target fused state quantities, and the corresponding joints are controlled for tracking accordingly. By introducing fused state quantities as the control basis, the continuity of actions, master-slave consistency, and joint tracking accuracy can be improved, enhancing control stability and teleoperation reliability under communication anomalies.

Description

Technical Field

[0001] This application relates to the field of robot teleoperation control, and more particularly to a teleoperation control method, apparatus and intelligent device for an embodied robot. Background Technology

[0002] Exoskeleton gloves, which combine sensor data acquisition and master-slave teleoperation control, have been widely used for remote precision operations in industrial, medical, and high-risk environments.

[0003] However, existing systems often rely on fixed structures and single sensing methods, making them susceptible to asynchrony between master and slave actions when communication links experience delays or packet loss. Furthermore, the sensed information is easily affected by environmental interference, leading to trajectory distortion and making it difficult to simultaneously maintain operational continuity, accuracy, and stability in complex scenarios.

[0004] Therefore, how to improve the synchronization, control accuracy and operational stability of the remote operating system under conditions of communication anomalies and sensor fluctuations has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides a remote control method, apparatus, and intelligent device for an android to solve the aforementioned technical problems. The method is geared towards a slave android, receiving, parsing, and performing joint-level tracking control on control information output from its associated exoskeleton glove. This enables the joints of the slave android to execute actions based on control results that consider both the actual state and the trend of motion changes, thereby improving the consistency of actions, control accuracy, and operational stability during master-slave teleoperation in the event of communication anomalies or sensory fluctuations.

[0006] In a first aspect, this application provides a remote control method for an android, applied to a slave android, comprising:

[0007] Receive a first control command generated by the exoskeleton glove based on fused state quantities, wherein the fused state quantities are obtained by combining the predicted state quantities determined based on the current motion state of each joint of the exoskeleton glove and the corresponding historical motion state at the previous moment with the actual state quantities.

[0008] Parse the first control command to obtain the target fusion state variables corresponding to each joint;

[0009] Based on the target fusion state quantities, the corresponding joints of the embodied robot are controlled to perform tracking operations.

[0010] Secondly, this application provides a remote control method for an android, applied to an exoskeleton glove, the method comprising:

[0011] The current motion state of each joint is obtained through preset sensors at the pivot point of the exoskeleton glove;

[0012] When the preset prediction conditions are met, the predicted state quantity of each joint is determined based on each current motion state and the corresponding historical motion state at the previous moment.

[0013] The fusion state quantities are determined based on the actual state quantities of each joint and the corresponding predicted state quantities.

[0014] A first control command is generated based on the fused state variables and sent to the slave embodied robot to control the corresponding joint of the slave embodied robot to perform tracking based on the fused state variables.

[0015] In one possible implementation, the acquisition of the current motion state of each joint includes:

[0016] The current absolute angle of each joint is collected by the preset sensor to obtain the joint angle of each joint;

[0017] Perform a first-order difference operation on the joint angle and the corresponding adjacent historical joint angle to obtain the current angular velocity;

[0018] The current angular velocity is obtained by performing a first-order difference operation between the current angular velocity and the corresponding adjacent historical angular velocity.

[0019] In one possible implementation, determining the predicted state quantities of each joint based on each current motion state and the corresponding historical motion state at the previous moment includes:

[0020] For each joint, the predicted joint angle is determined based on the historical joint angle, historical angular velocity, historical angular acceleration, and predicted time step from the previous moment.

[0021] The predicted angular velocity is determined based on the historical angular velocity, angular acceleration, and the predicted time step.

[0022] In one possible implementation, the fusion state quantity is a fusion joint angle or a fusion angular velocity, and the determination of the fusion state quantity based on the actual state quantity of each joint and the corresponding predicted state quantity includes:

[0023] The weighting coefficient is determined based on the ratio of the current angular acceleration to the preset proportional parameter, wherein the larger the ratio, the smaller the corresponding weighting coefficient.

[0024] For each joint, a first fused state quantity is determined by multiplying the predicted state quantity of the joint with the weighting coefficient.

[0025] The second fusion state quantity is determined by multiplying the actual state quantity of the joint with the product of 1 and the difference between the weighting coefficient;

[0026] The fusion state quantity is determined based on the sum of the first fusion state quantity and the second fusion state quantity.

[0027] In one possible implementation, the method further includes:

[0028] The operator's intended joint torque is derived by inversely from the current angular acceleration of each joint, wherein the intended joint torque is positively correlated with the moment of inertia of the corresponding joint and the current angular acceleration;

[0029] The joint torque intention is introduced as a feedforward parameter into the force control loop of the slave-end android, and superimposed with the actual contact force fed back by the force sensor of the exoskeleton glove to generate a second control command that includes position tracking and force feedforward compensation.

[0030] The second control command is sent to the slave robot so that the slave robot can respond in advance to the operator's force intention and adjust the clamping force at the end of the actuator of the slave robot.

[0031] In one possible implementation, the preset prediction conditions include at least one of the following:

[0032] The communication delay between the exoskeleton glove and the slave-end android exceeds a preset delay threshold.

[0033] Alternatively, the safety threshold of the embody robot may be reduced.

[0034] In one possible implementation, the finger linkage of the exoskeleton glove adopts a carbon fiber nested telescopic structure, the preset sensor is a dual-axis magnetic encoder, and the method further includes:

[0035] The length of the finger link and the rotation angle of the pivot are adjusted based on the fusion state quantity so that the finger link is adapted to the physiological state of the operator's fingers.

[0036] Thirdly, this application provides a remote control device for an android, applied to a slave android, comprising:

[0037] The receiving module is used to receive a first control command generated by the exoskeleton glove based on the fusion state quantity, wherein the fusion state quantity is a predicted state quantity determined based on the current motion state of each joint of the exoskeleton glove and the corresponding historical motion state at the previous moment, and is obtained from the actual state quantity.

[0038] The parsing module is used to parse the first control command and obtain the target fusion state quantities corresponding to each joint;

[0039] The execution module is used to control the corresponding joints of the embodied robot to perform tracking operations based on the fusion state variables of each target.

[0040] Fourthly, this application provides a remote control device for a bodysuit robot, applied to an exoskeleton glove, comprising:

[0041] The acquisition module is used to acquire the current motion state of each joint through preset sensors at the pivot of the exoskeleton glove;

[0042] The calculation module is used to determine the predicted state quantity of each joint based on each current motion state and the corresponding historical motion state at the previous moment when the preset prediction conditions are met.

[0043] The calculation module is also used to determine the fused state quantities based on the actual state quantities of each joint and the corresponding predicted state quantities;

[0044] The control module is used to generate a first control command based on the fused state quantity and send it to the slave embodied robot to control the corresponding joint of the slave embodied robot to perform tracking based on the fused state quantity.

[0045] Fifthly, this application provides an intelligent device, including: a processor, and a memory communicatively connected to the processor;

[0046] The memory stores computer-executed instructions;

[0047] The processor executes computer execution instructions stored in the memory to implement the remote control method for the embodied robot as described above.

[0048] Sixthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a computer, are used to implement the telescopic control method for the embodied robot as described above.

[0049] In a seventh aspect, this application also provides a computer program product, including a computer program that, when executed by a processor, can implement the steps of the scheme recommendation method as described in any of the preceding claims.

[0050] This application provides a remote control method, device, and intelligent device for an android. By receiving a first control command generated by the exoskeleton glove based on fused state quantities, and parsing the first control command to obtain the target fused state quantities corresponding to each joint, the android can simultaneously refer to the actual state quantities of each joint of the exoskeleton glove and the predicted state quantities determined when preset prediction conditions are met during joint control. Thus, even with communication delays, packet loss, or fluctuations in perception information, it can still maintain good motion synchronization, trajectory tracking accuracy, and control continuity, thereby improving the overall stability of the remote operating system. Attached Figure Description

[0051] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0052] Figure 1 A flowchart illustrating a remote control method for an embodied robot provided in this application embodiment. Figure 1 ;

[0053] Figure 1a This is a schematic diagram of the structure of an exoskeleton glove provided in an embodiment of this application;

[0054] Figure 2 A flowchart illustrating a remote control method for an embodied robot provided in this application embodiment. Figure 2 ;

[0055] Figure 3 A schematic diagram of the structure of a remote control device for an embodied robot provided in this application embodiment. Figure 1 ;

[0056] Figure 4 A schematic diagram of the structure of a remote control device for an embodied robot provided in this application embodiment. Figure 2 ;

[0057] Figure 5 This is a schematic diagram of the structure of a smart device provided in an embodiment of this application.

[0058] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0059] Teleoperation control technology for androids is mainly applied in scenarios such as industrial manufacturing, medical assistance, high-risk environment operations, and remote precision operations. Its core lies in acquiring the operator's hand movement information through a wearable data acquisition device at the master end and transmitting this information to the slave android, enabling the slave joints to perform synchronized movements according to the operator's intentions. In actual deployment, the system typically consists of an exoskeleton glove, a communication link, the slave android, and a control processing unit. The exoskeleton glove is responsible for acquiring the movement status of the operator's hand joints, the communication link is responsible for transmitting control commands between the master and slave ends, and the slave android drives the corresponding joints to perform actions such as grasping, pinching, rotating, assembling, or contacting based on the received commands.

[0060] For example, in industrial assembly scenarios, operators can remain in a safe area and remotely control the end-of-line robotic arm through exoskeleton gloves to complete precision assembly, hazardous material handling, and maintenance in confined spaces. In medical settings, doctors can use teleoperation to drive mechanical actuators to perform minimally invasive interventions, tissue retrieval, and precise positioning. In deep-sea, nuclear environments, or disaster relief scenarios, operators can also use such systems to grasp targets and manipulate equipment in isolated environments. Because these applications generally have high requirements for motion continuity, joint tracking accuracy, and system stability, the quality of motion state acquisition by the exoskeleton gloves, the consistency of master-slave control commands, and the tracking ability of the slave robot to the target state directly affect whether the entire teleoperation task can be completed safely, accurately, and continuously. Especially when the work object is fragile, the operating space is confined, or motion switching is frequent, any control deviation caused by communication fluctuations, state distortion, or command asynchrony can lead to grasping failure, trajectory deviation, assembly errors, or even safety risks. Therefore, this type of technology has long been an important research direction in the field of high-end robot control.

[0061] Existing teleoperation solutions that link exoskeleton gloves with androids typically employ a master-slave control model to map operator movements to slave-side execution actions. The master exoskeleton glove collects finger joint angles, bending degrees, or posture changes through various sensors, encodes these actual state quantities into control commands, and transmits them to the slave android controller via wired or wireless networks. The slave android controller parses the received data, extracts the target values ​​for each joint, and drives the corresponding joints of the android to follow the movement. The basic working principle of this type of solution is to use the joint state currently detected by the master as the direct basis for slave-side execution, enabling the slave's mechanical structure to reproduce the operator's hand movements as closely as possible.

[0062] However, in practical applications, such background technologies typically rely on ideal conditions: stable communication links, continuous sensor data, and real-time delivery of state changes. Once the link experiences latency jitter, momentary congestion, data loss, or local anomalies, the actual state output from the master end cannot be completely and continuously transmitted to the slave end, resulting in delayed, missing, or abrupt control data obtained by the slave end. This problem is further amplified in scenarios requiring high-frequency motion tracking, as the slave-end embodied robot often performs joint control based on received discrete state points. When the state at certain critical moments is not received in time, the slave end can only rely on outdated data for action, leading to motion tailing, grasping misalignment, or sudden posture jumps. Simultaneously, existing solutions typically use the actual acquired state directly as the control basis, lacking a mechanism to compensate for future joint change trends when preset prediction conditions are met, thus lacking buffering capacity against sudden communication anomalies. Especially in complex operations, the operator's hand movements have continuous and inertial characteristics; simply relying on the currently successfully uploaded actual state quantity makes it difficult to reflect the short-term motion evolution trend, causing intermittent joint control at the slave end. Furthermore, traditional solutions do not adequately consider the consistency of state between sending and parsing instructions. Even if the master end can temporarily adopt a local correction strategy, if the slave end still executes only according to a single actual state, local asynchrony can easily occur at the joint level. Therefore, while existing technologies can achieve basic teleoperation under ideal network and normal operating conditions, in real-world environments where communication anomalies and sensory fluctuations coexist, it is still difficult to simultaneously ensure motion synchronization, control accuracy, and operational stability, especially to meet the requirements of highly reliable and highly continuous remote operation of embodied robots.

[0063] In light of this, ensuring that the slave android continuously receives effective control data reflecting the operator's intentions during master-slave teleoperation, even in the face of communication delays, data loss, or discontinuous state updates, has become a pressing technical problem. To address this issue, a teleoperation control method for androids is provided, applied to a slave android. In this application scenario, the system still operates based on a master-slave collaborative architecture. The master end, an exoskeleton glove associated with the slave android, handles motion state acquisition and control command generation. The slave android receives and executes corresponding joint control tasks, and the two exchange control information via a communication link. Unlike methods that directly control the slave end based solely on actual state, this method first receives a first control command sent by the exoskeleton glove. This first control command is not solely formed from actual state quantities but is generated from fused state quantities. The fused state quantities are determined jointly based on the actual state quantities of each joint of the exoskeleton glove and the predicted state quantities determined according to the current motion state of each joint and the corresponding historical motion state at the previous moment, when preset prediction conditions are met. Subsequently, the slave-end embodied robot parses the first control command to obtain the target fused state variables corresponding to each joint, and controls the corresponding joints of the embodied robot to perform tracking operations based on each target fused state variable. By introducing fused state variables as an intermediate control basis in the control link, the slave-end embodied robot can still perform joint tracking based on more continuous and stable target fused state variables even when there are fluctuations in the master-end state updates. This improves the consistency of master-slave actions, enhances control accuracy and operational stability under communication anomalies, and provides a more reliable control foundation for fine teleoperation in complex scenarios.

[0064] Figure 1 A flowchart illustrating a remote control method for an embodied robot provided in this application embodiment. Figure 1 .like Figure 1 As shown, the method includes:

[0065] S101. Obtain the current motion state of each joint through the preset sensor at the pivot of the exoskeleton glove.

[0066] The execution entity for this step can be a data acquisition and control unit located within the exoskeleton glove itself. This control unit can include a microcontroller, embedded processor, edge computing module, and data acquisition interfaces connected to various sensors. The exoskeleton glove can be constructed as a wearable mechanical structure corresponding to each finger joint of the operator's hand. A pivot is set at the rotational connection point of each finger joint, and preset sensors for characterizing joint motion information are arranged at these pivots. The current motion state refers to the set of states characterizing joint motion directly acquired by sensors or obtained through basic calculations at the current sampling moment. Preset sensors can include a dual-axis magnetic encoder, a linear displacement sensor, an inertial measurement unit, a tactile sensor, and a pressure sensor. The dual-axis magnetic encoder acquires absolute joint angle information, the linear displacement sensor acquires extension / retraction length information related to the glove's linkage or flexible transmission structure, the inertial measurement unit acquires attitude angular velocity and linear acceleration, and the tactile and pressure sensors characterize contact state and grip strength. Technically, a pivot refers to the mechanical connection center in the exoskeleton glove used to achieve relative rotation between adjacent finger joints, and preset sensors refer to state detection elements that are pre-installed and have undergone position calibration, zero-point calibration, and sampling parameter setting.

[0067] In one possible implementation, because the length of the finger links and the rotation angle of the shaft in the exoskeleton glove are fixed, they cannot be dynamically adapted to the operator's actual finger size and movement posture. This results in poor fit and low motion transmission accuracy when worn by different operators, and prolonged use can easily lead to fatigue and even joint damage. Therefore, to improve the fit and motion transmission accuracy of the exoskeleton glove and reduce operator fatigue, this application introduces a carbon fiber nested telescopic structure finger link and a dual-axis magnetic encoder. Based on fused state variables, the length of the finger link and the rotation angle of the shaft are adjusted in real time to dynamically adapt the finger link to the physiological state of the operator's fingers. The specific implementation method is as follows:

[0068] The finger links of the exoskeleton glove adopt a carbon fiber nested telescopic structure, and the preset sensor is a dual-axis magnetic encoder.

[0069] In this embodiment, the carbon fiber nested telescopic structure can be composed of an outer connecting rod, an inner sliding section, and a limiting retainer. The outer connecting rod provides basic load-bearing stiffness, while the inner sliding section expands and contracts along the connecting rod axis to accommodate different finger lengths and flexion / extension ranges while maintaining overall lightweight design. Carbon fiber material has high specific strength and fatigue resistance, reducing the wear load on the glove and improving structural stability under repetitive movements. In practical applications, other models of this component can also be selected, and this embodiment does not limit this. A dual-axis magnetic encoder is installed at the finger pivot point; one axis detects the flexion / extension angle, and the other detects the yaw angle, thereby outputting an angular position signal corresponding to the finger posture. This type of sensor features non-contact measurement, wear resistance, and stable response, making it suitable for wearable applications involving frequent bending.

[0070] Furthermore, when adjusting the finger link length and pivot rotation angle of the exoskeleton glove, the target end-effector position can be calculated based on a forward kinematics algorithm. Specifically, based on the current joint angle, link length, and target end-effector position, the exoskeleton glove derives the adjustment requirements for link length and pivot angle using a forward kinematics model (such as the Denavit-Hartenberg parametric method). Through an iterative optimization strategy that minimizes end-effector position error, the link length and pivot rotation angle are dynamically adjusted to ensure that the end-effector position of the exoskeleton glove aligns with the natural bending trajectory of the operator's fingers. This process is achieved through the collaborative action of a nested carbon fiber link telescopic mechanism and micro-drive components (such as stepper motors or linear actuators), ensuring the real-time performance and accuracy of the adjustment process.

[0071] Figure 1a This is a schematic diagram of the structure of an exoskeleton glove provided in an embodiment of this application. Figure 1a As shown, the exoskeleton glove includes a palm base, multiple sets of finger links, and multiple pivot joints. The finger links adopt a carbon fiber nested telescopic structure, which consists of an outer link, an inner sliding section, and a limiting retainer. The outer link provides basic load-bearing stiffness, while the inner sliding section expands and contracts along the link axis to accommodate different finger lengths and flexion / extension ranges while maintaining overall lightweight design. Each finger link is sequentially hinged to the palm base and adjacent links via pivot joints, forming a kinematic chain corresponding to the joints of the human finger. Each pivot joint is equipped with a dual-axis magnetic encoder as a preset sensor. One axis detects the flexion / extension angle, and the other axis detects the yaw angle, thereby outputting an angular position signal corresponding to the finger posture. The current motion state of each joint is obtained through this dual-axis magnetic encoder, and the length of the finger link and the rotation angle of the pivot are adjusted in real time based on the fused state variables. This allows the finger links to dynamically adapt to the physiological state of the operator's fingers, thereby improving the wearability and motion transmission accuracy of the exoskeleton glove and reducing operator fatigue.

[0072] In practical implementation, the exoskeleton glove first performs an initialization process after startup. The acquisition and control unit reads the zero-point data and calibration parameters of each sensor, establishing a mapping relationship between sensor numbers and target joint numbers. For example, the proximal joint of the thumb, the proximal joint of the index finger, and the middle joint of the middle finger are each mapped to an independent sampling channel. In one possible embodiment, the dual-axis magnetic encoder outputs the raw angle value at a sampling frequency of not less than 100 Hz, and the linear displacement sensor synchronously outputs the displacement value. The acquisition and control unit performs noise reduction, temperature drift compensation, and outlier removal on the raw values. Noise reduction can be achieved using sliding mean filtering, median filtering, or first-order low-pass filtering; temperature drift compensation can be corrected based on a pre-stored temperature-offset lookup table; outlier removal can be accomplished through a combination of threshold and continuity detection. Subsequently, the acquisition and control unit determines the current motion state based on the data collected by the preset sensors in two adjacent sampling cycles.

[0073] In one possible implementation, since the preset sensors can typically only directly collect the current absolute angle of the joints, and the determination of the predicted state variables and the subsequent generation of fused state variables both require multi-dimensional motion state information such as angular velocity and angular acceleration as input, if control is based solely on the raw angle data, it cannot fully reflect the dynamic motion trend of the joints, resulting in insufficient prediction accuracy and lag of the fused state variables behind the actual operational intent, thereby affecting the tracking response speed and motion smoothness of the slave-end embodied robot. Therefore, in order to improve the completeness and prediction accuracy of joint motion state information and enhance the slave-end embodied robot's ability to track the operator's dynamic intent, this application uses the current absolute angle, current angular velocity, and current angular acceleration of each joint to constitute a complete current motion state, providing a sufficient data foundation for the subsequent determination of predicted state variables and the generation of fused state variables. The implementation process for obtaining the current motion state is as follows:

[0074] The current absolute angle of each joint is collected by preset sensors to obtain the joint angle of each joint; the first-order difference operation is performed on the joint angle and the corresponding adjacent historical joint angle to obtain the current angular velocity; the first-order difference operation is performed on the current angular velocity and the corresponding adjacent historical angular velocity to obtain the current angular acceleration.

[0075] Preset sensors are positioned near the joint axes of each joint in the exoskeleton glove to sense the rotational position of the finger or wrist joints in real time. The acquired absolute angles are used to characterize the instantaneous attitude information of each joint relative to a reference posture. Joint angles can be constructed using one or more of a magnetic encoder, photoelectric encoder, or inertial measurement unit. The sensor output signal is filtered and calibrated by the acquisition circuit before being converted into angle data to reduce measurement errors caused by zero bias, jitter, and mechanical backlash. For stable sampling, the sensors can be electrically connected to the control unit via flexible wiring and installed inside the joint housing using a fixed mount or embedding method. In practical applications, other models of this component can also be selected; this embodiment does not limit this choice.

[0076] After acquiring the current absolute angle, the data acquisition and control unit determines the joint angle of each joint at the current moment. It then performs a first-order difference operation between the joint angle of each joint and its corresponding adjacent historical joint angle, and combines this with the sampling period to calculate the current angular velocity, reflecting the speed and trend of each joint's movement over a short period. Subsequently, the current angular velocity is again subjected to a first-order difference operation with its corresponding adjacent historical angular velocity to obtain the current angular acceleration, which characterizes the acceleration, deceleration, and transient change intensity of each joint's movement. Historical joint angles and historical angular velocities are typically stored in a buffer or circular queue to ensure that data from adjacent moments can be continuously accessed. The difference results can be further smoothed before being output to the subsequent state fusion module, thus forming a more continuous motion description.

[0077] The above method allows for the step-by-step derivation of angular velocity and angular acceleration from an absolute perspective, enabling the current motion state to simultaneously encompass position, velocity, and dynamic change information. This state representation method more fully reflects the operator's hand gestures and is particularly suitable for teleoperation scenarios with high requirements for motion continuity. Since both angular velocity and angular acceleration are calculated based on adjacent historical data, even short-term sampling fluctuations provide a more stable foundation for subsequent prediction and fusion. This enhances the completeness and dynamic representation capability of joint motion state acquisition and provides a more accurate state basis for the continuous tracking control of the embodied robot.

[0078] In another possible implementation, the current motion state is not limited to a single scalar value, but can be constructed as a state vector, such as "current motion state = [joint angle, joint angular velocity, joint angular acceleration, link length change]". Using a state vector approach allows subsequent prediction and fusion processes to utilize position, velocity, and trend information simultaneously, rather than being limited to static angles. To ensure consistency of outputs from different sensors on the time axis, the acquisition and control unit can attach a unified timestamp to the sampled data from each channel and perform alignment processing on data from different sampling periods based on a time synchronization mechanism. For example, high-frequency sampled data can be resampled, and low-frequency data can be interpolated for compensation. This ensures the comparability of joint states at the same moment and avoids motion distortion caused by sampling phase differences.

[0079] Based on the above analysis, by directly deploying preset sensors at the pivot points of the exoskeleton glove and acquiring the current motion state of each joint, basic state data highly consistent with the operator's actual hand movements can be generated on the master-end exoskeleton glove. This step provides the initial basis for subsequent prediction of state variables and generation of fused state variables. Since the current motion state includes not only angles but also angular velocity, angular acceleration, and displacement changes, it can more fully characterize the continuity and transient changes of the movement, thus establishing a reliable data foundation for solving the problem of discontinuous control data at the slave end under network fluctuation conditions.

[0080] It should be noted that, to ensure reliable perception in complex environments, the exoskeleton gloves can further integrate tactile sensors and inertial measurement units (IMUs). The tactile sensors detect changes in the grip strength of the operator's fingers, while the IMU captures the overall hand movement trajectory. The control module dynamically fuses data from the magnetic encoder, linear displacement sensor, tactile sensor, and IMU using an adaptive weighting algorithm (such as Kalman filtering) to eliminate errors from individual sensors. For example, in environments with strong electromagnetic interference, magnetic encoder data may be distorted due to noise. In this case, the system dynamically adjusts the weighting coefficients using redundant data from the tactile sensor and IMU to ensure the accuracy of motion state calculations.

[0081] S102. When the preset prediction conditions are met, the predicted state quantities of each joint are determined based on each current motion state and the corresponding historical motion state at the previous moment.

[0082] In this step, the preset prediction conditions refer to the judgment rules that trigger the state prediction mechanism. Historical motion states refer to the state data of the same joint at the previous sampling time or several previous sampling times, which have been cached and saved. The predicted state quantity refers to the estimated joint state at the next control time or target transmission time, calculated based on the current motion state and historical motion states. The preset prediction conditions can be jointly determined by the acquisition and control unit, communication management unit, or edge computing unit. The judgment criteria can include situations such as communication latency exceeding a set threshold, control command acknowledgment delay exceeding limits, data packet loss rate exceeding the safety margin, sensor data continuity being impaired, master-slave clock asynchrony exceeding limits, or sudden changes in the state of certain joints that do not conform to physiological motion constraints. The caching method for historical motion states can be a circular queue, a time window buffer, or a joint-level state register structure. Each joint can save at least the state of the previous moment, or it can save a sequence of states from multiple consecutive moments.

[0083] In one possible implementation, considering that data transmission between the exoskeleton glove and the slave robot typically occurs via a wireless communication link, when the communication delay exceeds a preset delay threshold, control commands generated directly based on the current motion state may not reflect the operator's true intentions by the time they reach the slave robot, resulting in tracking lag and asynchronous movement. Furthermore, when the safety threshold of the slave robot decreases (e.g., when entering areas with dense human-robot collaboration or performing delicate tasks), conventional state feedback control lacks the ability to predict motion trends, easily leading to collision risks or decreased operational accuracy due to response delays. Therefore, to improve the real-time tracking performance and operational safety of the slave robot under scenarios with changing communication delays or safety constraints, this application introduces preset prediction conditions, specifically:

[0084] The preset prediction conditions include at least one of the following: the communication delay between the exoskeleton glove and the slave animate robot exceeds a preset delay threshold; or, the safety threshold of the slave animate robot is reduced.

[0085] Communication latency refers to the time interval between the exoskeleton glove outputting control information and the slave robot completing the parsing. The preset latency threshold can be pre-set based on the task's real-time requirements, joint response requirements, and link stability. When the link uses a high-speed wireless module or wired bus, this threshold can be set in the millisecond range; when used for precision assembly, the threshold can be further reduced to ensure motion synchronization.

[0086] The safety threshold of an end-effector robot characterizes its maximum withstand speed, displacement, torque, or contact force. This threshold can be dynamically adjusted by the controller based on the work scenario, end-effector status, obstacle distance, and human-robot collaboration safety strategies. When the system detects that the gripped object is fragile, the activity space is shrinking, or the proximity to the human body is decreasing, the safety threshold can be lowered, allowing the controller to enter a restricted control state earlier to reduce the risk of malfunctions and overload. When the safety threshold is lowered, the system will trigger a prediction mechanism based on the current joint status, enabling subsequent control commands to anticipate the movement trend.

[0087] Specifically, when the communication delay exceeds a preset delay threshold, it is determined that the master-slave state transmission has lagged behind, and predictive state quantity generation and fusion control are then activated to compensate for the delay gap in the transmitted state. When the safety threshold of the slave-end embodied robot decreases, the joint targets are pre-corrected based on more conservative control constraints, so that the slave end can still maintain continuous and stable tracking output when the risk tightens.

[0088] This preset prediction condition limits the prediction triggering time through two dimensions: communication anomalies and changes in safety constraints. This enables the slave device to maintain stable control even with incomplete information or enhanced constraints. As a result, the embodied robot can reduce motion tailing, abrupt posture changes, and overload impacts during teleoperation, improving response timeliness, execution safety, and control consistency in complex scenarios.

[0089] Furthermore, after each sampling cycle, the acquisition and control unit writes the processed current motion state into the historical cache, and before entering the next transmission cycle, the prediction trigger module determines whether the preset prediction conditions are met. For example, when the round-trip delay of the wireless network link is greater than 20 milliseconds, no confirmation is received from the slave end for two consecutive transmission cycles, or the packet loss rate exceeds 5% within a predetermined time window, it can be determined that the current communication environment is no longer sufficient to stably support the control mode that relies solely on real-time measured values, and the prediction mechanism is activated at this time.

[0090] After startup, the historical running state of each joint at the previous moment is read, and the predicted state quantity is determined based on the historical running state and the current motion state of each joint.

[0091] For example, if the current angle of the joint is denoted as The angle at the previous moment is recorded as The sampling period is denoted as Then the current angular velocity can be expressed as: If the angular velocity at the previous moment was Then the current angular acceleration can be expressed as: .

[0092] In one possible implementation, the predicted state variables are not limited to calculations based on analytical formulas; they can also be implemented using prediction models oriented towards time series. Edge computing units can pre-deploy lightweight recurrent neural networks, Transformer temporal prediction networks, or temporal convolutional networks. The current motion state and historical motion state sequences are input into the model, and the predicted state variables for the next moment or several subsequent moments are output, including predicted angle, predicted angular velocity, and predicted angular acceleration. For high-frequency repetitive actions, such as grasping, releasing, and pinching, prediction methods based on learning models can better fit nonlinear motion patterns. For slow, fine-grained operations, such as precision assembly, the ability to distinguish subtle motion trends in the prediction results can be improved by increasing the length of the short-term historical window. To avoid deviations between model predictions and reality, online corrections can be performed after each new, stable measured data is received, feeding the prediction error back to the model parameters or bias terms.

[0093] In another possible implementation, the preset prediction conditions can be further refined into joint-level trigger conditions, meaning prediction is only initiated for local joints where data continuity is impaired or changes are significant, while the measured values ​​are still used for the remaining joints. This reduces unnecessary prediction intervention, decreases computational burden, and maintains the transmission of the true state of stable joints. Regardless of the prediction method used, the generated predicted state quantities are written to the current cycle state cache and the data source type is marked for the next step of fusion processing. Through this processing, the data output from the exoskeleton glove no longer relies entirely on single communication and single sampling results, but introduces explicit modeling of motion trends.

[0094] Based on the above analysis, it can be seen that when the preset prediction conditions are met, the predicted state quantity is determined by using the current motion state and the corresponding historical motion state at the previous moment. This can provide the master end with an effective state compensation basis for the next moment in the event of communication delay, packet loss, or interruption of state update. This step directly addresses the problem in existing solutions where the slave end can only rely on expired states for action control, transforming the control link from "passively waiting for measured values" to "actively compensating based on action trends," thereby improving the continuity of action and time consistency in abnormal network environments.

[0095] S103. Determine the fusion state variables based on the actual state variables and the corresponding predicted state variables of each joint.

[0096] In this step, the actual state quantity can be understood as the effective state data measured by sensors from the current motion state and corrected by calibration. The fused state quantity refers to the target state data used for control output obtained by combining the actual state quantity and the predicted state quantity according to a predetermined rule. The actual state quantity focuses on reflecting the actual motion result measured by the sensors at the current sampling moment, while the predicted state quantity focuses on reflecting the short-term evolution direction of the joint under the current trend. Each has its advantages and limitations: the former has strong realism but is significantly affected by communication and sampling interruptions, while the latter has good continuity but may be offset by model errors. Therefore, by forming a unified control state through fusion processing, both real-time realism and short-term continuity can be taken into account. Fusion can be performed separately at the joint level or at the state dimension level, for example, by setting different weights for angle, angular velocity, and angular acceleration.

[0097] In practice, the acquisition control unit or edge computing unit first reads the actual state quantity and predicted state quantity of each joint in the current cycle, and then calculates the fused state quantity according to the preset fusion rules.

[0098] For example, the preset fusion rule is dynamic weighted linear fusion, and the actual state quantity is . The predicted state variables are The fusion state variable is Therefore, it can be based on: "Perform calculations."

[0099] in This represents the actual state weighting coefficient, and its value can be set between 0 and 1. To adapt the fusion result to different action intensities, It can be dynamically determined, for example, based on the ratio between the current absolute value of angular acceleration and a preset proportional parameter k. When the angular acceleration is small and the movement is relatively smooth, it indicates that the operator is performing low-speed, fine-tuning operations. Increase the angular acceleration to make the fused state variables more dependent on the actual state variables. When the angular acceleration is large and the action switching is frequent, it indicates that short-term trend information is more helpful in compensating for the effects of time delay. In this case, increase the angular acceleration to make the fused state variables more dependent on the actual state variables. Adjust the value to a smaller percentage and increase the proportion of predicted state variables.

[0100] In one possible implementation, fusion is not limited to simple linear weighting; it can also employ Kalman filtering, adaptive weighted fusion, multi-source consistency verification, or a confidence-based state selection mechanism. When using Kalman filtering, the actual state variables can be treated as observations, and the predicted state variables as state priors. The uncertainties of the two types of data are modeled using the observation noise covariance and the process noise covariance, and then the optimal estimated state is output. If the sensor is significantly affected by environmental interference, the observation noise covariance can be increased; if communication jitter significantly increases the prediction error, the process noise covariance can be increased. When using adaptive weighting, the fusion parameters can be dynamically updated by combining sensor health, link quality index, historical prediction error mean, and action mode labels, thereby avoiding the problem of insufficient adaptability of fixed weights in different scenarios.

[0101] In another implementation, to prevent abnormal results from occurring in the fused state variables that do not conform to the physiological constraints of hand movement, the system can also add consistency checks and constraint corrections after the fusion calculation. For example, it checks whether the angles of adjacent joints exceed mechanical limits, and whether the coupling relationship of multiple joints of the same finger exceeds the range of motion. If the limits are exceeded, the system will trim or revert according to a preset constraint model. For the grasping contact state collected by tactile sensors or pressure sensors, it can also be used as an additional constraint condition for the fusion process. When it is detected that the object has made contact and the pressure is rising rapidly, the predicted proportion of angle changes is appropriately reduced to avoid the slave manipulator from continuing to over-close.

[0102] After obtaining the fused state variables for each joint, they are organized into control state frames and written to the transmission buffer. Each state frame can include a joint number, fused state variable, state timestamp, data validity bits, and checksum. Thus, in subsequent command generation and issuance stages, the transmitted data is no longer simply discrete, anomaly-prone raw detection values, but a comprehensive state with temporal continuity and robustness. Based on the above analysis, by fusing the actual state variables of each joint with their corresponding predicted state variables, control jumps caused by sensing errors, network fluctuations, or missing local states can be significantly reduced, improving the smoothness and reliability of the master-end output state. This step constitutes a crucial stabilization step in the entire method, ensuring that the control basis obtained from the slave end retains both the operator's current intent and includes short-term trend compensation, thereby improving tracking accuracy and system stability in complex operating environments.

[0103] Furthermore, when the finger links of the exoskeleton glove adopt a carbon fiber nested telescopic structure, the exoskeleton glove adjusts the length of the finger links and the rotation angle of the pivot based on the fusion state quantity, so that the finger links are adapted to the physiological state of the operator's fingers.

[0104] Specifically, after the exoskeleton glove obtains the fused state data, the acquisition and control unit calculates the target length and target angle of the finger linkage based on the current finger flexion, grip width, and historical movement trends. It then drives the telescopic and rotational mechanisms to adjust the finger linkage's outline to match the operator's finger length, interphalangeal distance, and joint range of motion. The adjustment process can be accomplished by micro-drive components, a lead screw and slider mechanism, or a motor reduction transmission mechanism. Length changes compensate for differences in finger geometry among different users, while angle changes correct the deviation between the linkage axis and the natural bending trajectory of the fingers after wearing the glove, thereby reducing pressure, slippage, and movement lag. The fused state data reflects both the actual real-time acquired state and the predicted trend state, thus providing continuous data for structural adjustment even during communication fluctuations or rapid posture changes.

[0105] This structure, through length self-adaptation and axis angle linkage adjustment, allows the exoskeleton glove to conform to the physiological characteristics of different operators' fingers, maintaining good synchronization during finger flexion, extension, pinching, and lateral fine-tuning. The carbon fiber nested telescopic structure combines lightweight and rigidity advantages, and, together with the highly stable angle detection capability of the dual-axis magnetic encoder, reduces wearing inertia and measurement drift, thereby improving the accuracy of hand motion acquisition and the consistency of remote control. With this implementation, the exoskeleton glove has a wider adaptability to different hand shapes, offers greater comfort during extended wear, and the finger linkages maintain reliable support during dynamic adjustments, which is beneficial for improving the continuity and precision of actions performed by the slave-end android.

[0106] S104. Generate a first control command based on the fused state variables and send it to the slave embodied robot to control the corresponding joint of the slave embodied robot to perform tracking based on the fused state variables.

[0107] In this step, the first control command refers to the control message or control data packet constructed by the master exoskeleton glove based on the fused state variables, which can be recognized and executed by the slave embodied robot. The corresponding joint refers to the execution joint in the slave embodied robot that has a control mapping relationship with each acquisition joint of the exoskeleton glove; this can be a one-to-one mapping or a multi-joint coupling mapping. Execution tracking refers to the slave embodied robot converting the fused state variables in the first control command into motor drive quantities, servo target poses, impedance parameters, or force control reference quantities, and driving each execution joint to converge towards the target state. The embodied robot can be a robotic hand, a humanoid hand, an end effector, or other execution mechanisms with multi-degree-of-freedom joint structures. It should be noted that the conversion process performed in the above execution tracking process can be executed on the exoskeleton glove side or on the slave embodied robot side.

[0108] When executed on the exoskeleton glove side, the edge computing unit first performs master-slave mapping based on the fused state variables. If the joint structure of the master exoskeleton glove is the same as the joint structure of the robotic arm of the slave robot, the fused state variables can be directly used as the slave target values. If the structures are different, conversion is required based on pre-established kinematic mapping relationships. For example, the fused state variables can be converted into servo angles of the robotic fingers and end-effector posture commands according to joint ratio mapping, cooperative joint mapping, or task space mapping. After mapping, the master exoskeleton glove encapsulates the target values ​​into a first control command. The control command may include a message header, device identifier, timestamp, fused state variables of each joint target, force control parameters, anomaly flags, and a check field. The message encoding method can adopt a binary fixed-length format, a lightweight custom protocol, or an industrial bus protocol compatible format to adapt to different communication links.

[0109] When executing on the slave embodied robot side, the exoskeleton glove generates a first control command containing fused state variables and sends it to the slave embodied robot via a wireless communication link. The communication link can be Wi-Fi (Wireless Fidelity), 5G, Bluetooth Low Latency, a proprietary RF link, or wired Ethernet. To enhance transmission reliability, the master exoskeleton glove can attach a sequence number and a cyclic redundancy check (CRC) code to the first control command and enable transmission confirmation and timeout retransmission mechanisms. When the link quality is poor, double-buffered transmission and priority queue scheduling can also be used to prioritize the transmission of the latest control frame, avoiding the backlog of old frames that could cause execution lag in the slave embodied robot.

[0110] By receiving the first control command generated from the fused state variables, the embodied robot no longer obtains a single actual acquired value, but rather a control basis that takes into account both the current real state and short-term motion trends. This ensures the continuity of target information even during communication jitter, packet loss, or discontinuous state updates. Furthermore, parsing the first control command yields the target fused state variables corresponding to each joint, enabling the slave to obtain a unified and executable target input at the joint level. This results in smoother and more synchronized tracking processes for each joint. Controlling the corresponding joints to perform tracking operations based on the target fused state variables reduces motion tailing, posture jumps, and grasping deviations caused by outdated or abrupt data, thereby improving the control accuracy, operational stability, and operational reliability of the embodied robot during teleoperation.

[0111] S105. Receive the first control command generated by the exoskeleton glove based on the fusion state variables, parse the first control command, and obtain the target fusion state variables corresponding to each joint. Based on each target fusion state variable, control the corresponding joint of the embodied robot to perform tracking operations.

[0112] Among them, the fused state variables are the predicted state variables determined based on the current motion state of each joint of the exoskeleton glove and the corresponding historical motion state at the previous moment, and are obtained together with the actual state variables.

[0113] In this step, after receiving the first control command, the slave-end embodied robot's controller analyzes the target fusion state variables of each joint and executes tracking control according to the control mode. The control mode can be position control, velocity control, impedance control, model predictive control, or hybrid force-position control. For example, when the target task is stable grasping, the slave-end embodied robot's controller can use position control with gripping force feedforward compensation to first track the joints to the fusion angle, and then limit the continued closing amplitude according to the contact force parameters in the command. When the target task is compliant contact, impedance control can be used to allow the slave joints to maintain compliance with the environment while tracking the target fusion state variables.

[0114] In another implementation, the first control command may also include a local adaptive gain parameter, which allows the slave end to automatically adjust the servo controller parameters based on the current load, friction estimation, and execution error, in order to improve tracking stability under different working objects and environmental conditions.

[0115] To ensure master-slave consistency, the slave android can periodically feed back its actual execution status to the master exoskeleton glove. The master exoskeleton glove compares this with the previously sent fused state variables. If the deviation exceeds a set range, it updates the mapping parameters, fusion parameters, or prediction parameters for the next cycle. Although this application focuses on the control generation process on the master exoskeleton glove side, this distribution and closed-loop coordination mechanism allows the fused state variables to be truly implemented as continuous tracking movements of the slave android's joints, avoiding the slave android performing mechanical execution based solely on discrete and potentially distorted actual state points. Based on the above analysis, generating the first control command based on the fused state variables and distributing it to the slave android enables the corresponding joints of the slave android to perform tracking based on the fused state variables. This ensures the stable transmission of the master exoskeleton glove's comprehensive expression of the operator's intentions to the execution side. Even with communication anomalies and data fluctuations, it maintains continuous joint movements, smooth posture transitions, and task execution accuracy, thereby improving the overall reliability and safety of the android's teleoperation system.

[0116] This application provides a remote control method for an embodied robot. Through multi-dimensional motion state acquisition, predictive compensation triggered by abnormal conditions, state fusion for motion continuity, and generation and issuance of control commands for slave execution, the control basis of the exoskeleton glove output is no longer limited to the measured state at a single moment, but can comprehensively reflect the operator's current action result and subsequent short-term motion trends. Thus, even in the event of communication latency, data loss, or discontinuous state updates, the slave embodied robot can still continuously obtain a stable and trend-continuous target state, thereby reducing motion tailing, posture jumps, and grasping misalignment, and improving master-slave motion consistency, joint tracking accuracy, and system operational stability.

[0117] It should be understood that the above examples are merely illustrative and not limiting. In one possible embodiment, the sensor type, prediction model form, fusion algorithm type, and communication protocol structure can all be adjusted according to the specific application scenario. As long as it can generate a fused state based on the actual state and the predicted state and drive the slave tracking control, it can fall within the technical concept of the embodiments of this application.

[0118] Figure 2 A flowchart illustrating a remote control method for an embodied robot provided in this application embodiment. Figure 2 This embodiment provides a detailed explanation of the steps involved in determining the fusion state quantity of the exoskeleton glove based on the actual state of each joint and the corresponding predicted state. For example... Figure 2 As shown, the method includes:

[0119] S201. Determine the weighting coefficient based on the ratio of the current angular acceleration to the preset proportional parameter. The larger the ratio, the smaller the corresponding weighting coefficient.

[0120] In this step, in the remote control scenario, joint angles directly correspond to the target positions of each joint of the slave-end embodied robot, suitable for trajectory tracking tasks in position servo mode. Angular velocity, on the other hand, reflects the trend of motion change and is suitable for smooth following tasks in speed servo mode. Since different slave-end embodied robots have different servo drive architectures and control interfaces, some robots prioritize receiving angle commands for position closed-loop control, while others prioritize receiving speed commands for speed closed-loop control. Therefore, setting the fused state variables to fused joint angles or fused angular velocities allows for flexible adaptation to various control interface types of slave-end embodied robots, improving the method's versatility and deployment convenience.

[0121] Correspondingly, the predicted state variables include the predicted angular velocities and predicted joint angles of each joint in the exoskeleton glove. The specific implementation process is as follows:

[0122] For each joint, the predicted joint angle is determined based on the historical joint angle, historical angular velocity, historical angular acceleration, and predicted time step from the previous moment; the predicted angular velocity is determined based on the historical angular velocity, angular acceleration, and predicted time step.

[0123] In this embodiment, historical joint angles are used to characterize the joint's attitude reference at the previous moment, historical angular velocity is used to characterize the rotational change trend of the joint between adjacent moments, historical angular acceleration is used to characterize the rate of change of angular velocity, and the prediction time step is used to limit the time span extrapolated to the target prediction moment. These parameters can be obtained by time-aligning the sensor sampling results of the exoskeleton glove and stored in the historical state cache one joint at a time for retrieval at the current moment.

[0124] In practical implementation, the predicted joint angle can be expressed as: The predicted angular velocity can be expressed as: In the formula and These represent the predicted joint angle and predicted angular velocity at the target prediction time, respectively.

[0125] When the sampling frequency is high, a discrete kinematic integral model can be used to superimpose the cumulative change of angular velocity over the time step onto the historical joint angle, thus forming a continuous and smooth angle prediction result. The predicted angular velocity can be obtained by superimposing the velocity increment caused by angular acceleration on the historical angular velocity and updating it in conjunction with the prediction time step to obtain the estimated angular velocity value at the target time. To reduce the impact of instantaneous noise on the prediction results, the historical angular velocity and historical angular acceleration can be processed by low-pass filtering or moving average before participating in the extrapolation calculation.

[0126] This predictive state generation mechanism plays a forward compensation role in the control link, enabling the master exoskeleton glove to still infer the future change trend of the joints based on the historical motion state of the previous moment when the current motion state update is incomplete or there is a short time delay. It also uses the predicted joint angle and predicted angular velocity as the basis for the fusion state variables, thereby outputting more continuous control basis to the slave embodied robot.

[0127] By adopting this implementation method, the continuity of joint states in the time dimension can be improved, control abrupt changes caused by communication jitter or data loss can be reduced, the embodied robot can maintain a more stable tracking effect during grasping, assembly and fine operation, and the response accuracy and motion consistency of the teleoperation system can be improved.

[0128] Furthermore, the fused joint angle refers to the angle quantity used for control obtained by weighted synthesis of the current joint angle and the predicted joint angle, while the fused angular velocity refers to the velocity quantity used for control obtained by weighted synthesis of the current angular velocity and the predicted angular velocity. Weighting coefficients reflect the degree of confidence the current joint motion intensity has in the actual and predicted states. Preset proportional parameters can be pre-written by the controller and used to limit the sensitivity of angular acceleration to weight changes. A larger angular acceleration indicates more intense joint motion; the exoskeleton glove can correspondingly reduce the weighting coefficients, thereby increasing the proportion of the actual state in the fusion result and reducing the involvement of the predicted state.

[0129] In one implementation, the weighting coefficients can be obtained by normalizing the ratio of angular acceleration to a preset proportional parameter to ensure that their values ​​are within a preset range, facilitating stable weighting. The calculation process is as follows: ,in Indicates the current angular acceleration. This represents the preset proportional parameter. The purpose of this expression is to reduce the weight of the actual state as the intensity of the action decreases, so that the fusion result transitions more smoothly to trend prediction control.

[0130] Therefore, when the angular acceleration is small, the fusion result is closer to the predicted state, thus reducing the impact of sampling fluctuations and time delays on control continuity. When the angular acceleration increases, the fusion result gradually deviates from the actual acquired state to maintain the timeliness of the action response. This fusion calculation can be performed in real time by the control processing unit inside the exoskeleton glove. The calculated fusion state is then encoded and sent to the slave-end embodied robot to drive the corresponding joint tracking motion.

[0131] S202. For each joint, determine the first fusion state quantity by multiplying the predicted state quantity of the joint by the weight coefficient. And determine the second fusion state quantity by multiplying the actual state quantity of the joint by the product of 1 and the difference between the weight coefficient.

[0132] In this step, both the first fused state quantity and the second fused state quantity can be calculated one by one according to the joint dimension. The first fused state quantity corresponds to the weighted result of the predicted state quantity, and the second fused state quantity corresponds to the weighted result of the actual state quantity. The two are added together to form the fused state quantity, which serves as the basis for the generation of subsequent control commands.

[0133] The first fusion state variable can be expressed as: ;

[0134] The second fusion state variable can be expressed as: ;

[0135] in, These are the weighting coefficients. To predict state variables, This refers to the actual state quantity.

[0136] S203. Determine the fusion state quantity based on the sum of the first fusion state quantity and the second fusion state quantity.

[0137] In this step, the fusion state variables can be expressed as:

[0138] .

[0139] in, This refers to the fusion state variable.

[0140] The aforementioned fusion method enables the exoskeleton glove to dynamically balance the effects of the actual and predicted states under different motion intensities. This avoids the jitter and lag caused by relying solely on current sampled values, as well as the deviation risks associated with relying entirely on predicted values, thus maintaining the continuity and smoothness of joint control during teleoperation. Because the fused state variables can adaptively adjust their weights according to changes in angular acceleration, they can still provide relatively stable target joint variables for the slave embodied robot in scenarios such as rapid grasping, wrist rotation, or posture switching, improving motion synchronization accuracy and control robustness.

[0141] In one possible implementation, control is achieved solely through fused state variables at the position or velocity levels. The end-effector robot can only replicate the operator's hand movements, unable to sense or respond to the operator's intended torque. This leads to a discrepancy between the gripping force at the end effector of the end-effector robot and the operator's desired force in tasks requiring force interaction, such as grasping, pressing, or fine assembly. This can result in problems like loose gripping causing objects to slip or excessive gripping causing damage. Furthermore, since the force control loop passively adjusts based solely on the actual contact force fed back from the exoskeleton glove's force sensor, there is a significant lag in force adjustment when there is a delay in force sensor signal transmission or inertia in the end-effector robot's dynamic response, making rapid and compliant force tracking difficult.

[0142] Therefore, in order to enable the slave embodied robot to perceive and respond to the operator's force intentions in advance, and improve the operational compliance and execution accuracy in force interaction tasks, this application introduces a feedforward compensation mechanism that derives the joint torque intentions based on the current angular acceleration. The specific implementation process is as follows:

[0143] The operator's intended joint torque is derived by inversely from the current angular acceleration of each joint. The intended joint torque is positively correlated with the rotational inertia and current angular acceleration of the corresponding joint. The intended joint torque is introduced as a feedforward parameter into the force control loop of the slave robot and superimposed with the actual contact force fed back by the force sensor of the exoskeleton glove to generate a second control command that includes position tracking and force feedforward compensation. The second control command is sent to the slave robot so that the slave robot can respond to the operator's force intention in advance and adjust the clamping force at the end of the actuator of the slave robot.

[0144] The joint torque is intended to characterize the operator's force application trend at the joint level. It can be determined by the current angular acceleration and joint rotational inertia. The larger the rotational inertia and the higher the angular acceleration, the more significant the torque requirement. The actual contact force can be obtained in real time by the force sensor of the exoskeleton glove, and its value reflects the contact change between the gripped object and the actuator end effector. The second control command combines the position tracking quantity and the force feedforward compensation quantity and outputs them, enabling the slave-end embodied robot to compensate for force changes while maintaining trajectory synchronization. The above force control loop can be implemented by a digital controller in the control processing unit and calculated in combination with impedance control or admittance control. The sensor used can be a strain gauge, piezoresistive, or six-dimensional force sensor. In practical applications, other models can also be selected for this component, and this application embodiment does not limit this.

[0145] In its implementation, after acquiring the current angular acceleration of each joint, the control processing unit inverts the equivalent torque based on pre-calibrated joint rotational inertia parameters and inputs this torque as a feedforward compensation term into the closed-loop controller of the slave-end embodied robot. Simultaneously, the controller reads the actual contact force feedback from the exoskeleton glove side, superimposes or weights the feedforward and feedback terms to form a second control command that balances trajectory tracking and contact force adjustment. This second command is then sent to the slave-end embodied robot via the communication link. The slave-end embodied robot adjusts the output current or driving force of its end-effector gripping mechanism according to the second control command, thereby changing the gripping force and enabling the slave to respond in advance when the operator increases or decreases their force.

[0146] This method infers the force intention through angular acceleration during the use of exoskeleton gloves and feeds it forward to compensate the force control loop. This allows the slave robot to adjust the output in advance without waiting for the contact force error to accumulate, reducing control lag. Since the second control command contains both position tracking information and force compensation information, it can maintain the continuity of action during grasping, pinching, or assembly, and reduce the phenomenon of excessive or insufficient clamping caused by force feedback delay.

[0147] By adopting this method, the end-device robot can more accurately follow the operator's force trend, improve the timeliness and stability of clamping force adjustment, reduce the impact of sudden contact changes on teleoperation accuracy, and enhance safety and task completion reliability in complex scenarios.

[0148] Figure 3 A schematic diagram of the structure of a remote control device for an embodied robot provided in this application embodiment. Figure 1 .like Figure 3 As shown, the remote control device 30 for the android is applied to the slave android and includes:

[0149] The receiving module 301 is used to receive a first control command generated by the exoskeleton glove based on the fusion state quantity, wherein the fusion state quantity is a predicted state quantity determined based on the current motion state of each joint of the exoskeleton glove and the corresponding historical motion state at the previous moment, and is obtained from the actual state quantity.

[0150] The parsing module 302 is used to parse the first control command and obtain the target fusion state quantities corresponding to each joint;

[0151] The execution module 303 is used to control the corresponding joints of the embodied robot to perform tracking operations based on the fused state variables of each target.

[0152] This embodiment provides a remote control device for a slave-end embodied robot, which can execute the remote control method for a slave-end embodied robot provided in the above-described method embodiment. The implementation principle and technical effect are similar, and will not be described in detail here.

[0153] Figure 4 A schematic diagram of the structure of a remote control device for a body-worn robot applied to an exoskeleton glove, provided in an embodiment of this application. Figure 2 .like Figure 4 As shown, the remote control device 40 of the android is applied to the exoskeleton glove, including:

[0154] The acquisition module 401 is used to acquire the current motion state of each joint through a preset sensor at the pivot of the exoskeleton glove;

[0155] The calculation module 402 is used to determine the predicted state quantity of each joint based on each current motion state and the corresponding historical motion state of the previous moment when the preset prediction conditions are met.

[0156] The calculation module 402 is also used to determine the fused state quantity based on the actual state quantity of each joint and the corresponding predicted state quantity;

[0157] The control module 403 is used to generate a first control command based on the fused state variables and send it to the slave embodied robot to control the corresponding joint of the slave embodied robot to perform tracking based on the fused state variables.

[0158] In one possible implementation, the acquisition module 401 is further configured to:

[0159] The joint angle of each joint is obtained by collecting the current absolute angle of each joint based on the preset sensors.

[0160] Perform a first-order difference operation on the joint angle and the corresponding adjacent historical joint angle to obtain the current angular velocity;

[0161] Perform a first-order difference operation between the current angular velocity and the corresponding adjacent historical angular velocities to obtain the current angular acceleration.

[0162] In one possible implementation, the computing module 402 is further configured to:

[0163] For each joint, the predicted joint angle is determined based on the historical joint angle, historical angular velocity, historical angular acceleration, and predicted time step from the previous moment.

[0164] The predicted angular velocity is determined based on the historical angular velocity, angular acceleration, and predicted time step.

[0165] In one possible implementation, the fusion state variable is a fusion joint angle or a fusion angular velocity, and the calculation module 402 is further used for:

[0166] The weighting coefficient is determined based on the ratio of the current angular acceleration to the preset proportional parameter. The larger the ratio, the smaller the corresponding weighting coefficient.

[0167] For each joint, the first fused state quantity is determined by multiplying the predicted state quantity of the joint with the weighting coefficient.

[0168] The second fusion state quantity is determined by multiplying the actual state quantity of the joint with the product of 1 and the difference between the weighting coefficients.

[0169] The fusion state quantity is determined based on the sum of the first fusion state quantity and the second fusion state quantity.

[0170] In one possible implementation, the control module 403 is further configured to:

[0171] The operator's intended joint torque is derived by inversely from the current angular acceleration of each joint. The intended joint torque is positively correlated with the moment of inertia and current angular acceleration of the corresponding joint.

[0172] The joint torque intention is introduced as a feedforward parameter into the force control loop of the end-body robot and superimposed with the actual contact force fed back by the force sensor of the exoskeleton glove to generate a second control command that includes position tracking and force feedforward compensation.

[0173] A second control command is sent to the slave robot so that the slave robot can respond in advance to the operator's force intention and adjust the clamping force at the end of the actuator of the slave robot.

[0174] In one possible implementation, the preset prediction conditions within the calculation module 402 include at least one of the following:

[0175] The communication latency between the exoskeleton glove and the slave robot exceeds a preset latency threshold.

[0176] Alternatively, the safety threshold for end-body robots may be lowered.

[0177] In one possible implementation, the finger linkage of the exoskeleton glove adopts a carbon fiber nested telescopic structure, the preset sensor is a dual-axis magnetic encoder, and the control module 403 is also used for:

[0178] The length of the finger link and the rotation angle of the pivot are adjusted based on the fusion state variables to adapt the finger link to the physiological state of the operator's fingers.

[0179] This embodiment provides a remote control device for an exoskeleton glove-based robot, which can execute the remote control method for an exoskeleton glove-based robot provided in the above-described method embodiment. The implementation principle and technical effects are similar, and will not be described in detail here.

[0180] Figure 5 This is a schematic diagram of the structure of a smart device provided in an embodiment of this application. Figure 5 As shown, the smart device 50 includes a processor 501 and a memory 502 communicatively connected to the processor 501. Optionally, the smart device 50 also includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus 504.

[0181] Memory 502 stores instructions executed by the computer;

[0182] The processor 501 executes computer execution instructions stored in the memory 502 to implement the remote control method of the embodied robot as described above.

[0183] At least one processor 501 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0184] Optionally, in specific implementations, the processor 501 and memory 502 are implemented independently. In this case, the processor 501 and memory 502 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc., but this does not imply that there is only one bus or one type of bus.

[0185] Optionally, in a specific implementation, if the processor 501 and the memory 502 are integrated on a single chip, the processor 501 and the memory 502 can communicate through an internal interface.

[0186] This application also provides a computer storage medium storing computer execution instructions, which, when executed by a processor, implement the aforementioned remote control method for the embodied robot.

[0187] The aforementioned computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The computer-readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0188] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Alternatively, the readable storage medium can be an integral part of the processor. Both the processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the control device of a garment handling apparatus.

[0189] This unit division is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0190] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0191] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0192] If this function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods indicated in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0193] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0194] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it is readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A remote control method for an embodied robot, characterized in that, Applications include: [List of applications] Receive a first control command generated by the exoskeleton glove based on fused state quantities, wherein the fused state quantities are obtained by combining the predicted state quantities determined based on the current motion state of each joint of the exoskeleton glove and the corresponding historical motion state at the previous moment with the actual state quantities. Parse the first control command to obtain the target fusion state variables corresponding to each joint; Based on the target fusion state quantities, the corresponding joints of the embodied robot are controlled to perform tracking operations.

2. A remote control method for an embodied robot, characterized in that, Applied to exoskeleton gloves, the method includes: The current motion state of each joint is obtained through preset sensors at the pivot point of the exoskeleton glove; When the preset prediction conditions are met, the predicted state quantity of each joint is determined based on each current motion state and the corresponding historical motion state at the previous moment. The fusion state quantities are determined based on the actual state quantities of each joint and the corresponding predicted state quantities. A first control command is generated based on the fused state variables and sent to the slave embodied robot to control the corresponding joint of the slave embodied robot to perform tracking based on the fused state variables.

3. The method according to claim 2, characterized in that, The acquisition of the current motion state of each joint includes: The current absolute angle of each joint is collected by the preset sensor to obtain the joint angle of each joint; Perform a first-order difference operation on the joint angle and the corresponding adjacent historical joint angle to obtain the current angular velocity; The current angular velocity is obtained by performing a first-order difference operation between the current angular velocity and the corresponding adjacent historical angular velocity.

4. The method according to claim 3, characterized in that, The step of determining the predicted state quantities of each joint based on each current motion state and the corresponding historical motion state at the previous moment includes: For each joint, the predicted joint angle is determined based on the historical joint angle, historical angular velocity, historical angular acceleration, and predicted time step from the previous moment. The predicted angular velocity is determined based on the historical angular velocity, angular acceleration, and the predicted time step.

5. The method according to claim 4, characterized in that, The fusion state quantity is a fusion joint angle or a fusion angular velocity. Determining the fusion state quantity based on the actual state quantity of each joint and the corresponding predicted state quantity includes: The weighting coefficient is determined based on the ratio of the current angular acceleration to the preset proportional parameter, wherein the larger the ratio, the smaller the corresponding weighting coefficient. For each joint, a first fused state quantity is determined by multiplying the predicted state quantity of the joint with the weighting coefficient. The second fusion state quantity is determined by multiplying the actual state quantity of the joint with the product of 1 and the difference between the weighting coefficient; The fusion state quantity is determined based on the sum of the first fusion state quantity and the second fusion state quantity.

6. The method according to any one of claims 3-5, characterized in that, The method further includes: The operator's intended joint torque is derived by inversely from the current angular acceleration of each joint, wherein the intended joint torque is positively correlated with the moment of inertia of the corresponding joint and the current angular acceleration; The joint torque intention is introduced as a feedforward parameter into the force control loop of the slave-end android, and superimposed with the actual contact force fed back by the force sensor of the exoskeleton glove to generate a second control command that includes position tracking and force feedforward compensation. The second control command is sent to the slave robot so that the slave robot can respond in advance to the operator's force intention and adjust the clamping force at the end of the actuator of the slave robot.

7. The method according to claim 2, characterized in that, The preset prediction conditions include at least one of the following: The communication delay between the exoskeleton glove and the slave-end android exceeds a preset delay threshold. Alternatively, the safety threshold of the embody robot may be reduced.

8. The method according to claim 2, characterized in that, The finger linkage of the exoskeleton glove adopts a carbon fiber nested telescopic structure, the preset sensor is a dual-axis magnetic encoder, and the method further includes: The length of the finger link and the rotation angle of the pivot are adjusted based on the fusion state quantity so that the finger link is adapted to the physiological state of the operator's fingers.

9. A remote control device for an embodied robot, characterized in that, Applications include: [List of applications] The receiving module is used to receive a first control command generated by the exoskeleton glove based on the fusion state quantity, wherein the fusion state quantity is a predicted state quantity determined based on the current motion state of each joint of the exoskeleton glove and the corresponding historical motion state at the previous moment, and is obtained from the actual state quantity. The parsing module is used to parse the first control command and obtain the target fusion state quantities corresponding to each joint; The execution module is used to control the corresponding joints of the embodied robot to perform tracking operations based on the fusion state variables of each target.

10. A smart device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1 to 8.

Images