Flexible somatosensory interaction system and method based on intention probability mapping

Abstract

The present application relates to the field of human-computer interaction and intelligent control technology, in particular to a flexible somatosensory interaction system and method based on intention probability mapping. By setting up a flexible somatosensory acquisition module, an intention modeling module, a flexible-rigid mapping field construction module, a dynamic intention screening module and an execution control module, the multi-modal somatosensory information of the operator is collected, the probability distribution model of the action intention is established, and the flexible-rigid mapping field is constructed based on the information retention and minimum semantic loss principle, the continuous action space of the human body is converted into a limited control instruction set executable by the mechanical arm. The system dynamically screens the optimal control instruction under multiple candidate intentions using the confidence driving mechanism, and realizes the precise response of the mechanical arm to the operator's action through closed-loop feedback. The problem of mismatch between flexible action and rigid control freedom and intention recognition ambiguity in the prior art is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction and intelligent control, specifically to a flexible haptic interaction system and method based on intent probability mapping. Background Technology

[0002] With the rapid development of human-machine collaboration technology, robotic arms have been widely used in fields such as medical-assisted surgery, remote operation, hazardous environment work, and human-machine collaborative manufacturing. In order to achieve natural and intuitive human-machine interaction, researchers usually collect motion information of the operator's upper limbs through wearable sensing devices, and map human movements into control commands for the robotic arm, thereby realizing real-time control of the robotic arm.

[0003] However, existing technologies generally employ point-to-point geometric mapping or least-squares inverse kinematics to directly map human joint poses to the end effector pose of a robotic arm. While this method is simple to implement, it has significant limitations: First, the human upper limb has a continuously deformable flexible structure and extremely high degrees of freedom, far exceeding the finite rigid joint degrees of freedom of a robotic arm. Traditional low-dimensional mapping models cannot fully express these complex nonlinear motion characteristics, leading to accuracy loss and posture deviations during motion transmission. Second, the same motion signal often corresponds to multiple operational intentions, and existing control strategies lack mechanisms for describing and handling intention ambiguity, making the system prone to control misjudgments when faced with vague or transitional actions. Third, in complex work scenarios, the operator's intentions often exhibit continuous changes and the coexistence of multiple candidate states. Existing technologies treat the motion space as a deterministic input, lacking modeling of the uncertainty of intentions, and therefore cannot achieve probabilistic expression and optimization of the operator's true intentions at the control level.

[0004] In summary, existing methods of mapping human movements to robotic arm control generally suffer from the following problems:

[0005] The large difference in degrees of freedom between the flexible structure of the human body and the rigid structure of the robotic arm leads to inaccurate motion mapping.

[0006] The mapping model only supports deterministic single-solution output and cannot handle the diversity and uncertainty of intent;

[0007] There is no flexible mapping mechanism between continuous and discrete spaces, and there is a lack of evaluation methods for information preservation and semantic consistency. Summary of the Invention

[0008] This application provides a flexible somatosensory interaction system and method based on intent probability mapping, which solves the control deviation problem caused by the mismatch between the human body's degree of freedom of movement and the rigid structure of the robotic arm in existing flexible somatosensory interaction systems. It overcomes the challenges of intent recognition errors and semantic loss caused by the limited number of gestures, and achieves accurate mapping and intent recognition from flexible movements to rigid execution. Figure 1Human-machine collaborative control.

[0009] To achieve the above objectives, the embodiments of this application disclose the following technical solutions:

[0010] On the one hand, this solution discloses a flexible haptic interaction system based on intent probability mapping, characterized by comprising:

[0011] The flexible motion sensing acquisition module is used to acquire multimodal motion information of the operator's limbs, including joint bending angles, postural acceleration and electromyographic signals, and convert the information into standardized flexible motion vectors;

[0012] The intention modeling module is used to perform continuous space analysis on the flexible motion vector and establish an intention probability distribution model to represent the operator's potential control intention under multi-degree-of-freedom motion, wherein the intention probability distribution includes target direction, force feeling trend and motion confidence.

[0013] The flexible-rigid mapping field construction module is used to construct a nonlinear mapping field between the flexible human body action space and the rigid action space of the robotic arm based on the intention probability distribution. The mapping field maps the continuously deforming flexible action into a discrete set of robotic arm control commands by maintaining the principle of minimizing the change in information entropy.

[0014] The dynamic intent filtering module is used to dynamically filter and sort candidate control instructions in the set of robotic arm control instructions based on action confidence threshold and historical execution deviation, and determine the optimal execution instruction.

[0015] The execution control module is used to convert the optimal execution command into a drive signal that the robotic arm can execute, and to perform closed-loop correction based on real-time motion feedback, so that the robotic arm's motion trajectory is consistent with the operator's original motion semantics.

[0016] The robotic arm body is used to execute the drive signals output by the execution control module to realize the corresponding posture adjustment and action output;

[0017] Through the above structure, the somatosensory interaction system realizes the mapping of human flexible movements into rigid motion commands that can be executed by the robotic arm while preserving the intention distribution information, thereby improving the control accuracy and semantic consistency.

[0018] On the other hand, this solution discloses a flexible haptic interaction method based on intent probability mapping, including the following steps:

[0019] Step S1: Multimodal somatosensory signal acquisition. Through a flexible somatosensory acquisition module worn on the operator's arm and wrist, multimodal data including bending sensing signals, inertial signals, electromyographic signals, and temperature signals are acquired. The multimodal data is then time-synchronized and normalized to obtain flexible motion vectors for subsequent modeling.

[0020] Step S2: Model the intent probability distribution. Divide the flexible action vector into time windows and extract the posture feature sequence. Calculate the intent probability distribution of the action in the semantic action space based on the posture feature sequence. The intent probability distribution includes direction probability, velocity trend probability, and target intention probability. Output a distribution matrix containing multiple candidate intents as input for subsequent mapping.

[0021] Step S3: Construct a flexible-rigid mapping field to establish a mapping field that describes the correspondence between the flexible motion space and the rigid motion space of the robotic arm; adjust the mapping weights of the flexible space and the rigid space according to the intention probability distribution, and calculate the information entropy change of each candidate mapping path using the information retention criterion; select the mapping path with the minimum information loss, and generate multiple candidate control commands for the robotic arm with corresponding confidence levels.

[0022] Step S4: Dynamic intent filtering, performing semantic clustering on the candidate control commands, aggregating similar action commands into candidate clusters; calculating the comprehensive confidence of each candidate cluster, and determining the final execution command based on an adaptive threshold; if the confidence is lower than a preset threshold, triggering a visual or voice confirmation request to prevent erroneous actions.

[0023] Step S5: Execution and closed-loop correction. The final execution command is calculated into robotic arm drive signal parameters to control each joint of the robotic arm to perform corresponding actions; feedback signals of robotic arm pose, torque and end force are collected; the mapping field parameters are dynamically corrected according to the feedback error to achieve long-term adaptive control of the robotic arm's movements.

[0024] Step S6: Continuous Intent Update. The execution results are compared with historical action features to dynamically update the intent probability model; this allows the system to gradually optimize the matching accuracy of the flexible-rigid mapping in repetitive tasks, improving the accuracy of intent interpretation and the stability of control.

[0025] Through the above steps, the robotic arm motion-sensing interaction method can establish a continuous-discrete fusion mapping relationship between flexible human body movements and rigid robotic arm movements, thereby achieving accurate recognition and low-latency control execution of multi-intent commands.

[0026] This invention establishes a semantically consistent mapping relationship between human flexible movements and rigid robotic arm control by introducing a flexible-rigid mapping field and an intent probability distribution model, significantly improving the control accuracy and intent reproduction capability of the somatosensory interaction system. By fusing multimodal signals in the flexible somatosensory acquisition module, comprehensive capture of posture, acceleration, electromyography, and force trends is achieved, enabling the system to accurately depict the operator's true motion state. The probabilistic design of the intent modeling module eliminates reliance on single gestures or fixed thresholds for action recognition, instead dynamically inferring the operator's control target based on intent distribution, effectively avoiding misjudgments caused by motion ambiguity or posture overlap in traditional systems. The establishment of the flexible-rigid mapping field adopts the principles of information preservation and minimum semantic loss, ensuring that human movements maintain consistency in motion trends and force direction during mapping to robotic arm commands, guaranteeing synchronization between output actions and human intent. Combined with a confidence-driven dynamic filtering mechanism, the system can automatically select the optimal execution path from multiple candidate intents and continuously correct mapping errors through closed-loop feedback, achieving adaptive and precise control of the robotic arm. Attached Figure Description

[0027] Figure 1 This is a system overall structure block diagram according to an embodiment of the present invention;

[0028] Figure 2 This is a schematic diagram of the flexible-rigid mapping field model according to an embodiment of the present invention;

[0029] Figure 3 This is a flowchart illustrating the modeling process for the intent probability distribution in an embodiment of the present invention.

[0030] Figure 4 This is a logic diagram of dynamic intent filtering and confidence decision-making in an embodiment of the present invention;

[0031] Figure 5 This is a flowchart illustrating the closed-loop execution process of the system according to an embodiment of the present invention.

[0032] Figure 6 This is a schematic diagram of human-computer interaction according to an embodiment of the present invention. Detailed Implementation

[0033] Specific embodiments of the invention will now be described in detail. Although the invention is described in conjunction with these specific embodiments, it should be understood that the invention is not intended to be limited to these specific embodiments. Rather, these embodiments are intended to cover alternative, modified, or equivalent embodiments that may be included within the spirit and scope of the invention as defined by the claims. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. The invention may be practiced without some or all of these specific details. In other instances, well-known processes have not been described in detail so as not to unnecessarily obscure the invention.

[0034] When used in conjunction with the terms "comprising," "method comprising," or similar language in this specification and appended claims, the singular forms "a," "some," and "the" include plural references unless the context clearly indicates otherwise. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0035] Application Overview: In existing technologies, flexible haptic control systems typically rely on single gestures or motion mapping to drive robotic arms. Because human limbs exhibit continuous and multi-degree-of-freedom flexible movements, while the robotic arm's motion space is a finite, rigid structure, there are significant differences in their motion representation. For example, the human body has far more degrees of freedom than a robotic arm, and this difference in degrees of freedom leads to pose deviations and semantic losses during motion transitions, preventing the robotic arm from accurately reflecting the operator's true intentions. Furthermore, the number of gestures is limited while operational intentions are diverse. When the system relies solely on posture or angle information for recognition, it is prone to misjudging intentions, resulting in erroneous movements or delayed responses from the robotic arm. If these problems are not addressed, the flexible haptic interaction system will ultimately suffer from decreased control accuracy, fragmented interactive semantics, and limited human-machine collaboration performance.

[0036] To address the aforementioned challenges, this application first considers establishing a continuous and semantically consistent mapping between flexible human body movements and rigid robotic arm control. This application attempts to map the continuous human body movement space into a finite control space for the robotic arm by constructing a flexible-rigid mapping field and an intent probability distribution model, and introduces an information preservation criterion to minimize semantic loss in the movements. To resolve the issues of mismatched degrees of freedom and intent recognition bias, the system employs a confidence-driven filtering mechanism to dynamically determine the optimal execution path from multiple candidate intents. This enables the robotic arm to accurately respond to complex operator movements under multi-degree-of-freedom conditions, making the movement execution more aligned with human intent and significantly improving the control accuracy and intelligence level of the flexible wearable somatosensory interaction system.

[0037] Example 1

[0038] A flexible haptic interaction system based on intent probability mapping includes:

[0039] The flexible motion sensing acquisition module is used to acquire multimodal motion information of the operator's limbs, including joint bending angles, postural acceleration and electromyographic signals, and convert the information into standardized flexible motion vectors;

[0040] The intention modeling module is used to perform continuous space analysis on the flexible motion vector and establish an intention probability distribution model to represent the operator's potential control intention under multi-degree-of-freedom motion, wherein the intention probability distribution includes target direction, force feeling trend and motion confidence.

[0041] The flexible-rigid mapping field construction module is used to construct a nonlinear mapping field between the flexible human body action space and the rigid action space of the robotic arm based on the intention probability distribution. The mapping field maps the continuously deforming flexible action into a discrete set of robotic arm control commands by maintaining the principle of minimizing the change in information entropy.

[0042] The dynamic intent filtering module is used to dynamically filter and sort candidate control instructions in the set of robotic arm control instructions based on action confidence threshold and historical execution deviation, and determine the optimal execution instruction.

[0043] The execution control module is used to convert the optimal execution command into a drive signal that the robotic arm can execute, and to perform closed-loop correction based on real-time motion feedback, so that the robotic arm's motion trajectory is consistent with the operator's original motion semantics.

[0044] The robotic arm body is used to execute the drive signals output by the execution control module to realize the corresponding posture adjustment and action output;

[0045] Through the above structure, the somatosensory interaction system realizes the mapping of human flexible movements into rigid motion commands that can be executed by the robotic arm while preserving the intention distribution information, thereby improving the control accuracy and semantic consistency.

[0046] In flexible state space With robotic arm motion space Establish a probability mapping field between them:

[0047] ;

[0048] in, It is a flexible-rigid mapping function. For the rigid motion space of the robotic arm, Let be the set of probability distributions in the robotic arm's motion space; In a flexible state The corresponding rigid action probability field.

[0049] During implementation, the flexible motion sensing module collects multimodal signals from the operator's arm, wrist, or upper limb in real time. An embedded data fusion circuit synchronizes bending angle, acceleration, electromyographic potential, and temperature parameters into a unified flexible motion vector, which is then denoised, normalized, and transmitted to the control core. Based on this flexible motion vector, the intent modeling module uses a time-segment-based sequence analysis model to transform the continuity of the movement into a multidimensional intent distribution, forming a probability matrix of movement direction, speed trend, and force weights. This module's design allows the system to infer the operator's potential intent before the movement is completed, thereby reducing response latency.

[0050] The flexible-rigid mapping field module establishes a mapping function from the flexible space to the robotic arm space based on the principle of minimizing information entropy, maintaining the diversity and continuity of input signals through a probability distribution. The dynamic intent filtering module filters and sorts multiple candidate instructions by setting a confidence threshold and introducing historical execution deviation weights, ensuring that the output control instructions best match the operator's semantic intent. Finally, the execution control module converts the optimal instruction into a driveable signal, which is transmitted to the robotic arm's execution unit via the bus.

[0051] By introducing the concept of probabilistic semantic mapping, the error in the flexible-to-rigid conversion is reduced while the ambiguity of the action intention is preserved, thereby improving control accuracy and significantly reducing the error rate.

[0052] This solution further proposes that the intent modeling module includes:

[0053] The motion pattern solving unit is used to divide continuous flexible motion vectors into time window segments and extract posture feature sequences.

[0054] The intent probability generation unit is used to calculate the intent probability distribution of the action in the multidimensional semantic space based on the posture feature sequence, wherein each intent component includes direction probability, velocity trend probability and target intention probability.

[0055] The above design enables the system to predict the operator's potential intentions before the action is completed.

[0056] Establish a flexible detection and state space, acquire human motion signals, and establish a continuous flexible state space:

[0057] ;

[0058] in, For the flexible body's motion state space; The human posture vector acquired by the flexible detection unit at time i includes angle, acceleration, strain, or electromyographic signal parameters.

[0059] In this embodiment, the intent modeling module is responsible for converting raw action data into semantic intent probabilities. The system first divides the continuous action stream into fixed time windows through the action pattern solving unit, and extracts feature parameters such as posture vector, angular velocity, and electromyographic amplitude within each time window.

[0060] Subsequently, the intent probability generation unit calculates the action distribution in a multidimensional semantic space based on these features. This semantic space uses direction, velocity, and target inclination as the main coordinate axes, and establishes a one-dimensional probability density function on each coordinate axis. Together, these three elements form a multimodal intent representation of the action.

[0061] The system considers the continuity of attitude changes during the calculation process, enabling the model to predict potential intentions before the action is completed, thus achieving feedforward control.

[0062] This modeling approach avoids the delays and misjudgments caused by traditional threshold recognition, making the control of the robotic arm closer to the rhythm of human movements.

[0063] The establishment of a flexible state space allows each sampling point to correspond to a high-dimensional attitude vector, which contains angle, acceleration, strain and electromyographic components, thus forming a continuous and analyzable motion flow, providing a data foundation for subsequent probability mapping.

[0064] This solution further proposes that the flexible-rigid mapping field construction module includes:

[0065] Feature coupling unit, used to dynamically adjust the mapping weights between the flexible space and the rigid space of the robotic arm according to the intention probability distribution;

[0066] The minimum information entropy unit is used to select the mapping path with the least information loss while preserving the semantics of the operator's actions, based on the optimization criterion of information entropy change.

[0067] The instruction generation unit is used to output multiple candidate instructions for robotic arm control, along with corresponding confidence levels.

[0068] Based on the input flexibility state Generate multiple candidate actions And calculate its intention probability: ;

[0069] in: For input state Next action The probability of being judged as the operator's intention; Candidate actions The semantic cost function; β is a temperature parameter used to adjust the sharpness of the distribution; when β>0, the probability concentration increases.

[0070] The feature coupling unit couples the intent probability of the flexible input with the robotic arm's motion space, dynamically adjusting the mapping weights to ensure that different motion directions have different response sensitivities in the robotic arm space. The minimum information entropy unit calculates the change in information entropy of the mapping path, minimizing information loss to ensure that the semantic information of the flexible motion is preserved as much as possible when it is transmitted to the robotic arm.

[0071] For example, when an operator simultaneously intends to extend and rotate, the system determines the weights of both based on a probability distribution and generates a series of candidate actions, each accompanied by a confidence coefficient. The instruction generation unit converts these candidate actions into specific control parameters, such as end-effector pose or joint angles, and outputs them via a standard bus format.

[0072] This mapping strategy enables the robotic arm to exhibit composite motion characteristics similar to those of the human body with limited degrees of freedom, thereby solving the motion deviation problem caused by the mismatch of degrees of freedom in traditional systems.

[0073] This solution further proposes that the dynamic intent filtering module includes:

[0074] A confidence assessment unit is used to adaptively adjust the confidence threshold based on historical execution deviations.

[0075] Candidate action clustering unit is used to cluster control candidate instructions according to semantic similarity to reduce duplicate or conflicting actions;

[0076] The optimal selection unit is used to select the candidate instruction with the highest confidence among the cluster centers as the final execution instruction.

[0077] The system determines the optimal mapping by minimizing the semantic loss function:

[0078] ;

[0079] in: This represents the semantic loss value. For flexible input actions semantic embedding vector; Perform actions for the robotic arm semantic embedding vector; This is the Kullback–Leibler information divergence function, used to measure the information difference between two semantic distributions.

[0080] The confidence assessment unit adaptively adjusts the confidence threshold by comparing historical execution deviations with the current prediction error. When the system detects significant recent operational fluctuations, it automatically raises the threshold to avoid misjudgments. The candidate action clustering unit aggregates all candidate instructions based on semantic similarity and uses feature embedding distance to determine similarity, thereby avoiding the repeated execution of multiple similar actions.

[0081] The optimal selection unit is compared among cluster centers, and the candidate instruction with the highest confidence and the smallest historical deviation is selected as the execution instruction.

[0082] This process is equivalent to finding the path with the minimum semantic loss within the semantic distribution space. The semantic loss function uses Kullback-Leibler information divergence as a metric to minimize the semantic difference between the final selected action and the human intention.

[0083] Through this design, the system can establish an optimal semantic matching relationship between high-dimensional flexible input and low-dimensional rigid output, thereby significantly improving the accuracy of interaction.

[0084] This solution further proposes that the execution control module includes:

[0085] The signal processing unit is used to process the optimal execution instruction into drive signal parameters;

[0086] A closed-loop feedback unit is used to correct motion trajectory errors in real time based on feedback from the robotic arm's sensors;

[0087] The deviation adaptive unit is used to dynamically correct the mapped field parameters based on the feedback error, thereby achieving long-term adaptive control of the action.

[0088] The probabilistic reasoning and execution module selects the final robotic arm action to be executed based on the confidence level:

[0089] ;

[0090] in,

[0091] in, Dynamically adjusted to an uncertain function:

[0092] ;

[0093] in, This is a dynamic function calculated based on the stability and autocorrelation of the input signal.

[0094] The execution control module is responsible for converting the selected action into low-level instructions that the robotic arm can execute. The signal processing unit, based on the robotic arm's kinematic model, resolves the semantic instructions into joint drive parameters or end-effector pose targets. During the processing, link length, joint limits, and load constraints are considered to ensure that the output instructions are physically feasible.

[0095] The closed-loop feedback unit monitors the robot arm's pose, torque, and end effector force in real time. By comparing these parameters with the desired trajectory, it generates an error term, which is then fed back to the mapping field to correct subsequent control parameters. The deviation adaptive unit gradually adjusts the mapping parameters over extended operation, enabling the system to adapt to the operator's movement style.

[0096] The confidence-driven mechanism in this module enables the system to automatically select the action with the highest probability from multiple candidate instructions for execution, and controls the response sensitivity through a dynamic threshold function, thereby improving execution efficiency while ensuring safety.

[0097] This solution further proposes that the flexible motion sensing acquisition module includes:

[0098] The multimodal sensing unit includes a flexible bending sensor, an inertial measurement unit, an electromyography signal acquisition unit, and a temperature sensor.

[0099] The data fusion unit is used to perform time synchronization and normalization processing on multi-source data and output a unified flexible motion vector.

[0100] The robotic arm actuator performs the determined optimal action. Execute the corresponding rigid motion and update the mapping field parameters and intent distribution through the feedback module to achieve a closed-loop interaction:

[0101] ;

[0102] in, Map the field parameters at the current moment; For the updated mapping field parameters; To update the step size factor; The gradient of semantic loss with respect to the parameters of the mapped field.

[0103] The flexible motion sensing module achieves high-precision acquisition of motion signals through a multimodal sensing unit. A flexible bending sensor measures changes in limb angle; an inertial measurement unit detects acceleration and angular velocity; an electromyography (EMG) signal acquisition unit senses the intensity of muscle electrical activity to infer force application trends; and a temperature sensor compensates for measurement offsets caused by differences in ambient and skin temperature. The data fusion unit synchronizes multi-source data in the time domain, normalizes amplitudes, and generates motion vectors in a unified format.

[0104] After obtaining the optimal action, the robotic arm actuator completes the corresponding pose adjustment according to the control signal and feeds back the execution status to the main control module.

[0105] The system fine-tunes the mapping field parameters through the gradient of the semantic loss function, achieving an adaptive interactive closed loop that allows the model to continuously optimize mapping accuracy across multiple rounds of operation. This multimodal fusion acquisition method effectively suppresses the uncertainty caused by noise from a single sensor channel, significantly improving the stability of action recognition.

[0106] This solution further proposes that the system also includes an intent confirmation module, which is used to confirm with the operator through visual or voice means when the confidence level is lower than a set threshold, so as to prevent erroneous actions.

[0107] To prevent accidental operations under low confidence levels, the system includes an intent confirmation module. When the dynamic intent filtering module detects that the highest confidence level is below a threshold, the system pauses command execution and prompts the operator for confirmation via visual indicators, voice announcements, or tactile vibrations. The confirmation module communicates bidirectionally with the main control core, allowing the operator to provide confirmation feedback through gestures or voice responses.

[0108] This secondary confirmation mechanism effectively prevents accidental triggering in complex working conditions, and is especially suitable for high-risk scenarios such as medical assistance and precision assembly.

[0109] This solution further proposes that the robotic arm body includes:

[0110] The drive execution unit is used to receive drive signals and execute corresponding joint movements;

[0111] The state sensing unit is used to detect the robot arm's pose, torque, and end-effector force information in real time, and to feed the state back to the execution control module.

[0112] The feedback communication unit is used to enable bidirectional data transmission between the robotic arm and the control module.

[0113] The robotic arm body, as the system execution layer, mainly consists of a drive execution unit, a state perception unit, and a feedback communication unit.

[0114] The drive actuator employs a servo motor or electric cylinder structure to achieve independent drive and coordinated control of each joint. The state sensing unit is equipped with an angle encoder, torque sensor, and a six-dimensional force sensor at the end effector to detect joint load and external interference.

[0115] The feedback communication unit communicates bidirectionally with the main control module via CAN or EtherCAT bus to achieve high-frequency data interaction and security monitoring.

[0116] This robotic arm can be configured with six to eight degrees of freedom depending on the task requirements and can perform compliant control in space. By coupling with a somatosensory system, it can be used in various scenarios such as human-machine collaborative assembly, rehabilitation training, or remote operation.

[0117] Compared with traditional rigid controllers, the robotic arm in this system can respond more naturally to the operator's slight posture changes, improving the accuracy and controllability of the operation.

[0118] Example 2

[0119] The flexible haptic interaction method based on intent probability mapping includes the following steps:

[0120] Step S1: Multimodal somatosensory signal acquisition. Through a flexible somatosensory acquisition module worn on the operator's arm and wrist, multimodal data including bending sensing signals, inertial signals, electromyographic signals, and temperature signals are acquired. The multimodal data is then time-synchronized and normalized to obtain flexible motion vectors for subsequent modeling.

[0121] Step S2: Model the intent probability distribution. Divide the flexible action vector into time windows and extract the posture feature sequence. Calculate the intent probability distribution of the action in the semantic action space based on the posture feature sequence. The intent probability distribution includes direction probability, velocity trend probability, and target intention probability. Output a distribution matrix containing multiple candidate intents as input for subsequent mapping.

[0122] Step S3: Construct a flexible-rigid mapping field to establish a mapping field that describes the correspondence between the flexible motion space and the rigid motion space of the robotic arm; adjust the mapping weights of the flexible space and the rigid space according to the intention probability distribution, and calculate the information entropy change of each candidate mapping path using the information retention criterion; select the mapping path with the minimum information loss, and generate multiple candidate control commands for the robotic arm with corresponding confidence levels.

[0123] Step S4: Dynamic intent filtering, performing semantic clustering on the candidate control commands, aggregating similar action commands into candidate clusters; calculating the comprehensive confidence of each candidate cluster, and determining the final execution command based on an adaptive threshold; if the confidence is lower than a preset threshold, triggering a visual or voice confirmation request to prevent erroneous actions.

[0124] Step S5: Execution and closed-loop correction. The final execution command is calculated into robotic arm drive signal parameters to control each joint of the robotic arm to perform corresponding actions; feedback signals of robotic arm pose, torque and end force are collected; the mapping field parameters are dynamically corrected according to the feedback error to achieve long-term adaptive control of the robotic arm's movements.

[0125] Step S6: Continuous Intent Update. The execution results are compared with historical action features to dynamically update the intent probability model; this allows the system to gradually optimize the matching accuracy of the flexible-rigid mapping in repetitive tasks, improving the accuracy of intent interpretation and the stability of control.

[0126] Through the above steps, the robotic arm motion-sensing interaction method can establish a continuous-discrete fusion mapping relationship between flexible human body movements and rigid robotic arm movements, thereby achieving accurate recognition and low-latency control execution of multi-intent commands.

[0127] This scheme further proposes that, in step S3, the information retention criterion is specifically as follows: by calculating the mutual information gain between the flexible motion space and the robotic arm motion space, the dynamic weight distribution of the mapping field is determined, so that the selected mapping path minimizes the motion semantic loss rate while maintaining the consistency of motion semantics.

[0128] During the mapping field construction process, the realization of the information preservation criterion relies on the calculation of mutual information gain. The system first establishes a joint distribution model of the flexible motion space and the robotic arm motion space, and calculates the mutual information value between them. The gain of mutual information reflects the amount of information that can be transmitted to the robotic arm control layer in the flexible motion. The mapping weights are dynamically adjusted based on this mutual information: when the input motion contains more semantic features, the system automatically increases the weight of the corresponding mapping channel; conversely, it decreases it.

[0129] This self-regulating mechanism ensures that semantic loss is minimized while maintaining semantic consistency of actions.

[0130] Compared with traditional linear mapping methods, this method can achieve nonlinear multi-intent fusion in complex, multi-degree-of-freedom scenarios, making the robotic arm response more robust and human action more coherent.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation methods of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications should be covered within the scope of the technical solutions claimed in the present invention.

Claims

1. A flexible haptic interaction system based on intent probability mapping, characterized in that, include: The flexible motion sensing acquisition module is used to acquire multimodal motion information of the operator's limbs, including joint bending angles, postural acceleration and electromyographic signals, and convert the information into standardized flexible motion vectors; The intention modeling module is used to perform continuous space analysis on the flexible motion vector and establish an intention probability distribution model to represent the operator's potential control intention under multi-degree-of-freedom motion, wherein the intention probability distribution includes target direction, force feeling trend and motion confidence. The flexible-rigid mapping field construction module is used to construct a nonlinear mapping field between the flexible human body action space and the rigid action space of the robotic arm based on the intention probability distribution. The mapping field maps the continuously deforming flexible action into a discrete set of robotic arm control commands by maintaining the principle of minimizing the change in information entropy. The dynamic intent filtering module is used to dynamically filter and sort candidate control instructions in the set of robotic arm control instructions based on action confidence threshold and historical execution deviation, and determine the optimal execution instruction. The execution control module is used to convert the optimal execution command into a drive signal that the robotic arm can execute, and to perform closed-loop correction based on real-time motion feedback; The robotic arm body is used to execute the drive signals output by the execution control module.

2. The flexible haptic interaction system based on intent probability mapping according to claim 1, characterized in that, The intent modeling module includes: The motion pattern solving unit is used to divide continuous flexible motion vectors into time window segments and extract posture feature sequences. The intent probability generation unit is used to calculate the intent probability distribution of the action in the multidimensional semantic space based on the posture feature sequence, wherein each intent component includes direction probability, velocity trend probability and target intention probability.

3. The flexible haptic interaction system based on intent probability mapping according to claim 1, characterized in that, The flexible-rigid mapping field construction module includes: Feature coupling unit, used to dynamically adjust the mapping weights between the flexible space and the rigid space of the robotic arm according to the intention probability distribution; The minimum information entropy unit is used to select the mapping path with the least information loss while preserving the semantics of the operator's actions, based on the optimization criterion of information entropy change. The instruction generation unit is used to output multiple candidate instructions for robotic arm control, along with corresponding confidence levels.

4. The flexible haptic interaction system based on intent probability mapping according to claim 1, characterized in that, The dynamic intent filtering module includes: A confidence assessment unit is used to adaptively adjust the confidence threshold based on historical execution deviations. Candidate action clustering unit, used to cluster control candidate instructions based on semantic similarity; The optimal selection unit is used to select the candidate instruction with the highest confidence among the cluster centers as the final execution instruction.

5. The flexible haptic interaction system based on intent probability mapping according to claim 1, characterized in that, The execution control module includes: The signal processing unit is used to process the optimal execution instruction into drive signal parameters; A closed-loop feedback unit is used to correct motion trajectory errors in real time based on feedback from the robotic arm's sensors; The bias adaptive unit is used to dynamically correct the mapping field parameters based on the feedback error.

6. The flexible haptic interaction system based on intent probability mapping according to claim 1, characterized in that, The flexible motion sensing acquisition module includes: The multimodal sensing unit includes a flexible bending sensor, an inertial measurement unit, an electromyography signal acquisition unit, and a temperature sensor. The data fusion unit is used to perform time synchronization and normalization processing on multi-source data and output a unified flexible motion vector.

7. The flexible haptic interaction system based on intent probability mapping according to claim 1, characterized in that, The system also includes an intent confirmation module, which is used to confirm with the operator through visual or voice means when the confidence level is lower than a set threshold.

8. The flexible haptic interaction system based on intent probability mapping according to claim 1, characterized in that, The robotic arm body includes: The state perception unit is used to detect the robot arm's pose, torque, and end-effector force information in real time, and to feed the state back to the execution control module. The feedback communication unit is used to enable bidirectional data transmission between the robotic arm and the control module.

9. A flexible haptic interaction method based on intent probability mapping, characterized in that, Includes the following steps: Step S1: Multimodal somatosensory signal acquisition. Through a flexible somatosensory acquisition module worn on the operator's arm and wrist, multimodal data including bending sensing signals, inertial signals, electromyographic signals, and temperature signals are acquired. The multimodal data is then time-synchronized and normalized to obtain flexible motion vectors for subsequent modeling. Step S2: Model the intent probability distribution, divide the flexible action vector into time windows, and extract the posture feature sequence; calculate the intent probability distribution of the action in the semantic action space based on the posture feature sequence, the intent probability distribution includes direction probability, velocity trend probability and target intention probability; The output contains a distribution matrix of multiple candidate intentions, which serves as the input for subsequent mappings; Step S3: Construct a flexible-rigid mapping field to establish a mapping field that describes the correspondence between the flexible motion space and the rigid motion space of the robotic arm; adjust the mapping weights of the flexible space and the rigid space according to the intention probability distribution, and calculate the information entropy change of each candidate mapping path using the information retention criterion; select the mapping path with the minimum information loss, and generate multiple candidate control commands for the robotic arm with corresponding confidence levels. Step S4: Dynamic intent filtering, performing semantic clustering on the candidate control instructions, aggregating similar action instructions into candidate clusters; calculating the comprehensive confidence of each candidate cluster, and determining the final execution instruction based on an adaptive threshold; If the confidence level is lower than a preset threshold, a visual or voice confirmation request will be triggered. Step S5: Execution and closed-loop correction. The final execution command is calculated into robotic arm drive signal parameters to control each joint of the robotic arm to perform corresponding actions; feedback signals of robotic arm pose, torque and end-effector force are collected. The mapping field parameters are dynamically corrected based on the feedback error; Step S6: Continuously update the intent, compare the execution result with historical action features, and dynamically update the intent probability model; gradually optimize the matching accuracy of the flexible-rigid mapping in repetitive tasks.

10. The flexible haptic interaction method based on intent probability mapping according to claim 9, characterized in that, In step S3, the information retention criterion is specifically: the dynamic weight distribution of the mapping field is determined by calculating the mutual information gain between the flexible motion space and the robotic arm motion space.

Images