A humanoid robot caressing interaction system and control method

Abstract

This invention discloses a humanoid robot touch interaction system and control method, belonging to the field of robot intelligent control technology. In this invention, a flexible bionic hand incorporates a pressure sensor array, a low-frequency vibration module is integrated inside the hand, and a multimodal perception unit includes a depth camera, a pressure sensor, and a joint posture sensor. The control unit automatically identifies and maps the touch area based on visual data. A force-position dual closed-loop constant force control is employed to stabilize the contact pressure within a preset range. Simultaneously, the vibration frequency and amplitude are dynamically and collaboratively adjusted according to the touch speed and trajectory, and a safety shutdown is executed when the pressure exceeds the limit or the posture is abnormal. This invention achieves gentle and stable touch interaction through a flexible bionic hand and constant force control, reduces human intervention through automatic visual positioning, enriches the interactive experience through vibration coordination, and possesses a comprehensive safety protection mechanism, solving the problems of poor consistency in human touch, discomfort from rigid contact in traditional robots, and low automation levels.

Description

Technical Field

[0001] This invention relates to the field of robot intelligent control technology, specifically to a humanoid robot touch interaction system and control method. Background Technology

[0002] Tactile stimulation is an important non-pharmacological soothing technique widely used in rehabilitation nursing, emotional calming, muscle relaxation, and adjunctive physical therapy. Currently, tactile stimulation is mainly performed manually by professional caregivers. Its technical characteristics lie in the operator using their palms and fingers to perform pushing, kneading, and pressing movements on the recipient's body surface with specific force, frequency, and trajectory. Tactile signals are transmitted through the skin's mechanoreceptors, achieving physiological and psychological effects such as muscle relaxation and tension relief.

[0003] However, manual touch therapy has several inherent limitations. First, there are significant individual differences in the strength control, rhythm, and trajectory consistency among different operators. Even the same operator is prone to fatigue-induced strength reduction or trajectory disorder after prolonged work, making it difficult to standardize and reproducibly verify the effects of touch therapy. Second, the training period for highly skilled touch therapists is long and the labor costs are high, and the need for large-scale, long-term continuous operation cannot be met, which limits the widespread application of touch therapy.

[0004] With the development of robotics technology, some research has attempted to apply robotic arms to human body surface manipulation, such as massage robots and rehabilitation training robots. In existing technologies, massage robots typically employ rigid end effectors, relying on position control or simple force-position switching to achieve pressing actions. However, this approach has significant shortcomings when applied to tactile interaction. On the one hand, rigid mechanical structures lack flexibility when contacting soft human tissue, easily generating a sense of impact or pressure discomfort, making it difficult to simulate the tactile experience of gentle human touch. On the other hand, existing systems mostly rely on manual teaching or offline programming to set the work trajectory, lacking the ability to automatically recognize and adaptively position themselves according to the human body surface morphology. When the body shape of the person being touched changes or moves slightly, the contact point easily deviates from the target area, resulting in a low degree of automation. Therefore, this invention proposes a humanoid robot touch interaction system and control method. Summary of the Invention

[0005] The purpose of this invention is to provide a humanoid robot touch interaction system and control method to solve the problems mentioned in the background art.

[0006] According to a first aspect of the present invention, in order to achieve the above-mentioned objective, the present invention provides the following technical solution: a humanoid robot touch interaction system, comprising: The humanoid robot itself; A flexible bionic hand actuator is installed on the humanoid robot body and includes multiple degrees of freedom joints, a flexible covering layer covering the outside of the joints, and a pressure sensing array built into the flexible covering layer. A low-frequency vibration module, integrated inside the flexible bionic hand actuator, is used to output mechanical vibrations with adjustable frequency and amplitude. The multimodal sensing unit includes at least a visual positioning sensor, a joint posture sensor, and the pressure sensing array; The control unit is connected to the flexible bionic hand actuator, the low-frequency vibration module, and the multimodal sensing unit, respectively. The control unit is configured as follows: Receive human image data collected by the visual positioning sensor, identify the human touch area and locate key points, establish the mapping relationship between the robot coordinate system and the human surface coordinates, and generate and output the spatial coordinates of the target touch area. The system receives the spatial coordinates of the target touch area, drives the flexible bionic hand actuator to approach the target area, and receives the contact pressure signal fed back in real time by the pressure sensor array. It then uses a compliant control algorithm to perform closed-loop adjustment of the touch force, outputs joint driving torque, and stabilizes the contact pressure within a preset force range. The system receives the preset touch trajectory and the spatial coordinates of the target touch area, drives the flexible bionic hand actuator to complete the touch path movement, and synchronously outputs vibration control signals to the low-frequency vibration module to adjust the vibration parameters according to the current touch action stage and movement speed. It receives pressure signals from the pressure sensor array and attitude signals from the joint attitude sensor in real time, and outputs a safety shutdown command when the pressure exceeds the limit or the attitude is abnormal.

[0007] Furthermore, the flexible coating layer is made of flexible silicone material, and the detection range of the pressure sensing array is 0–20 N.

[0008] Furthermore, the frequency adjustment range of the low-frequency vibration module is 20Hz to 200Hz, and the amplitude adjustment range is 0.1mm to 3mm.

[0009] Furthermore, the visual positioning sensor uses an Intel RealSense D435 depth camera to capture human body contours, achieve region segmentation and key point recognition, with an effective depth range of 0.2m to 10m, an RGB resolution of 1920×1080, and a depth resolution of 1280×720.

[0010] Furthermore, the joint posture sensor includes an AS5047P magnetic encoder installed in each joint for real-time acquisition of joint rotation angles. The end effector integrates an MPU6050 six-axis IMU for monitoring the overall hand posture.

[0011] Furthermore, the pressure sensing array employs FlexiForce A201 ultra-thin film pressure sensors, which are distributed at multiple points under the flexible covering layer of the bionic hand. The control unit polls each sensing point at a sampling rate of not less than 100Hz to obtain contact pressure distribution data for constant force closed-loop control and overpressure safety protection.

[0012] Furthermore, the control unit is an embedded processor, specifically including: a microcontroller (MCU) or a digital signal processor (DSP).

[0013] According to a second aspect of the present invention, a control method for a humanoid robot touch interaction system is provided, which applies the humanoid robot touch interaction system described in the first aspect and includes the following steps: S1: Collect images of the human body surface through a vision sensor, perform region segmentation and key point localization, establish a mapping relationship between the robot coordinate system and the human body surface coordinates, and output the spatial coordinates of the target touch area; S2: Receive the spatial coordinates of the target touch area output by S1, and preset the touch trajectory mode, touch force range, vibration frequency range and amplitude range; S3: Using the touch trajectory pattern preset in S2 and the spatial coordinates of the target touch area as input, the robot joint movement is controlled by inverse kinematics solution to drive the flexible bionic hand to approach the target area, and the contact pressure is collected in real time through the pressure sensor array and the contact pressure signal is output. S4: Receive the contact pressure signal output by S3 and the touch force range preset by S2, calculate the pressure deviation, use PID control or impedance control algorithm to perform force feedback and position feedback dual closed-loop adjustment, output joint driving torque, stabilize the contact pressure in the preset touch force range, and realize constant force touch. S5: During the constant force stroking process described in S4, the current stroking trajectory position and movement speed are received, and the vibration frequency and amplitude are dynamically adjusted according to the mapping relationship between movement speed and vibration parameters. The vibration control signal is output to the low-frequency vibration module to realize the coordinated control of stroking motion and low-frequency vibration. S6: During the execution of S4 and S5, the contact pressure signal and joint posture signal are received in real time. When the pressure exceeds the threshold, the movement posture is abnormal, or the contact is lost, a deceleration and stop command is output to automatically decelerate and stop the action.

[0014] Furthermore, by acquiring images of the human body surface through a visual sensor, region segmentation and key point localization are performed to establish a mapping relationship between the robot coordinate system and the human body surface coordinates, and the spatial coordinates of the target touch area are output, specifically including: Human RGB-D images are acquired using a depth camera, and the three-dimensional spatial coordinates of multiple joints in the human body are extracted using a skeletal key point detection model. The extracted joint points are matched with predefined touch regions in the key point database. Each touch region is associated with a specific set of joint points and region boundary rules constructed from the geometry of the joint points. Based on the actual coordinates of the matched joints, the boundary of the touching area and the sequence of internal target points are dynamically calculated and converted to the robot's base coordinate system; When a match fails or the confidence level of a key point is below a threshold, region segmentation based on depth images or a preset offset is used as a backup localization method.

[0015] Furthermore, the preset touch trajectory includes linear pushing, circular kneading, and fixed-point holding.

[0016] The present invention has at least the following beneficial effects: 1. This invention forms a flexible contact with the human body surface through a flexible bionic hand actuator. By utilizing the passive compliance properties of the flexible silicone coating layer, the rigid movement of the robot joint is transformed into a soft surface fit, which significantly reduces the contact impact force and local pressure concentration, avoids discomfort or damage to human tissue, and improves the safety and comfort of the human-computer interaction process.

[0017] 2. This invention adopts a dual closed-loop control structure of force feedback and position feedback. The contact pressure detected in real time by the pressure sensor array is used as the feedback signal. The joint driving torque is continuously adjusted through a compliant control algorithm to keep the contact pressure stable within a preset range. This overcomes the problems of uneven force and drift in manual operation and realizes standardized constant force massage that is quantifiable and reproducible.

[0018] 3. This invention acquires human images through a visual positioning sensor and performs region segmentation and key point localization, establishing a mapping relationship between the robot coordinate system and the human body surface coordinates. This enables the system to automatically identify the target touch area and generate spatial coordinates, reducing manual teaching intervention, improving the autonomy and continuity of the work process, and adapting to the body shape differences of different individuals.

[0019] 4. This invention integrates a low-frequency vibration module into the interior of a flexible bionic hand, and dynamically adjusts the vibration frequency and amplitude according to the current stage of the stroking action and the movement speed, so as to achieve deep synergy between vibration output and mechanical movement, enrich the sense of hierarchy of the interaction mode, and improve the uniformity of stroking and the effect of tissue penetration.

[0020] 5. This invention simultaneously collects contact pressure and joint posture information through a multimodal sensing unit, triggering a safety shutdown logic when the pressure exceeds the limit or the posture is abnormal, forming a rapid protection closed loop from sensing to response, avoiding mechanical damage caused by overpressure impact and abnormal posture, and enhancing the practical reliability of the system.

[0021] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the system described in this invention; Figure 2 This is a flowchart illustrating the control method described in this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Example 1: Please see Figure 1 This invention provides a technical solution: a humanoid robot touch interaction system, comprising: The humanoid robot itself; The flexible bionic hand actuator is installed on the humanoid robot body and includes multiple degrees of freedom joints, a flexible covering layer covering the outside of the joints, and a pressure sensing array built into the flexible covering layer. The flexible overlay layer uses flexible materials such as silicone, giving the hand natural compliance when in contact with the human body. It can adapt to the contours of the human body, transforming rigid contact into a flexible fit. This significantly reduces contact impact and local pressure concentration, avoiding discomfort or damage to human tissue and improving the safety and comfort of the interaction process. The built-in pressure sensor array is directly embedded in the flexible overlay layer, forming an integrated structure of sensing and execution. This allows the pressure detection points to be close to the actual contact surface, enabling direct and real-time perception of the mechanical distribution of the contact interface, providing high-fidelity raw data for subsequent force closed-loop control. Multiple degrees of freedom joints give the hand flexible movement capabilities. Combined with the passive compliance characteristics of the flexible overlay layer, it forms a composite structure of active drive and passive compliance. It can actively complete complex movements such as pushing and kneading according to a planned trajectory, and can also passively adapt to the curvature changes of the human body surface during contact, achieving a combination of rigidity and flexibility in the interaction. The low-frequency vibration module, integrated within the flexible bionic hand actuator, outputs mechanical vibrations with adjustable frequency and amplitude. This integrated design eliminates the need for additional installation space and mechanical connections required by external vibration devices, resulting in a compact and lightweight end effector that reduces the load on the arm joints. This facilitates high-speed, flexible trajectory movements and simplifies system assembly and maintenance. Furthermore, the independently adjustable vibration parameters allow the vibration output to dynamically adapt to the type, speed, and human feedback of the current stroking action. For example, the vibration amplitude can be reduced during rapid stroking to avoid excessive stimulation, while vibration can be increased during point holding or slow kneading to promote blood circulation. This achieves deep synergy between vibration and movement, enriching the layers of the interaction mode. The multimodal perception unit includes at least a visual positioning sensor, a joint posture sensor, and a pressure sensor array. The visual positioning sensor undertakes the task of global spatial perception, and can pre-complete contour acquisition, region segmentation, and key point localization without contacting the human body, mapping the target area from the human image space to the robot coordinate space. The joint posture sensor provides real-time feedback on the angles of each joint of the robotic arm and the spatial posture of the end effector, enabling the control unit to accurately grasp the actual movement state of the hand and cross-validate with the visual positioning results to promptly detect posture abnormalities caused by mechanical deformation, load disturbance, or trajectory deviation. The pressure sensor array focuses on the micro-mechanical perception of the contact interface, converting the contact force information that the visual and posture sensors cannot directly obtain into the feedback signal required for closed-loop control. When the three work together, vision is responsible for target guidance, posture is responsible for motion monitoring, and pressure is responsible for contact adjustment, forming a hierarchical perception chain from macro-positioning to micro-force control. The control unit is connected to the flexible bionic hand actuator, the low-frequency vibration module, and the multimodal sensing unit, respectively. The control unit is configured as follows: It receives human image data collected by a visual positioning sensor, identifies the human touch area and locates key points, establishes a mapping relationship between the robot coordinate system and the human surface coordinates, and generates and outputs the spatial coordinates of the target touch area. The system receives the spatial coordinates of the target touch area, drives the flexible bionic hand actuator to approach the target area, and receives the contact pressure signal fed back in real time by the pressure sensor array. It uses a compliant control algorithm to perform closed-loop adjustment of the touch force and outputs joint driving torque to stabilize the contact pressure within the preset force range. It receives the preset stroking trajectory and the spatial coordinates of the target stroking area, drives the flexible bionic hand actuator to complete the stroking path movement, and simultaneously outputs vibration control signals to the low-frequency vibration module to adjust the vibration parameters according to the current stroking action stage and movement speed. It receives pressure signals from the pressure sensor array and attitude signals from the joint attitude sensor in real time, and outputs a safety shutdown command when the pressure exceeds the limit or the attitude is abnormal.

[0026] Regarding the technical solution of this embodiment, the flexible covering layer uses flexible silicone material. Flexible silicone material has excellent biocompatibility and skin affinity. Its texture is soft and elastic. When in contact with the human body surface, it can adapt to the body contour and deform accordingly, transforming the rigid movement of the robot joints into a soft surface fit. This significantly reduces the mechanical impact force at the moment of contact and avoids causing local pressure or abrasion to human tissue. The detection range of the pressure sensor array is 0 to 20 N. Setting the detection range of the pressure sensor array to 0 to 20 Newtons precisely covers the conventional force range required for human touch interaction. It can effectively detect everything from very slight touch to moderate pressure.

[0027] Regarding the technical solution of this embodiment, the frequency adjustment range of the low-frequency vibration module is 20Hz to 200Hz, and the amplitude adjustment range is 0.1mm to 3mm. This frequency range extends from the upper edge of the infrasound band to the mid-low frequency vibration zone, which precisely covers the low-frequency vibration range where the mechanoreceptors of human skin are most sensitive. It can effectively activate the tactile corpuscles and ring corpuscles, producing a clear and comfortable vibration sensation, and avoiding tingling or numbness caused by excessively high frequencies.

[0028] For the technical solution of this embodiment, the visual positioning sensor adopts an Intel RealSense D435 depth camera to acquire human body contours, achieve region segmentation and key point recognition. Its effective depth range is 0.2m to 10m, with an RGB resolution of 1920×1080 and a depth resolution of 1280×720. The effective depth range extends from very close distance to a relatively long distance, which can meet the needs of the robot hand for high-precision depth information at close range when it approaches the human body surface, ensuring accurate obstacle avoidance and positioning before contact, and can also cover the complete human body contour acquisition, realizing full-body or half-body region segmentation and key point detection without frequent movement of the robot base.

[0029] Regarding the technical solution of this embodiment, the joint posture sensor includes an AS5047P magnetic encoder installed in each joint for real-time acquisition of joint rotation angles. Each joint is equipped with an independent magnetic encoder, which can directly provide the true rotation angle for each degree of freedom. The control unit does not need to indirectly calculate or compensate for transmission backlash through the motor encoder, making the inverse kinematics solution more accurate, thus reliably ensuring the positioning accuracy of the end effector. The end effector integrates an MPU6050 six-axis IMU for monitoring the overall hand posture. The joint encoder reflects the positional relationship of each segment of the robotic arm, while the end effector independently senses the absolute posture of the hand. The two sets of data can be cross-validated: when an inconsistency is found between the end effector posture calculated from the joint angles and the posture measured by the IMU, the system can determine that there is an external collision, joint slippage, or sensor malfunction, and immediately trigger the safety protection logic to effectively prevent accidental actions from injuring the user.

[0030] Regarding the technical solution of this embodiment, the pressure sensing array adopts FlexiForce A201 ultra-thin film pressure sensor, which is arranged at multiple points under the flexible covering layer of the bionic hand. The control unit polls each sensing point at a sampling rate of not less than 100Hz to obtain contact pressure distribution data for constant force closed-loop control and overpressure safety protection. The sensor is extremely thin and has flexible and bendable characteristics. After being embedded under the flexible silicone covering layer of the bionic hand, it hardly changes the original soft touch and deformation ability of the hand. It will not produce local hard points or foreign body sensation due to the addition of sensing elements, so that the person being touched still gets a soft experience close to the touch of a human hand. By deploying sensors at multiple points on the palm and fingertips of the bionic hand, the system no longer relies on a single pressure reading but can instead sense the pressure distribution across the entire contact surface. This helps determine whether the robot hand is evenly conforming to the human body surface, avoiding excessive local pressure due to angular deviations while other areas are not properly aligned, thus guiding the control unit to adjust the posture to achieve uniform force distribution.

[0031] In this embodiment, the control unit is an embedded processor, specifically including a microcontroller (MCU) or a digital signal processor (DSP). Both the MCU and DSP are designed for real-time control tasks, with extremely short interrupt response latency. They can complete sensor data acquisition, force closed-loop calculation, motor drive updates, and vibration parameter adjustment in a defined time sequence, ensuring smooth and continuous stroking actions. Furthermore, the DSP has a hardware multiplier and parallel processing architecture, suitable for rapidly executing mathematical operations in PID control, filtering, attitude calculation, and vibration coordination algorithms. The MCU can efficiently perform task scheduling and communication processing. Both ensure smooth multi-sensor data fusion and real-time control.

[0032] Example 2: This invention provides a control method for a humanoid robot touch interaction system, using the humanoid robot touch interaction system described in the first aspect, comprising the following steps: S1: Collect images of the human body surface through a vision sensor, perform region segmentation and key point localization, establish a mapping relationship between the robot coordinate system and the human body surface coordinates, and output the spatial coordinates of the target touch area; Human RGB-D images are acquired using a depth camera, and the three-dimensional spatial coordinates of multiple joints in the human body are extracted using a skeletal key point detection model. The extracted joint points are matched with predefined touch regions in the key point database. Each touch region is associated with a specific set of joint points and region boundary rules constructed from the geometry of the joint points. Based on the actual coordinates of the matched joints, the boundary of the touching area and the sequence of internal target points are dynamically calculated and converted to the robot's base coordinate system; When matching fails or the confidence of key points is below the threshold, depth image-based region segmentation or preset offset is used as a backup localization method. S2: Receives the spatial coordinates of the target touch area output by S1, and presets the touch trajectory mode, touch force range, vibration frequency range and amplitude range; S2.1 Receives coordinates output by S1 Receive the target touch area spatial coordinate set output by step S1. This coordinate set contains the three-dimensional spatial location information of each key point in the area, which serves as the spatial reference for subsequent trajectory planning and parameter configuration. S2.2 Geometric Features of the Analytical Region Using the spatial coordinate set output by S1 as input, the surface curvature, normal vector distribution, and area of ​​the target touch area are calculated. Specifically, adjacent key points within the area are selected to form local triangular patches. The normal vectors of each patch are obtained through cross product operation, and then the average normal vector of the area is obtained through weighted averaging. The local quadratic surface is fitted using multi-point coordinates to estimate the principal curvature of the surface, determine the flatness or curvature of the area, and output the geometric feature descriptor of the area. S2.3 Matching touch trajectory mode The region geometric feature descriptor output by S2.2 is input into the trajectory pattern database, and pattern matching is performed based on curvature and area: if the region has small curvature and large area, the straight pushing pattern is matched to generate an equally spaced straight line path parallel to the long axis of the region; if the region has significant curvature changes and has a circular topology, the circular kneading pattern is matched to generate a spiral or concentric circle path around the center point of the region; if the region has a small area and is located at a sensitive position such as a joint or acupoint, the fixed point holding pattern is matched to generate a single-point hovering path. Output the selected trajectory pattern identifier and the corresponding sequence of planned path points; S2.4 Set the range of stroking intensity Using the trajectory mode identifier output by S2.3 as input, the force range mapping table is queried: the straight pushing and stroking mode corresponds to the lower force range, the circular kneading and pressing mode corresponds to the medium force range, and the fixed-point holding mode corresponds to the higher force range. At the same time, the depth information along the normal vector direction of the spatial coordinate set output by S2.1 is received. If the soft tissue on the surface of the region is thick, the lower and upper limits of the force range are adjusted upward; if the region is close to bony prominences, the upper limit of the force range is adjusted downward, and the adapted stroking force range is output. S2.5 Set the vibration frequency range and amplitude range Using the trajectory mode identifier output by S2.3 and the stroking force range output by S2.4 as joint inputs, the vibration parameter mapping table is queried: For linear stroking mode with low force, a lower frequency range and a smaller amplitude range are set; for circular kneading mode with medium force, a medium frequency range and a medium amplitude range are set; for fixed-point holding mode with high force, a higher frequency range and a larger amplitude range are set. Simultaneously, the area curvature output by S2.2 is used as a correction factor; if the curvature is large, the upper limit of the amplitude is lowered to avoid excessive local stimulation. The adapted vibration frequency range and amplitude range are then output. S2.6 Parameter Integration and Output Loading The trajectory pattern identifier and planned path point sequence output by S2.3, the touch force range output by S2.4, and the vibration frequency range and amplitude range output by S2.5 are integrated to form a complete touch task parameter package. This parameter package is loaded into the control unit cache and output to step S3 as the input command for robot approach motion and contact control. S3: Using the preset touch trajectory pattern and the spatial coordinates of the target touch area as input, the robot joint movement is controlled by inverse kinematics solution to drive the flexible bionic hand to approach the target area, and the contact pressure is collected in real time through the pressure sensor array and the contact pressure signal is output. S4: Receives the contact pressure signal output by S3 and the touch force range preset by S2, calculates the pressure deviation, and uses PID control or impedance control algorithm to perform force feedback and position feedback dual closed-loop adjustment, outputs joint driving torque, stabilizes the contact pressure within the preset touch force range, and achieves constant force touch. The system receives the contact pressure signal output from step S3. This signal is obtained by real-time sampling and filtering of the pressure sensor array, representing the actual contact pressure value at the current moment. Simultaneously, it receives the stroking force range output from step S2, and takes the midpoint of the range as the target reference pressure value for constant force control.

[0033] S4.2 Calculate pressure deviation Using the actual contact pressure and the target reference pressure as inputs, the difference between the two is calculated to obtain the pressure deviation. This deviation value is the core input for subsequent force closed-loop control. A positive deviation indicates that the actual pressure is insufficient and the contact force needs to be increased, while a negative deviation indicates that the actual pressure is too strong and the contact force needs to be decreased.

[0034] S4.3 Force Feedback Closed-Loop Calculation Using pressure deviation as input, proportional-integral-differential (PID) calculations are performed. The proportional term outputs a correction amount in real time based on the current deviation magnitude. The integral term accumulates historical deviations to eliminate steady-state pressure offset, ensuring that the pressure converges to the target range during prolonged contact. The differential term responds to the rate of change of deviation to suppress sudden pressure changes and improve dynamic response stability. The output is a force correction value.

[0035] S4.4 Position Feedback Closed-Loop Calculation Using the current target path point and its normal vector from the planned path point sequence output by S2, and the current actual position of the end effector output by S3 as inputs, the position deviation is calculated. This deviation is decomposed into a component along the normal direction of the contact surface and a tangential component. The normal component reflects the degree of penetration or retraction of the end effector along the normal direction of the contact surface, while the tangential component ensures trajectory tracking accuracy. The normal position deviation and tangential position deviation are output.

[0036] S4.5 Dual Closed-Loop Fusion and Impedance Mapping Using force correction values, normal position deviation, and tangential position deviation as inputs, a dual closed-loop fusion is performed. The normal channel, with force control as its core and position constraints as its auxiliary, combines the force correction value with the normal position deviation and normal motion velocity to calculate the expected normal force, ensuring stable contact pressure. The tangential channel, with position tracking as its core, combines the tangential position deviation with the tangential motion velocity to calculate the expected tangential force, ensuring trajectory accuracy.

[0037] S4.6 Inverse kinematics solution for joint torques Using the desired force vector at the end effector and the current joint angle as inputs, force mapping is performed using the robot's Jacobian matrix. By transposing the Jacobian matrix, the desired force in the end effector space is mapped to the driving torque in the joint space, realizing the conversion of the Cartesian space force control target to the joint motor command.

[0038] S4.7 Output joint drive torque Using the joint torque vector as input, the current loop is converted by the motor driver and output to the servo motors of each joint, driving the flexible bionic hand actuator to generate corresponding contact force adjustments. At the same time, the current actual contact pressure and joint angle are fed back to S4.1 and S4.4, forming a closed-loop iteration.

[0039] S4.8 Pressure Range Convergence Determination The system monitors the actual contact pressure in real time to determine if it falls within the preset touch force range. If it does, the current control output is maintained; if it does not and the deviation continues to exceed the preset tolerance, the S6 safety protection logic is triggered. The constant force touch status flag is output to step S5 as the enable condition for vibration coordinated control. S5: During the constant force stroking process in S4, the current stroking trajectory position and movement speed are received. Based on the mapping relationship between movement speed and vibration parameters, the vibration frequency and amplitude are dynamically adjusted, and the vibration control signal is output to the low-frequency vibration module to realize the coordinated control of stroking motion and low-frequency vibration. S6: During the execution of S4 and S5, it receives contact pressure signals and joint posture signals in real time. When the pressure exceeds the threshold, the movement posture is abnormal, or the contact is lost, it outputs a deceleration and stop command, automatically decelerates and stops the movement.

[0040] In the description of this specification, the references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0041] The foregoing has shown and described the basic principles, main features, and advantages of this application. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this application. Various changes and modifications can be made to this application without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of the claims of this application.

Claims

1. A humanoid robot haptics interaction system, characterized by, include: The humanoid robot itself; A flexible bionic hand actuator is installed on the humanoid robot body and includes multiple degrees of freedom joints, a flexible covering layer covering the outside of the joints, and a pressure sensing array built into the flexible covering layer. A low-frequency vibration module, integrated inside the flexible bionic hand actuator, is used to output mechanical vibrations with adjustable frequency and amplitude. The multimodal sensing unit includes at least a visual positioning sensor, a joint posture sensor, and the pressure sensing array; The control unit is connected to the flexible bionic hand actuator, the low-frequency vibration module, and the multimodal sensing unit, respectively. The control unit is configured as follows: Receive human image data collected by the visual positioning sensor, identify the human touch area and locate key points, establish the mapping relationship between the robot coordinate system and the human surface coordinates, and generate and output the spatial coordinates of the target touch area. The system receives the spatial coordinates of the target touch area, drives the flexible bionic hand actuator to approach the target area, and receives the contact pressure signal fed back in real time by the pressure sensor array. It then uses a compliant control algorithm to perform closed-loop adjustment of the touch force, outputs joint driving torque, and stabilizes the contact pressure within a preset force range. The system receives the preset touch trajectory and the spatial coordinates of the target touch area, drives the flexible bionic hand actuator to complete the touch path movement, and synchronously outputs vibration control signals to the low-frequency vibration module to adjust the vibration parameters according to the current touch action stage and movement speed. It receives pressure signals from the pressure sensor array and attitude signals from the joint attitude sensor in real time, and outputs a safety shutdown command when the pressure exceeds the limit or the attitude is abnormal.

2. The humanoid robot petting interaction system according to claim 1, characterized in that: The flexible coating layer is made of flexible silicone material, and the detection range of the pressure sensing array is 0 to 20 N. 3.The humanoid robot hugging interaction system of claim 1, wherein: The frequency adjustment range of the low-frequency vibration module is 20Hz to 200Hz, and the amplitude adjustment range is 0.1mm to 3mm.

4. The humanoid robot petting interaction system according to claim 1, wherein: The visual positioning sensor uses an Intel RealSense D435 depth camera to capture human contours, achieve region segmentation and key point recognition. Its effective depth range is 0.2m to 10m, with an RGB resolution of 1920×1080 and a depth resolution of 1280×720.

5. The humanoid robot petting interaction system according to claim 1, wherein: The joint posture sensor includes an AS5047P magnetic encoder installed in each joint for real-time acquisition of joint rotation angles. The end effector integrates an MPU6050 six-axis IMU for monitoring the overall hand posture.

6. The humanoid robot petting interaction system according to claim 1, wherein: The pressure sensing array uses FlexiForce A201 ultra-thin film pressure sensors, which are arranged at multiple points under the flexible covering layer of the bionic hand. The control unit polls each sensing point at a sampling rate of not less than 100Hz to obtain contact pressure distribution data for constant force closed-loop control and overpressure safety protection.

7. The humanoid robot touch interaction system according to claim 1, characterized in that: The control unit is an embedded processor, specifically including: a microcontroller (MCU) or a digital signal processor (DSP).

8. A control method for a humanoid robot touch interaction system, using the humanoid robot touch interaction system according to any one of claims 1 to 9, characterized in that, Includes the following steps: S1: Collect images of the human body surface through a vision sensor, perform region segmentation and key point localization, establish a mapping relationship between the robot coordinate system and the human body surface coordinates, and output the spatial coordinates of the target touch area; S2: Receive the spatial coordinates of the target touch area output by S1, and preset the touch trajectory mode, touch force range, vibration frequency range and amplitude range; S3: Using the touch trajectory pattern preset in S2 and the spatial coordinates of the target touch area as input, the robot joint movement is controlled by inverse kinematics solution to drive the flexible bionic hand to approach the target area, and the contact pressure is collected in real time through the pressure sensor array and the contact pressure signal is output. S4: Receive the contact pressure signal output by S3 and the touch force range preset by S2, calculate the pressure deviation, use PID control or impedance control algorithm to perform force feedback and position feedback dual closed-loop adjustment, output joint driving torque, stabilize the contact pressure in the preset touch force range, and realize constant force touch. S5: During the constant force stroking process described in S4, the current stroking trajectory position and movement speed are received, and the vibration frequency and amplitude are dynamically adjusted according to the mapping relationship between movement speed and vibration parameters. The vibration control signal is output to the low-frequency vibration module to realize the coordinated control of stroking motion and low-frequency vibration. S6: During the execution of S4 and S5, the contact pressure signal and joint posture signal are received in real time. When the pressure exceeds the threshold, the movement posture is abnormal, or the contact is lost, a deceleration and stop command is output to automatically decelerate and stop the action.

9. A humanoid robot touch interaction method according to claim 8, characterized in that: Images of the human body surface are acquired using a visual sensor, and region segmentation and key point localization are performed. A mapping relationship between the robot's coordinate system and the human body surface coordinates is established, outputting the spatial coordinates of the target touch area. Specifically, this includes: Human RGB-D images are acquired using a depth camera, and the three-dimensional spatial coordinates of multiple joints in the human body are extracted using a skeletal key point detection model. The extracted joint points are matched with predefined touch regions in the key point database. Each touch region is associated with a specific set of joint points and region boundary rules constructed from the geometry of the joint points. Based on the actual coordinates of the matched joints, the boundary of the touching area and the sequence of internal target points are dynamically calculated and converted to the robot's base coordinate system; When a match fails or the confidence level of a key point is below a threshold, region segmentation based on depth images or a preset offset is used as a backup localization method.

10. A humanoid robot touch interaction method according to claim 8, characterized in that: The preset touch trajectory includes linear pushing, circular kneading, and fixed-point holding.

Images