A fast adaptive grasping device for dynamic targets and a method thereof

Abstract

This invention discloses a rapid adaptive grasping device and method for dynamic targets. The device includes a base, three fingers, a high-speed vision unit, and an FPGA control system. Each finger consists of a rigid segment and a flexible segment. The flexible segment includes a silicone epidermis, a silicone cavity, and a tactile sensor array. The FPGA processing unit performs hardware-level fusion processing on visual and tactile data, directly generating drive commands to achieve ultra-low latency closed-loop control. The grasping method consists of five stages: approach, contact detection, gripping, slip compensation and wrapping mode switching, and release. When micro-slippage is detected, the clamping force and air pressure are increased first. If the slippage is still not eliminated, the device automatically switches to wrapping mode. A miniature push rod and proximity sensor are also included to achieve active obstacle avoidance and collaborative operation. This invention features fast response speed, stable grasping of high-speed moving targets, strong adaptability, and the ability to handle objects of various materials and shapes, achieving safe and flexible grasping in cluttered environments.

Description

Technical Field

[0001] This invention relates to the field of robot grasping technology, and in particular to a rapid adaptive grasping device and method for dynamic targets. Background Technology

[0002] In fields such as industrial automation, medical surgery, and service robots, the need for rapid, stable, and non-destructive grasping of dynamically moving targets is becoming increasingly urgent. For example, on high-speed production lines, it is necessary to grasp irregularly shaped parts moving randomly on conveyor belts; in minimally invasive surgery, it is necessary to stably hold tissues that pulsate due to physiological rhythms; and in home service scenarios, it is necessary to grasp thrown objects in mid-air. These applications place extremely high demands on the response speed, adaptability, and environmental adaptability of grasping devices.

[0003] Existing web scraping technologies mainly face the following technical challenges: First, there is the uncertainty of target motion. The trajectory, velocity, and attitude of dynamic targets change rapidly over time, and traditional grasping methods based on static vision positioning are difficult to respond in real time due to the serial delays in image acquisition, transmission, processing, and motion planning. Although trajectory prediction algorithms based on deep learning have made some progress in recent years, complex neural network models have high computational overhead and limited generalization ability, making it difficult to simultaneously guarantee real-time performance and robustness under millisecond-level response requirements.

[0004] Second, there is the diversity of object shapes and materials. Grasping objects may include rigid metal parts, fragile glassware, soft food or biological tissue, and even irregularly shaped waste. A rigid gripper with a single structure struggles to adapt to multiple objects simultaneously, often damaging fragile items due to stress concentration or failing to stably grip smooth objects due to insufficient contact area. While conventional flexible grippers can achieve compliant contact, they typically lack sufficient holding force, causing objects to slip or fall during grasping. Balancing compliant contact and firm grip when grasping different objects has long been a technical challenge.

[0005] Third, occlusion and interference in complex environments. In densely stacked or cluttered scenes, target objects are often partially occluded by other objects, making it difficult for traditional graspers to directly approach the target location. Simultaneously, accidental collisions during the grasping process may damage surrounding objects or cause grasping failure. Existing technologies largely focus on planning obstacle avoidance paths through advanced vision algorithms, neglecting the crucial role of the grasper's mechanical structure and haptic feedback in dynamic scenes. This results in systems often requiring complex path replanning when facing occlusion, leading to slow response and insufficient reliability.

[0006] To address the aforementioned issues, some studies have attempted to improve the success rate of grasping by fusing visual and force sensors, but the following common drawbacks still exist: There is a long serial processing link between perception and execution, which makes it difficult to meet the high-speed grasping requirements of dynamic targets. The gripper structure lacks the ability to adjust stiffness online, making it impossible to simultaneously achieve impact absorption, shape self-adaptation, and strong retention on the same mechanical structure. The lack of hardware-level active obstacle avoidance and interference handling mechanisms, and the over-reliance on upper-level algorithm planning, limits its practicality in unstructured environments.

[0007] Therefore, there is an urgent need to provide a grasping device and its control method that can quickly respond to dynamic targets and adaptively handle different shapes and occlusion situations without relying on complex calculations, through hardware innovation and low-level control. Summary of the Invention

[0008] In order to overcome the above-mentioned defects of the prior art, the present invention provides a fast adaptive grasping device and method for dynamic targets, so as to solve the problems existing in the background art.

[0009] This invention provides the following technical solution: A fast adaptive grasping device for dynamic targets includes: The base contains a miniature air pump, a high-speed solenoid valve, and a tactile signal conditioning circuit. At least three fingers are mounted on the base, each finger including a proximal rigid segment and a distal flexible segment, the proximal rigid segment and the distal flexible segment being connected by a joint; The proximal rigid segment is mounted on the base via a first driving device, which is used to rotate the proximal rigid segment and the distal flexible segment connected thereto around the base. The distal flexible segment has a flexible structure with adjustable stiffness, and the interior of the flexible structure is equipped with a tactile sensor array 2023 for detecting the contact force distribution when in contact with a target. A high-speed vision unit, installed at the center of the base, includes a global shutter industrial camera and a laser structured light projector, used to acquire the target's position, velocity and depth information in real time; The control system includes an FPGA processing unit, which is connected to the high-speed vision unit, the tactile sensor array 2023, and the first driving device, respectively. The FPGA processing unit is used to receive and process visual data and tactile data to fuse and generate control commands to directly drive the first driving device and the variable stiffness structure of the distal flexible segment.

[0010] Preferably, the finger is further provided with a second driving device, which is installed at the joint between the proximal rigid segment and the distal flexible segment, for independently driving the distal flexible segment to rotate relative to the proximal rigid segment around the joint.

[0011] Preferably, both the first and second driving devices are brushless DC servo motors, giving each finger two active rotational degrees of freedom.

[0012] Preferably, the flexible structure further includes a silicone skin and a silicone cavity 2022. The silicone skin is disposed on the outer layer of the finger and has anti-slip textures on its surface. The silicone cavity 2022 is disposed on the middle layer of the finger and is connected to a miniature air pump and a high-speed solenoid valve disposed inside the base via an air tube. The tactile sensor array 2023 is disposed on the inner layer of the finger and is a piezoresistive tactile sensor array arranged in a 16×16 rectangular structure.

[0013] Preferably, the base has a miniature telescopic push rod on its side for pushing away obstructions; the front end of the miniature telescopic push rod is covered with a soft silicone tip; the FPGA processing unit controls the extension and retraction state of the miniature telescopic push rod according to the position of the obstruction detected by vision.

[0014] Preferably, a ring-shaped capacitive proximity sensor is mounted on the outer side of the rigid segment of each finger.

[0015] A fast adaptive grasping method for dynamic targets includes the following steps: S1: The high-speed vision unit tracks the dynamic target in real time, the FPGA processing unit predicts the target trajectory and moves the grasping head to the pre-grabbing position, presets the finger opening width, and keeps the distal flexible segment in a low stiffness state. S2: Controls finger closing. When the tactile sensor detects that the contact pressure of any finger exceeds the threshold, the closing movement stops immediately. S3: Based on the target material identified by visual recognition, preset the initial clamping force and flexible section air pressure, and increase the clamping force to the target value; S4: During the holding process, continuously monitor the micro-slip signal of the tactile array. If a slipping trend is detected, perform slip compensation or switch to wrapping mode. S5: Upon receiving the release command, the flexible section exhausts air, and the fingers open to reset.

[0016] Preferably, step S4 specifically includes: Step S41: Continuously monitor the micro-slip signal of the tactile array. The criteria for determining micro-slip is that the fluctuation of the pressure difference between two adjacent rows of sensors exceeds 0.1N and lasts for more than 5ms. Step S42: Once micro-slippage is detected, immediately increase the target clamping force by 0.5N and simultaneously increase the air pressure of the corresponding flexible segment of the finger by 5kPa; Step S43: Determine if the slip has disappeared; if it has, lock the current parameter. Step S44: If slippage still exists and the total grip force has exceeded 200% of the initial target value, automatically switch to wrapping mode: reduce the air pressure of the flexible segments of all fingers to 10 kPa, and increase the finger closing angle to completely wrap the object with the flexible segments.

[0017] Preferably, during the crawling process: When the visual system detects an obstruction in front of the target, the FPGA processing unit controls the micro telescopic push rod on the side of the base to extend and push away the obstruction, creating an unobstructed path for the main grasping finger, and then controls the push rod to retract. When the visual system detects that the target is partially obscured and the push rod cannot clear it, the FPGA processing unit controls two of the fingers to approach the target from both sides, and controls the third finger to first push away the obstruction, and then quickly return to participate in the grasping. When the proximity sensor detects that the distance between the finger and the obstacle is less than the safety threshold, the FPGA processing unit automatically decelerates or stops the finger's movement; if a collision is unavoidable, it controls the finger to move in the opposite direction and triggers local path replanning.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention completes all data acquisition, preprocessing, and fusion decision-making for the high-speed vision unit and distributed tactile sensor array within the FPGA processing unit, without the need for a host computer. The FPGA processing unit directly performs hardware-level preprocessing on the visual image, outputting the target edge contour and centroid coordinates; simultaneously, it processes the tactile array data in real time at a 1000Hz sampling rate, calculating the contact point, total grip force, and micro-slip index; and it fuses the visual target position and tactile contact state within the FPGA processing unit to directly generate finger drive commands. This makes the cycle from perception to decision-making and fall control of the system less than 5ms. Compared with traditional visual servo systems based on host computer serial processing, this greatly improves the tracking robustness and grasping success rate for dynamic target trajectories and rapidly changing speeds. 2. Each finger of the present invention consists of a proximal rigid segment and a distal flexible segment. The distal flexible segment adopts a flexible structure with adjustable stiffness. Combined with an adaptive gripping control method, it maintains low stiffness during the approach phase to absorb impact. After contact, the initial clamping force and stiffness are preset according to the material recognized by vision. If micro-slippage is detected during the holding phase, the clamping force and air pressure are increased first. If the slippage still does not disappear and the clamping force exceeds 200% of the initial target value, it automatically switches to the wrapping mode, which greatly improves the gripping system's adaptability to all types of objects, including rigid metals, fragile glass, soft tissues, and low-friction objects. 3. This invention features a miniature telescopic push rod on the side of the base. The FPGA processing unit controls the push rod to quickly extend and remove small obstructions based on visual detection results. When the push rod cannot clear the obstruction, the system utilizes the independent control capability of three fingers. First, two fingers are controlled to approach the target from both sides, and the third finger is used to push away the obstruction and then quickly return to participate in the grasping. At the same time, a ring-shaped capacitive proximity sensor is also installed on the outer side of the rigid segment of each finger. When the distance to other objects is less than 5mm, the FPGA processing unit automatically decelerates or stops the movement of that finger. If a collision is unavoidable, the finger is controlled to move in the opposite direction and triggers local path replanning. The three work together to form a multi-level active obstacle avoidance and collaborative operation mechanism. In densely stacked scenarios, compared with traditional single-arm grasping, the operational flexibility and safety of the grasping system in partially obstructed and densely stacked environments are improved. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the gripping device of the present invention.

[0020] Figure 2 This is a bottom view of the base of the present invention.

[0021] Figure 3 This is a schematic diagram of the interior of the base of the present invention.

[0022] Figure 4 This is a schematic diagram of the finger in this invention.

[0023] Figure 5 This is a hardware block diagram of the control system of the present invention.

[0024] Figure 6 This is a flowchart of the grasping control process of the present invention.

[0025] Figure 7 This is a schematic diagram of the grasping timing in the high-speed conveyor belt scenario of the present invention.

[0026] The attached figures are labeled as follows: 100, base; 101, miniature air pump; 102, high-speed solenoid valve; 103, tactile signal conditioning circuit; 104, embedded microcontroller; 105, industrial robot interface; 200, finger; 201, proximal rigid segment; 202, distal flexible segment; 203, first driving device; 2304, second driving device; 300, high-speed vision unit; 301, industrial camera; 302, laser structured light projector; 400, control system; 500, miniature telescopic push rod; 501, miniature telescopic push rod; 600, capacitive proximity sensor. Detailed Implementation

[0027] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0028] To address the problems existing in the prior art, this application provides a fast adaptive grasping device for dynamic targets, specifically, as follows: Figures 1 to 3 As shown, the device includes: a base 100, three fingers 200, a high-speed vision unit 300, a control system 400, a miniature telescopic push rod 500, and a ring capacitive proximity sensor 600.

[0029] The base 100 is a circular disc with a diameter of 80 mm, and houses a miniature air pump 101, a high-speed solenoid valve 102, a tactile signal conditioning circuit 103, and an embedded microcontroller 104. The top of the base 100 has an industrial robot interface 105 for quick installation onto the end effector of a multi-axis robotic arm. A mounting slot is provided on the side of the base 100 for securing a miniature telescopic push rod 500.

[0030] The three fingers 200 are symmetrically distributed at 120° along the circumference of the base 100. All three fingers 200 can be detachably installed on the bottom of the base 100, so that each finger can be replaced independently. Each finger 200 consists of a proximal rigid segment 201 and a distal flexible segment 202, which are connected by a precision joint.

[0031] The proximal rigid segment 201 is made of aluminum alloy and is 60 mm long. One end of it is mounted on the base 100 via a first drive device 203. The first drive device 203 is a brushless DC servo motor that can drive the proximal rigid segment 201 and the connected distal flexible segment 202 to rotate independently around the base 100. The rotation range is -30° to +90°, with a maximum angular velocity of 300° / s and a maximum angular acceleration of 1000° / s². The proximal rigid segment 201 integrates an absolute angle encoder, which can precisely control the opening and closing angle and speed of the fingers. A ring-shaped capacitive proximity sensor 600 is installed on the outside of the rigid segment, with a detection distance of 0-30 mm. It can actively decelerate or stop the fingers when they approach obstacles to prevent collisions. This enables each finger to have independent and rapid movement capabilities, while also providing active anti-collision functionality, laying the foundation for collaborative grasping.

[0032] The distal flexible segment 202 is 40 mm long and has a flexible structure with adjustable stiffness.

[0033] The flexible structure includes: a silicone skin 2021, a silicone cavity 2022, and a tactile sensor array 2023.

[0034] The silicone skin 2021 is applied to the outer layer of the finger, with a coefficient of friction ≥1.2 and a diamond-shaped anti-slip texture on its surface. The silicone skin 2021 directly contacts the target, and its high frictional properties allow for sufficient static friction with relatively low clamping force, making it particularly suitable for smooth or fragile objects.

[0035] The silicone cavity 2022 is located in the middle layer of the finger and is connected to a miniature air pump 101 and a high-speed solenoid valve 102 inside the base 100 via an independent air tube. By adjusting the air pressure inside the cavity, the stiffness of the distal flexible segment 202 can be continuously varied within the range of 5 to 50 N / mm, and the equivalent Shore hardness is adjustable from A10 to A70. This variable stiffness capability allows the finger to maintain low stiffness when approaching a target to absorb impact and avoid damage; after contact, it can quickly increase stiffness according to the characteristics of the object to provide sufficient clamping force; and when wrapping is required, it can reduce stiffness to achieve shape adaptation, realizing real-time controllable stiffness-flexibility transition.

[0036] The 2023 tactile sensor array is located on the inner layer of the finger. It is a piezoresistive tactile sensor array arranged in a 16×16 rectangular structure with a unit spacing of 1.5 mm. The pressure measurement range is 0-10 N, with a resolution of 0.02 N, and it acquires contact force distribution at a sampling rate of 1000 Hz. The sensor array can detect the contact point position, pressure center, total grip force, and local slippage trend in real time. When the pressure difference between two adjacent rows of sensors fluctuates rapidly, the system determines that micro-slippage has occurred. This detection is entirely based on hardware threshold comparison and requires no complex calculations. The rich tactile information provided by this array is the direct basis for subsequent slippage compensation and wrapping mode switching.

[0037] The three fingers are symmetrically distributed at 120°, and can achieve various modes such as concentric grasping, parallel grasping, or wrapping grasping through independent rotation. When the three fingers simultaneously close towards the center, they form an approximately spherical workspace; when two fingers are fixed and the third finger moves, a push-pull operation can be achieved. This layout combines the stability of multi-point contact with the flexibility of multiple degrees of freedom.

[0038] In another embodiment of the invention, each finger 200 may also be provided with a second drive device 204 at a precision joint. The second drive device is also a brushless DC servo motor, which is used to independently drive the distal flexible segment 202 to rotate relative to the proximal rigid segment 201 around the joint 230, so that each finger has two active rotational degrees of freedom and can perform more complex active wrapping actions.

[0039] A high-speed vision unit 300 is mounted at the center of the base 100, comprising a global shutter industrial camera 301 and a laser structured light projector 302. The industrial camera 301 has a resolution of 640×480 and a frame rate of 200 fps; the laser structured light projector 302 is used to acquire the depth information of the target, with a measurement range of 50-200 mm and an accuracy of ±0.5 mm. The camera performs image preprocessing through an FPGA processing unit, directly outputting the edge contour and centroid coordinates of the target area with a latency of less than 2 ms. The laser structured light projector works synchronously to acquire the target depth information and form three-dimensional spatial coordinates. Vision processing is entirely completed at the hardware level, without going through a host computer, ensuring extremely low latency.

[0040] The miniature telescopic actuator 500 is mounted on the side of the base 100, with a stroke of 20 mm, a thrust of 5 N, and is driven by a shape memory alloy wire with a response time of 10 ms. The front end of the actuator is covered with a soft silicone tip 501 to prevent damage to objects or equipment when removing obstructions. The FPGA processing unit controls the extension and retraction of the actuator based on the visually detected position of the obstruction, enabling single extension or continuous jiggle operation.

[0041] The control system 400 includes an FPGA processing unit, which can be a Xilinx Artix-7 series model. The FPGA processing unit is connected to the high-speed vision unit 300, the tactile sensor array 2023, the drive unit, the miniature air pump 101, the high-speed solenoid valve 102, the ring capacitive proximity sensor 600, and the miniature telescopic push rod 500. The FPGA processing unit internally includes an image preprocessing module, a tactile data processing module, a Kalman filter, a fusion decision module, and a motor drive module. All modules are implemented in hardware logic circuits and do not depend on software programs for operation.

[0042] This invention employs a hardware-based sensing method that combines high-speed vision with distributed tactile sensing, fundamentally reducing reliance on complex algorithms. The specific processing flow is as follows: Visual data processing: The industrial camera 310 acquires images at 200 fps. The image preprocessing module of the FPGA processing unit directly performs binarization, edge detection, and centroid calculation on the raw pixel data, outputting the edge contour and centroid coordinates of the target area. The entire process has a latency of less than 2 ms. Simultaneously, the laser structured light projector 302 projects structured light stripes, and the FPGA processing unit calculates the target's three-dimensional spatial coordinates and velocity vector based on the stripe deformation.

[0043] Tactile data processing: The 2023 tactile sensor array outputs a 16×16 pressure value matrix with a sampling rate of 1000 Hz. The tactile data processing module of the FPGA processing unit calculates the following characteristic parameters in real time: total grip force, pressure center, pressure distribution uniformity, and micro-slip index. The micro-slip is determined by the fluctuation of the pressure difference between two adjacent rows of sensors exceeding 0.1 N and lasting for more than 5 ms. This determination is implemented entirely based on a hardware comparator, without the need for complex calculations.

[0044] Multimodal data fusion: The FPGA processing unit receives the target position from visual output and the contact state from tactile output, and directly generates grasping control commands, such as finger closing speed, target gripping force, and stiffness adjustment, in the fusion decision module. The entire perception-decision-control cycle is less than 5 ms. This method tightly integrates perception and decision-making, enabling real-time tracking and rapid response to dynamic targets, avoiding the serial delay between visual processing, path planning, and motion control in traditional methods.

[0045] This invention designs a complete robotic arm trajectory planning and tracking control method based on the motion characteristics of dynamic targets, ensuring that the gripper head can quickly, smoothly, and accurately reach the target gripping position and remain synchronized with the target.

[0046] Target trajectory prediction and grasping point planning: The high-speed vision unit continuously outputs the target's 3D position and velocity vectors at 200 fps. The Kalman filter embedded in the FPGA processing unit predicts the target's motion trajectory in real time for the next 50 ms. Based on the predicted trajectory, the system selects the optimal grasping point on the target's motion path. The selection of the grasping point follows these principles: priority is given to the target's centroid or preset grasping feature points, and the grasping point is ensured to be within the robotic arm's workspace and that the finger's spread width can cover the target's size. When the target's motion speed is high, the system automatically moves the grasping point forward, ensuring the robotic arm is in position before the target arrives, thus shortening the response time.

[0047] Cartesian Space Trajectory Generation: The desired trajectory of the robotic arm's end effector is generated in Cartesian space using a fifth-order polynomial, ensuring continuity of position, velocity, and acceleration. The trajectory starts at the robotic arm's current pose and ends at the predicted grasping point's spatial coordinates. The trajectory duration is dynamically adjusted based on the distance between the robotic arm and the target. The coefficients of the fifth-order polynomial are uniquely determined by boundary conditions, generating a smooth displacement curve. If the robotic arm needs to track a continuously moving target, the trajectory planner will generate a velocity matching segment to keep the robotic arm's end effector relatively stationary with respect to the target during the grasping phase.

[0048] Inverse kinematics and constraint handling in joint space: The Cartesian space trajectory is mapped to the joint angles of the robotic arm in real time through inverse kinematics. The inverse kinematics solution employs analytical or numerical iterative methods to ensure that the joint angles are continuous and meet the physical constraints of the robotic arm (joint angle limits, maximum angular velocity, maximum angular acceleration). When the trajectory exceeds the reachable range of the robotic arm, the system automatically adjusts the gripping point or reduces the tracking speed and issues a warning.

[0049] Speed ​​Synchronization and Adaptive Tracking Control: In uniform motion scenarios such as conveyor belts, the robotic arm needs to maintain speed synchronization with the target. This invention employs a visual feedback-based adaptive speed tracking controller: the system calculates the relative position error between the robotic arm's end effector and the target in real time, and generates a speed correction value through a proportional controller, allowing the robotic arm's end effector speed to gradually approach the target speed. The controller parameters are adaptively adjusted according to the target's motion characteristics; when the target speed changes drastically, the proportional gain is increased to ensure rapid following; when the target speed is stable, the gain is decreased to avoid jitter. Experiments show that this controller can stabilize the tracking error within ±2 mm.

[0050] Obstacle avoidance path replanning: When the proximity sensor detects that the distance between the robotic arm and an obstacle is less than a safety threshold, the system immediately triggers local path replanning. Replanning employs a fast expanding random tree or elastic band algorithm to generate an alternative trajectory in Cartesian space to avoid the obstacle and redirect the robotic arm back to the target. The replanning process is executed entirely within the FPGA processing unit, taking less than 10 ms, ensuring safe grasping even in dynamic environments.

[0051] This invention proposes a variable stiffness adaptive grasping control method based on haptic feedback, comprising the following steps: S1: Approaching and pre-positioning After the system starts, a high-speed vision camera captures the dynamic target at 200 fps. The FPGA processing unit extracts the target contour and centroid coordinates, and predicts the target position in the next 50 ms using a Kalman filter. The robotic arm moves the center of the gripper head to the predicted position, while simultaneously pre-setting the finger spread width based on the target's shape and size. At this point, the flexible segments of the fingers maintain low stiffness to absorb any potential accidental collisions. This stage relies entirely on vision guidance; the fingers have not yet made contact with the target.

[0052] S2: Contact Detection and Stop The fingers close towards the center at a maximum angular velocity of 300° / s. During the closing process, the tactile sensor monitors the contact pressure at 1000 Hz. When the contact pressure of any finger exceeds 0.5 N, the system determines that it has made contact with the target, immediately stops the closing movement of all fingers, and records the current finger angle and contact point position. This threshold setting balances sensitivity and anti-interference capability, ensuring reliable triggering even when the dynamic target is slightly swaying. If the pressure does not reach 0.5 N during the closing process, the system will replan the grasping path based on visual feedback and repeat this stage.

[0053] S3: Grip and Stiffness Adjustment The system presets the initial clamping force and air pressure in the flexible section based on the target material identified visually. The preset values ​​are based on a material property database: clamping force 5 N, air pressure 50 kPa for metal parts; clamping force 3 N, air pressure 40 kPa for plastic parts; clamping force 2 N, air pressure 20 kPa for glass parts; and clamping force 1 N, air pressure 10 kPa for soft materials. Subsequently, the system gradually increases the clamping force by slowly increasing the finger closure angle, while simultaneously adjusting the air pressure to the preset value via a PID controller and monitoring the total grip force. When the total grip force reaches 90% of the target value, the increase stops and is maintained. During this stage, a proportional, integral, and derivative controller precisely adjusts the finger angle and air pressure to ensure a smooth increase in clamping force and avoid impact damage.

[0054] S4: Slide Compensation and Package Mode Switching During the holding process, the tactile array continuously monitors for micro-slippage. Micro-slippage is determined by a fluctuation in the pressure difference between two adjacent rows of sensors exceeding 0.1 N for a duration exceeding 5 ms. Once a slippage trend is detected, the system immediately increases the target gripping force by 0.5 N and simultaneously increases the air pressure of the corresponding finger by 5 kPa to increase contact surface friction. After this increase, the system again checks if the slippage has disappeared. If the slippage has disappeared, the current parameters are locked, and the system enters a stable holding phase. If the slippage still exists and the total gripping force exceeds 200% of the initial target value, the system determines that the surface friction of the object is insufficient, and automatically switches to a wrapping mode: reducing the air pressure of all fingers to 10 kPa while increasing the finger closure angle to completely wrap the object with the flexible segments, using shape locking and static friction to hold the object. In wrapping mode, the tactile sensors monitor the pressure distribution in real time to ensure uniform wrapping. This method achieves a seamless transition from active force control to passive shape adaptation, effectively solving the problem of grasping low-friction objects.

[0055] S5: Release and Reset Upon receiving a release command from the host computer, the flexible section first vents to atmospheric pressure, then the finger opens to its maximum angle, the tactile sensor is reset to zero, and finally, it returns to its initial position, completing one grasping cycle. If no release command is received, the system will remain in the stable holding phase, continuously monitoring for slippage and external interference.

[0056] Throughout the entire control process, all judgments and actions are completed within the FPGA processing unit and microcontroller, without the need for a host computer. The control cycle is less than 5 ms, ensuring real-time performance in high-speed dynamic scenarios.

[0057] To address the occlusion problem in dense and cluttered environments, a method combining three-finger independent control and an auxiliary push-pull mechanism is adopted, as detailed below: Independent finger control: The rigid segment servo motor of each finger can be independently controlled to achieve asymmetric grasping. When the vision system detects that the target is partially occluded, it can first control two fingers to approach the target from both sides, while the third finger is first used to push away the occlusion and then quickly returns to participate in the grasping. This process uses the vision unit to identify the position of the occlusion in real time, and the FPGA processing unit generates corresponding finger movement commands, eliminating the need for complex path planning and achieving a response time of less than 20 ms. Independent finger control also allows the system to dynamically adjust the finger posture during grasping. For example, when the target rotates, the angle of a single finger can be adjusted to adapt and prevent slippage.

[0058] Auxiliary Push-Push Mechanism: A miniature telescopic push rod 500 is located on the side of the base, with a stroke of 20 mm, a thrust of 5 N, driven by a shape memory alloy wire, and a response time of 10 ms. When the vision detects a small obstruction in front of the target, the push rod quickly extends and pushes it away, creating an unobstructed path for the main grasping finger. The front end of the push rod is covered with a soft silicone tip to prevent damage to the object. The extension and retraction state of the push rod is controlled by the FPGA processing unit based on visual feedback, and can be performed in a single motion or with continuous shaking to clear minor obstructions.

[0059] Proximity sensor collision avoidance: A ring-shaped capacitive proximity sensor is mounted on the outer side of the rigid segment of each finger, with a detection distance of 0-30 mm and adjustable sensitivity. When the sensor detects a distance of less than 5 mm from another object, the system automatically decelerates or stops the finger's movement to avoid collision. If a collision is unavoidable, the finger will immediately reverse its movement and replan its path. This function is especially important in dense environments, preventing damage to surrounding objects or the finger itself during grasping.

[0060] In some embodiments, during initial installation or periodic maintenance, the robotic arm is controlled to move to multiple known spatial points, and the image coordinates of these points are simultaneously recorded using a vision camera. The transformation matrix between the camera coordinate system and the robotic arm base coordinate system is then calculated. The calibration results are stored in the non-volatile memory of the FPGA processing unit. Before each grasping operation, the system automatically transforms the target position output by vision to the robotic arm base coordinate system to ensure grasping accuracy.

[0061] In some embodiments, the system records the success / failure of each grasping attempt, along with the corresponding tactile data (slippage degree, actual gripping force). When grasping of the same type of object fails consecutively, the system automatically adjusts the preset gripping force (0.5 N increments) and air pressure (5 kPa increments) for that type of object, and stores the updated parameters in the database. After a period of operation, the system parameters will adaptively optimize, further improving the grasping success rate.

[0062] In some embodiments, a fast decision-making method based on priority scoring is employed: the FPGA processing unit calculates the distance, velocity, size, and trajectory of each target in real time, and calculates the priority score according to the following formula: Score = w1 (1 / distance) + w2 (target size matching degree) + w3 (motion direction weight) w1, w2, and w3 are preset coefficients (adjustable according to the scenario), and the motion direction weight gives higher priority to targets facing the robotic arm. The system updates the priority queue every 10 ms, selecting the target with the highest score for grasping, achieving efficient operation in multi-target scenarios.

[0063] Flexible segment air pressure regulation and robotic arm movement are the main sources of energy consumption for the system. To reduce long-term energy consumption, the system introduces intelligent energy management: in standby mode, the air pressure of all fingers is automatically reduced to the minimum maintenance value, and the air pump enters sleep mode; when a target is detected about to enter the grasping area, the air pump is woken up 200 ms in advance and the preset air pressure is restored. When planning the robotic arm's motion trajectory, the path with the lowest joint movement energy consumption is prioritized under the premise of meeting time constraints, which is achieved through offline optimization or real-time table lookup. Actual measurements show that this strategy can reduce the system's average power consumption by approximately 30%.

[0064] The system integrates a real-time self-test function, monitoring the status of the tactile sensors, air pressure sensors, vision camera, and communication link every second. When an abnormality is detected in a tactile sensor unit, the system automatically switches to interpolation of data from an adjacent sensor, or abandons the tactile feedback from that finger, relying solely on the tactile information of the remaining fingers for control. If air pressure regulation fails, the system will lock the current air pressure and maintain gripping using a purely mechanical clamping method, while simultaneously issuing an alarm signal to notify maintenance personnel. This fault-tolerant design ensures that the system can continue to operate under partial failure conditions, improving the system's reliability and safety.

[0065] Example 1

[0066] irregularly shaped parts gripping on the high-speed conveyor belt of the production line A conveyor belt in an electronics factory continuously transports irregularly shaped plastic parts of various sizes (20 mm-50 mm in size, 10 g-30 g in weight) at a speed of 1 m / s. This device is installed at the end effector of a six-axis industrial robotic arm. A vision camera captures the moving parts at 200 fps, and the FPGA processing unit calculates the part's position and velocity vector within 2 ms, predicting its position 50 ms ahead using a Kalman filter. The robotic arm begins tracking the part using a speed-matching method. When the part enters the gripping range (150 mm from the center of the robotic arm), the system executes the five-stage process described above: the fingers close with low stiffness (10 kPa air pressure), stop closing at the moment of contact (pressure > 0.5 N), and set a target gripping force of 3 N and air pressure of 40 kPa based on the visually recognized material (plastic), slowly increasing the gripping force to the target value. During the tracking motion, a tactile sensor detects a tendency for the part to slide backward, and the system automatically increases the gripping force to 3.5 N and the air pressure to 45 kPa, eliminating the sliding. The robotic arm accurately places the parts into the material box. The average time for a single gripping cycle is 120 ms, with a gripping success rate of 99.5%.

[0067] like Figure 5 The timing diagram of the high-speed conveyor belt scenario illustrates the complete process of a robotic arm's gripper head accelerating to catch up, matching speed, grasping and holding, and lifting off a part moving at a speed of 1 m / s. The part's trajectory is represented by a solid blue horizontal line, while the gripper head's trajectory is represented by a dashed red line. The first segment is an accelerating descent arc, the second segment is the horizontal speed matching segment, and the third segment is the lifting off arc, corresponding to the approach, gripping, and release stages in the flowchart. The three time points T0, T1, and T2 marked on the timing diagram correspond to the key nodes of visual detection of the part and the robotic arm starting to catch up, grasping and accompanying movement, and grasping and leaving, respectively, showing the entire grasping cycle of approximately 120 ms.

[0068] Example 2

[0069] Stable clamping of pulsating tissues during medical surgery In minimally invasive cardiac surgery, surgeons need to suture the surface tissue of the beating heart. Since the heart beats 60-100 times per minute, the tissue moves at approximately 20 mm / s. This miniaturized device is mounted on the end effector of a robotic gripper. The flexible segment of the finger is made of medical-grade silicone, and the tactile sensors meet biocompatibility requirements. A vision camera tracks markers on the tissue surface at 200 fps, predicting its trajectory. The finger approaches the tissue with extremely low stiffness (5 kPa air pressure), with a contact force of only 0.2 N, avoiding damage. Upon contact, the system automatically adjusts the gripping force to 1 N based on the tissue's softness, and the air pressure increases to 30 kPa, maintaining tissue stability without obstructing blood flow. When the heartbeat causes a tendency for tissue to detach, the tactile array detects a periodic shift in the pressure center. The system immediately increases the air pressure to 40 kPa and fine-tunes the finger angle to ensure stable gripping. Throughout the entire grasping process, tissue displacement is less than 0.5 mm.

[0070] Example 3

[0071] Home service robot grabs thrown objects In a home environment, users might toss items such as remote controls and mobile phones to a service robot, requesting it to catch them mid-air. The items can travel at speeds up to 3 m / s, following a parabolic trajectory. This device is mounted on the end effector of a mobile robot arm. A high-speed vision camera captures the target within 20 ms of the item being thrown and calculates the parabolic equation. The robotic arm quickly moves to the vicinity of the predicted landing point, with its fingers spread. When the item enters the grasping area (50 mm from the center), the vision triggers a grasping command. Due to the item's potential rotation, the tactile array detects uneven pressure at multiple points upon finger contact. The system immediately reduces finger stiffness (air pressure 10 kPa), allowing the flexible section to completely envelop the item, absorbing impact energy through large-area contact. Then, the system gradually increases the air pressure to 40 kPa, simultaneously adjusting the finger angle based on tactile feedback to stably hold the item. The entire process, from the item entering the grasping area to stable holding, takes approximately 80 ms, with a grasping success rate exceeding 90%.

[0072] The above embodiments demonstrate that the present invention can flexibly meet the dynamic target grasping needs in different scenarios, and is superior to the prior art in terms of speed, adaptability, and security.

[0073] Several points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection" and "linkage" should be interpreted broadly, and can be mechanical or electrical connection, or internal connection between two components, or direct connection. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationship. When the absolute position of the described object changes, the relative positional relationship may change.

[0074] The above description is only a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. Any equivalent modifications or changes made by those skilled in the art based on the content disclosed in the present invention should be included within the scope of protection set forth in the claims.

Claims

1. A fast adaptive grasping device and method for dynamic targets, characterized in that, include: The base contains a miniature air pump, a high-speed solenoid valve, and a tactile signal conditioning circuit. At least three fingers are mounted on the base, each finger including a proximal rigid segment and a distal flexible segment, the proximal rigid segment and the distal flexible segment being connected by a joint; The proximal rigid segment is mounted on the base via a first driving device, which is used to rotate the proximal rigid segment and the distal flexible segment connected thereto around the base. The distal flexible segment has a flexible structure with adjustable stiffness, and the interior of the flexible structure is equipped with a tactile sensor array for detecting the contact force distribution when in contact with a target. A high-speed vision unit, installed at the center of the base, includes a global shutter industrial camera and a laser structured light projector, used to acquire the target's position, velocity and depth information in real time; The control system includes an FPGA processing unit, which is connected to the high-speed vision unit, the tactile sensor array, and the first driving device, respectively. The FPGA processing unit is used to receive and process visual data and tactile data to fuse and generate control commands to directly drive the first driving device and the variable stiffness structure of the distal flexible segment.

2. The fast adaptive grasping device and method for dynamic targets according to claim 1, characterized in that: The finger is also provided with a second driving device, which is installed at the joint between the proximal rigid segment and the distal flexible segment, for independently driving the distal flexible segment to rotate relative to the proximal rigid segment around the joint.

3. The fast adaptive grasping device and method for dynamic targets according to claim 2, characterized in that: Both the first and second drive devices are brushless DC servo motors, giving each finger two active rotational degrees of freedom.

4. The fast adaptive grasping device and method for dynamic targets according to claim 1, characterized in that: The flexible structure also includes a silicone skin and a silicone cavity. The silicone skin is placed on the outer layer of the finger and has anti-slip textures on its surface. The silicone cavity is placed on the middle layer of the finger and is connected to a miniature air pump and a high-speed solenoid valve inside the base via an air tube. The tactile sensor array is placed on the inner layer of the finger and is a piezoresistive tactile sensor array arranged in a 16×16 rectangular structure.

5. The fast adaptive grasping device and method for dynamic targets according to claim 1, characterized in that: The base has a miniature telescopic push rod on its side for pushing away obstructions; the front end of the miniature telescopic push rod is covered with a soft silicone tip; the FPGA processing unit controls the extension and retraction of the miniature telescopic push rod according to the position of the obstruction detected by vision.

6. The fast adaptive grasping device and method for dynamic targets according to claim 1, characterized in that: A ring-shaped capacitive proximity sensor is installed on the outer side of the rigid segment of each finger.

7. A fast adaptive grasping method for dynamic targets, based on the fast adaptive grasping device for dynamic targets according to any one of claims 1-6, characterized in that, Includes the following steps: S1: The high-speed vision unit tracks the dynamic target in real time, the FPGA processing unit predicts the target trajectory and moves the grasping head to the pre-grabbing position, presets the finger opening width, and keeps the distal flexible segment in a low stiffness state. S2: Controls finger closing. When the tactile sensor detects that the contact pressure of any finger exceeds the threshold, the closing movement stops immediately. S3: Based on the target material identified by visual recognition, preset the initial clamping force and flexible section air pressure, and increase the clamping force to the target value; S4: During the holding process, continuously monitor the micro-slip signal of the tactile array. If a slipping trend is detected, perform slip compensation or switch to wrapping mode. S5: Upon receiving the release command, the flexible section exhausts air, and the fingers open to reset.

8. A fast adaptive grasping method for dynamic targets according to claim 7, characterized in that: The aforementioned step S4 specifically includes: Step S41: Continuously monitor the micro-slip signal of the tactile array. The criteria for determining micro-slip is that the fluctuation of the pressure difference between two adjacent rows of sensors exceeds 0.1N and lasts for more than 5ms. Step S42: Once micro-slippage is detected, immediately increase the target clamping force by 0.5N and simultaneously increase the air pressure of the corresponding flexible segment of the finger by 5kPa; Step S43: Determine if the slip has disappeared; if it has, lock the current parameter. Step S44: If slippage still exists and the total grip force has exceeded 200% of the initial target value, automatically switch to wrapping mode: reduce the air pressure of the flexible segments of all fingers to 10 kPa, and increase the finger closing angle to completely wrap the object with the flexible segments.

9. A fast adaptive grasping method for dynamic targets according to claim 7, characterized in that, During the crawling process: When the visual system detects an obstruction in front of the target, the FPGA processing unit controls the micro telescopic push rod on the side of the base to extend and push away the obstruction, creating an unobstructed path for the main grasping finger, and then controls the push rod to retract. When the visual system detects that the target is partially obscured and the push rod cannot clear it, the FPGA processing unit controls two of the fingers to approach the target from both sides, and controls the third finger to first push away the obstruction, and then quickly return to participate in the grasping. When the proximity sensor detects that the distance between the finger and the obstacle is less than the safety threshold, the FPGA processing unit automatically decelerates or stops the finger's movement; if a collision is unavoidable, it controls the finger to move in the opposite direction and triggers local path replanning.

Images