Bionic picking manipulator and grabbing control method thereof

Abstract

The invention discloses a bionic picking manipulator and a grabbing control method thereof, relates to the technical field of automatic picking robots, and comprises a rope-driven three-finger bionic picking manipulator and a control method thereof, and the manipulator is provided with a plurality of pressure sensors for acquiring contact force in the grabbing process; one attitude sensor is used for acquiring the attitude of the manipulator, the plurality of steering engines are used for controlling the positions of the knuckles respectively, and the core of the control method is that a comprehensive grabbing quality index Q is calculated based on contact force, attitude and knuckle position data acquired in real time. The index Q is composed of Qp for reflecting the stability in a fruit plane and Qf for reflecting the axial anti-sliding capability. According to the system, the real-time Q value is compared with a target threshold value Qs set according to the ideal grabbing posture, output of all the joint steering engines is dynamically adjusted, and therefore self-adaptive optimal distribution of grabbing force is achieved under the multiple grabbing postures, and grabbing stability can still be kept when part of contact points fail.

Description

Technical Field

[0001] This invention relates to the field of automated harvesting robot technology, and in particular to a bionic harvesting manipulator and its control method. Background Technology

[0002] Automated fruit and vegetable harvesting is a research hotspot and challenge in the field of agricultural robotics. With the increasing aging population and the development of large-scale agricultural production, replacing repetitive and labor-intensive manual harvesting operations with intelligent equipment has become a key way to enhance agricultural competitiveness. In a harvesting robot system, the harvesting manipulator, as the terminal mechanism that directly interacts with the target crop, directly determines the harvesting success rate, operational efficiency, and degree of damage to the crops of the entire system, thus making it one of the core technological equipment.

[0003] In recent years, bionic principles have been widely applied to the design of gripping end effectors, aiming to simulate the multi-finger coordination and adaptive wrapping ability of the human hand to improve the gripping adaptability of irregular targets. However, existing bionic fruit and vegetable gripping solutions still generally have the following limitations:

[0004] (1) Insufficient force sensing and control capabilities: Most designs focus on the conformation of the mechanism, but fail to deeply integrate a highly sensitive distributed force sensing system, and cannot obtain accurate multi-point contact force information in real time, resulting in the lack of basis for the control system to achieve fine force modulation.

[0005] (2) Low level of intelligence in grasping strategy: The grasping process usually relies on a preset fixed trajectory or simple force threshold judgment. It lacks a quantitative indicator that can comprehensively evaluate the current grasping state. Therefore, it cannot dynamically and adaptively adjust the output force of each finger joint according to the specific size, weight and real-time grasping posture of different fruits to achieve the optimal balance between stability and flexibility.

[0006] (3) Weak environmental adaptability: In complex natural working environments, due to the different growth positions and orientations of fruits, as well as the obstruction and entanglement with branches and leaves, the robotic arm may be unable to contact the fruit in an ideal posture, resulting in partial failure of contact at the finger joints. Existing systems often have difficulty in making robust force compensation control for such "incomplete contact" states, which can easily lead to grasping failure or slippage of fruits and vegetables. Summary of the Invention

[0007] To address the problem that existing harvesting robots struggle to achieve stable and low-damage grasping of irregular and fragile targets in complex environments, particularly the lack of adaptive grasping force distribution based on real-time force sensing and the inability to maintain robustness in the event of partial contact failure, this invention designs a universal grasping quality evaluation index. This index enables the robot to dynamically and adaptively distribute the grasping force of each finger joint according to the weight, size, and real-time grasping posture of different targets, rather than relying on preset fixed force values ​​or trajectories.

[0008] To achieve the above objectives, the technical solution provided by this invention is as follows:

[0009] The aforementioned bionic harvesting robot and its grasping control method are characterized in that the robot includes three grasping fingers with joints, multiple pressure sensors set in the knuckles and palm, a drive servo motor that can provide feedback on the joint position, and an attitude sensor that can acquire the robot's attitude data.

[0010] The aforementioned capture control method includes the following steps:

[0011] S1. System Initialization and Data Acquisition: Initialize the system and acquire data from each sensor;

[0012] S2. Contact Detection and Flexible Contact: The target enters the grasping range by determining the threshold of the pressure sensor, and the admittance control algorithm is used to ensure flexible contact.

[0013] S3. Gravity Estimation: Based on the position coordinates of each contact point, estimate the gravity f acting on the fruit. g and its components in the preset coordinate system , ;

[0014] S4. Grasping metric calculation: Calculate the target grasping quality Q based on the attitude data. s The real-time capture metric Q is calculated using the following formula:

[0015]

[0016] Among them, the xy plane capture index Q p The calculation method is as follows:

[0017]

[0018] In the formula, convex hull , representing the set of contact forces With gravity set Minkowski and the boundary, in which Let E(W) be the set of edges of the convex hull, and d(O,e) be the distance from the origin O to edge e. for The vector after adding the gravity term;

[0019] The gripping index Q along the z-axis f The calculation method is as follows:

[0020]

[0021] In the formula, µ is the coefficient of friction at the contact point. Let k be the component of the fruit's weight on the z-axis. f This is a correction factor;

[0022] S5. Decision-making and execution: The real-time calculated Q-value is compared with the target capture quality Q. s The values ​​are compared, when Q≥Q s At that time, the system determines that the grasp is stable and locks each joint.

[0023] The aforementioned biomimetic harvesting robot and its grasping control method are characterized in that the target grasping mass Q s The method for determining it is as follows:

[0024] The gravity of the object to be grasped is set to point towards the palm of the robotic arm, and the normal force at each contact point reaches a preset reference value. Then, based on the real-time posture data and contact point coordinate data during grasping, the Q value in this state is calculated and set as Q. s .

[0025] The grasping control method for a bionic harvesting robot is characterized in that step S3, which estimates the gravity of the grasped object, includes:

[0026] (a) Fit a circular equation for the gripping cross section of the gripping object based on the spatial coordinates of at least three non-collinear contact points;

[0027] (b) Estimate the volume of the object to be grasped based on the fitted circle radius;

[0028] (c) Calculate gravity based on the known average density and estimated volume of the object to be grasped.

[0029] The bionic harvesting manipulator used in the control method is characterized by comprising: a rope-driven three-finger grasping mechanism, including a thumb, index finger, and middle finger, wherein the index finger and middle finger are both double-joint structures, and the proximal phalanges of the two fingers share a single drive joint.

[0030] The bionic harvesting robot is characterized in that the thin-film pressure sensor indirectly contacts the surface of the object being grasped through a flexible silicone sheet, and the contact surface of the flexible silicone sheet has an arched structure to enhance the sensitivity to small contact forces.

[0031] The bionic harvesting robot is characterized in that the drive rope of the distal phalanx passes through the interior of the proximal phalanx and through the rotation axis of the proximal phalanx, and finally wraps around the servo motor's rudder.

[0032] The bionic harvesting robot described above is characterized in that the main structure of the transmission base adopts an opening design at the top and bottom. The servo motor controlling the distal phalanx is installed from the top opening, and the servo motor controlling the proximal phalanx is installed from the bottom opening, which facilitates the installation and maintenance of the servo motor.

[0033] Compared with existing technologies, the bionic harvesting robot and its grasping control method provided by this invention have the following significant advantages:

[0034] (1) Achieve adaptive compliant gripping: By introducing a comprehensive gripping quality index Q as the control basis, the system can automatically adjust the gripping strategy according to the real-time status (weight, size, posture) of the gripping target, just like a human hand, and minimize the gripping force while ensuring stability, thus fundamentally reducing the risk of damage.

[0035] (2) Improved intelligence and robustness of grasping: This invention does not rely on precise preset models or fixed programs. It makes decisions based on real-time multi-sensor feedback, which can effectively cope with complex situations such as irregular shapes and variable growth positions of grasping targets. Even in a non-ideal state where some finger joints fail to contact, the system can maintain stability through force redistribution, and has strong environmental adaptability.

[0036] (3) Compact structure and sensitive sensing: The design of using rope drive and shared proximal joint significantly reduces the number of drive components and the overall volume. Combined with the arched flexible contact interface and thin film pressure sensor, it realizes high sensitivity and distributed measurement of weak contact force, providing a reliable data foundation for fine control.

[0037] (4) High versatility: The core of this invention lies in a universal force sensing and intelligent control framework, rather than a gripper for a specific shape. This robotic arm and control method can be widely used for grasping fruits and vegetables such as dragon fruit, peaches, tomatoes, and apples, and has good technical portability and application prospects. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of the overall structure of the bionic harvesting robot provided in the embodiment;

[0040] Figure 2 This is a schematic diagram of the internal cable-driven transmission principle of the robotic arm;

[0041] Figure 3 This is a magnified view of the contact point structure of a mechanical finger joint;

[0042] Figure 4 It involves capturing metrics Q and Q'. s Calculation diagram;

[0043] Figure 5 It is a flowchart for data capture;

[0044] Figure 6 These are experimental diagrams of a robotic arm grasping in multiple postures;

[0045] Figure 7 It is a bar chart of average contact force under different gripping postures;

[0046] Figure 8 This is a diagram of a gripping experiment when some knuckles fail to make contact.

[0047] Figure 9 This is a radar image of the contact force when some knuckles fail to make contact.

[0048] Numbered in the diagram: 1. Middle finger; 2. Index finger; 3. Distal knuckle drive rope; 4. Proximal knuckle of index finger; 5. Connecting bracket; 6. Proximal knuckle rudder; 7. Proximal knuckle drive rope; 8. Rudder; 9. Control module; 10. Transmission module; 11. Reset belt; 12. Proximal knuckle of thumb; 13. Thumb; 14. Thin-film pressure sensor; 15. Flexible silicone sheet; 101. First servo motor. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the embodiments of this invention will be described in detail below with reference to the accompanying drawings and specific examples. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of this invention.

[0050] Example 1: Structural Implementation of a Bionic Harvesting Robot

[0051] refer to Figure 1 and Figure 2This embodiment provides a bionic harvesting robot, the core of which is a rope-driven three-finger bionic grasping mechanism. The mechanism includes a transmission module (10), a control module (9), a thumb (13), an index finger (2), and a middle finger (1). The thumb (13) is a double-joint structure, and its proximal phalanx (12) is connected to the side of the drive module (10) through a rotating shaft. The index finger (2) and the middle finger (1) are both double-joint structures, including a proximal phalanx (4) and a distal phalanx, respectively. The proximal phalanxes (4) of the index finger and the middle finger are merged and connected to the transmission module (10), and are driven by the same servo motor through a wire rope. This design significantly reduces the number of drive units and achieves a compact structure.

[0052] All finger joints are driven by steel wire ropes. Specifically, the steel wire rope driving the distal finger joint is led out from its rotating part, passes through the internal cavity of the corresponding proximal finger joint, goes around the rotation center axis of the proximal finger joint, and is finally wound and fixed to the servo disk of the servo motor installed inside the transmission module (10). The steel wire rope driving the proximal finger joint is directly wound around the servo disk of the first servo motor (101). The main structure of the transmission module (10) adopts an open design at the top and bottom, which facilitates the installation and maintenance of the servo motor.

[0053] Regarding force sensing, refer to Figure 3 Thin-film pressure sensors are installed in each joint and palm area of ​​the robotic arm. Each sensor is covered with a specially made flexible silicone sheet (15). This flexible silicone sheet has a double-arched structure. When it comes into contact with the grasping target (if real), the arched structure amplifies the deformation caused by the small contact force, thereby improving the sensitivity of the thin-film pressure sensor to small force values. The sensor signal is sent to the main controller after passing through a linear voltage conversion module.

[0054] The control module (9) is also equipped with an attitude sensor (such as MPU6050) and a Bluetooth module, which are used to measure the three-dimensional attitude angle of the robot in real time and transmit data.

[0055] Example 2: Software Implementation of the Grabbing Control Method

[0056] Combination Figure 4 and Figure 5 This embodiment details the implementation process of the grasping control method on an embedded main controller (such as an STM32 series microcontroller), with dragon fruit as the grasping experimental object.

[0057] Step 1. System Initialization and Data Acquisition:

[0058] After the main controller powers on, it initializes all GPIOs, ADCs (analog-to-digital converters), timers, and communication interfaces (such as UART for Bluetooth and I2C for the attitude sensor). The system then enters a loop, executing the following tasks in real time:

[0059] The voltage values ​​converted from all thin-film pressure sensors are read by an ADC, and then converted into normal contact force according to the calibration curve. The robot arm's acceleration 'a' along the xyz axis is calculated by reading attitude sensor data via the I2C bus. x a y a z The current angle value θ of each servo motor is read via the servo motor protocol. i This serves as joint position feedback.

[0060] Step 2. Contact Determination and Flexible Contact Stage:

[0061] When the system detects a contact force exceeding a very small threshold (e.g., 0.3N) via the palm pressure sensor, it determines that the target has entered the grasping range and then enters the flexible contact phase. During this phase, an admittance control algorithm is used to ensure compliant contact.

[0062] Define a virtual desired position X. d (Initial value: gripper center). The measured contact force F... ext As the input to the admittance controller, according to the second-order admittance model Calculate the adjustment amount X for the desired position. d Among them, M d B d K d These represent the virtual mass, damping, and stiffness coefficients, respectively, with values ​​set to 0.01, 0.01, and 0.1. The adjusted X... d Converted into target angles q for each joint through inverse kinematics d The servo motor is driven by a PID position controller. This cycle continues, allowing the mechanical finger to smoothly conform to the dragon fruit surface under contact force constraints until stable contact is established at all expected contact points. The contact point coordinates (x, y, z) are calculated using joint angles and finger geometry models. i y i ).

[0063] Step 3. Gravity estimation and target threshold Q s set up:

[0064] Gravity estimation: Using the coordinates of the five contact points obtained in step 2, a circle is fitted using the least squares method to estimate the radius r of the fruit-grabbing cross section. Based on the approximate ellipsoidal model of the dragon fruit and its average density ρ, its gravity is estimated. , where k v The shape correction factor (take k) v =1.3), the density ρ of dragon fruit is taken as 1.008 g / cm³. 3 .

[0065] Setting Q s Assuming that the normal force at each contact point reaches a preset reference value 'a' at the current knuckle position, and based on the dragon fruit grasping experiment, 'a' is determined to be 1.5N. Selecting the ideal state where the fruit's gravity direction points towards the palm, and substituting this into the formula in step 4 to calculate the Q value, this value is set as the target grasping mass threshold Q. s This value represents the target state that stable crawling should achieve in the current crawling scenario.

[0066] Step 4. Real-time capture of quality indicator Q calculation and adaptive force control:

[0067] Calculate Q p The origin of the robot's coordinate system (xyz) is located at the grasping center. The normal contact force vectors of all currently valid contact points are... The estimated components of gravity in the grasping plane xy Perform operations to construct vector sets P and G. Calculate their Minkowski sums and generate the convex hull W. Calculate the distance from the origin O to each edge e of the convex hull W, and take the minimum value as Q. p Q p It visually reflects how easily the fruit resists being pushed in a plane.

[0068] Calculate Q f According to the formula Calculation. Where µ = 0.86 is the coefficient of friction between the flexible silicone sheet and the dragon fruit peel. To estimate the component of gravity along the gripping axis, k f =0.05 is the correction factor, Q f This reflects the safety margin to prevent the fruit from slipping off the z-axis.

[0069] Synthetic Q: Calculation This value comprehensively represents the overall stability of the current capture.

[0070] Step 5. Decision-making and Execution:

[0071] Compare the real-time calculated Q value with the target threshold Q. s . If Q<Q s The system was determined to have insufficient gripping stability. The controller then increases the target position command output to each servo motor by a certain increment, thereby increasing the contact force at each contact point. If Q≥Q s The controller determines that a stable gripping state has been reached. It locks the positions of each servo motor, maintains the current gripping force, and issues a gripping completion signal.

[0072] During the grasping process, if the system detects that the pressure sensor reading at a certain preset contact point is consistently zero, it determines that contact at that point is missing. In the subsequent Q...p In the calculation, the force vector corresponding to the missing point is set to zero. Since the Q value will decrease as a result, the control system will automatically increase the target grasping force at the remaining effective contact points to compensate for the stability provided by the missing point, thus achieving Q≥Q even under non-ideal conditions. s The goal.

[0073] Example 3: Grasping effect applied to dragon fruit harvesting

[0074] To verify the effectiveness of the robotic arm and its control method, a grasping experiment was conducted using dragon fruit weighing 450-550g as the object.

[0075] Multi-pose adaptive grasping: such as Figure 6 As shown, the robotic arm performs grasping movements in four different postures. The system can automatically calculate and distribute the grasping force of each finger joint. The experimental results are as follows: Figure 7 The results show that in posture 1 (gravity pointing towards the palm), the average gripping force is the lowest, at only about 1.86 N; in posture 2 (gravity pointing towards the index finger), the average gripping force increases to about 2.52 N to counteract the gravitational torque. This demonstrates that the Q index can effectively guide the system to adaptively distribute the contact force according to the gripping situation.

[0076] Robustness to missing contact points: Simulating knuckle contact failure, experiments are as follows. Figure 8 As shown, the simulation addresses the absence of the proximal index finger contact point, the distal index finger contact point, and the distal thumb contact point. The missing proximal index finger contact point is achieved by setting the contact force vector to zero in the program. From Figure 9 The contact force radar chart shows that the average grasping force is the lowest when all contact points are in contact, and the grasping force automatically increases when there are missing contact points to compensate. In all experiments with missing contact points, the grasping success rate was maintained at 100%, and no fruit slippage or damage occurred.

Claims

1. A biomimetic harvesting robot and its grasping control method, characterized in that, The robotic arm includes three jointed grasping fingers, multiple pressure sensors set in the knuckles and palm, a drive servo motor that can provide feedback on the joint position, and an attitude sensor that can acquire the robotic arm's attitude data. The aforementioned capture control method includes the following steps: S1. System Initialization and Data Acquisition: Initialize the system and acquire data from each sensor; S2. Contact Detection and Flexible Contact: The target enters the grasping range by determining the threshold of the pressure sensor, and the admittance control algorithm is used to ensure flexible contact. S3. Gravity Estimation: Based on the position coordinates of each contact point, estimate the gravity f acting on the fruit. g and its components in the preset coordinate system , ; S4. Grasping metric calculation: Calculate the target grasping quality Q based on the attitude data. s The real-time capture metric Q is calculated using the following formula:

2. Among them, The grasping index Q in the xy plane of the robot's coordinate system p The calculation method is as follows:

3. In the formula, convex hull , representing the set of contact forces With gravity set Minkowski and the boundary, in which Let E(W) be the set of edges of the convex hull, and d(O,e) be the distance from the origin O to edge e. for The vector after adding the gravity term; The gripping index Q along the z-axis f The calculation method is as follows:

4. In the formula, µ is the coefficient of friction at the contact point. Let k be the component of the fruit's weight on the z-axis. f This is a correction factor; S5. Decision-making and execution: The real-time calculated Q-value is compared with the target capture quality Q. s The values ​​are compared, when Q≥Q s At that time, the system determines that the grasp is stable and locks each joint.

5. The biomimetic harvesting robot and its grasping control method according to claim 1, characterized in that, The target grasping quality Q s The method for determining it is as follows: The gravity of the object to be grasped is set to point towards the palm of the robotic arm, and the normal force at each contact point reaches a preset reference value. Then, based on the real-time posture data and contact point coordinate data during grasping, the Q value in this state is calculated and set as Q. s .

6. The grasping control method for a bionic harvesting robot according to claim 1, characterized in that, Step S3, which involves estimating the gravity of the object to be grasped, includes: (a) Fit a circular equation for the gripping cross section of the gripping object based on the spatial coordinates of at least three non-collinear contact points; (b) Estimate the volume of the object to be grasped based on the fitted circle radius; (c) Calculate gravity based on the known average density and estimated volume of the object to be grasped.

7. A biomimetic harvesting robot used in implementing the control method as described in claim 1, characterized in that, include: The rope-driven three-finger grasping mechanism includes a thumb, index finger, and middle finger. The index and middle fingers are both double-jointed structures, and their proximal phalanges share a single drive joint.

8. The bionic harvesting robot according to claim 4, characterized in that, The thin-film pressure sensor indirectly contacts the surface of the object being grasped via a flexible silicone sheet. The contact surface of the flexible silicone sheet has an arched structure to enhance the sensitivity to small contact forces.

9. The bionic harvesting robot according to claim 4, characterized in that, The drive cable of the distal phalanx passes through the interior of the proximal phalanx and through the rotation axis of the proximal phalanx, eventually winding around the servo's rudder.

10. The bionic harvesting robot according to claim 4, characterized in that, The main structure of the transmission base adopts an open design at the top and bottom. The servo motor that controls the distal phalanx is installed from the top opening, and the servo motor that controls the proximal phalanx is installed from the bottom opening, which facilitates the installation and maintenance of the servo motor.

Images