Method for roughly sensing tail end force of under-actuated gripper based on current information of driver
By monitoring the driver current information of the underactuated gripper, and combining kinematic and dynamic models with an LSTM network, the problem of high cost of end-effector force sensors is solved, enabling coarse sensing of the end-effector force and improving the safety and performance of the system.
Patent Information
- Application Number
- CN202410704927.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-03
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-06-03
AI Technical Summary
Existing gripper end force sensors are expensive and complex to install, limiting their widespread use in industrial robots and automation systems.
By monitoring the driver current information of the underactuated gripper, and combining kinematic and dynamic models, an LSTM network is used to predict the gripper's attitude changes, establish a mapping relationship between the end force and the driver current, and achieve a rough perception of the end force.
Without increasing additional hardware costs, it enables simple and effective sensing of the force at the end of the gripper, improving the safety and performance of industrial robots and automation systems.
Smart Images

Figure CN118906042A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of mechanical grippers, and in particular relates to a method for roughly sensing the end force of an under-actuated gripper based on driver current information. Background Art
[0002] In today's industrial robots and automation systems, the end-of-gripper force sensing is essential for safe, precise and efficient operation. However, in some cases, the end-of-gripper force sensor is expensive and complex to install, which limits its widespread application in some fields. Summary of the invention
[0003] The present invention aims to provide a method for roughly sensing the end force of an under-actuated gripper based on driver current information to solve the above-mentioned technical problems.
[0004] In order to solve the above technical problems, the specific technical solution of the method for roughly sensing the end force of an underactuated gripper based on driver current information of the present invention is as follows:
[0005] A method for roughly sensing the end force of an underactuated grasping actuator based on driver current information comprises the following steps:
[0006] S1: Kinematics / dynamics modeling of underactuated grasping actuator: A theoretical model is constructed for the structure of the underactuated grasping actuator, and the position information and torque information of the driver of the underactuated grasping actuator are used as input, and the position of the end of the underactuated grasping actuator and the coupled motion state are used as output;
[0007] S2: Under-actuated grab actuator driver current information analysis: Directly obtain the position information of the driver output according to the driver encoder, and use the driver's current constant to obtain the driver's output torque, and use the filter to process the driver's current information;
[0008] S3: Posture prediction of underactuated grasping actuator: The posture of the finger is predicted by using the change of current information combined with the kinematic and dynamic models of the underactuated grasping gripper;
[0009] S4: Optimization of the end force accuracy of underactuated gripping actuators: Introducing machine learning to compensate for uncertainty errors and thus improve the prediction accuracy;
[0010] S5: Mapping the current information of the under-actuated gripper and the end force: Substitute the actual input torque of the under-actuated gripper actuator into the predicted gripper posture to obtain the end gripping force information.
[0011] Furthermore, the S1 uses an underactuated grasping actuator to construct a kinematic and dynamic model, and the establishment process is as follows:
[0012]
[0013] where t represents the input torque vector from the actuator and the torsion spring mounted on the joint; ω a represents the angular velocity vector of the torsion spring installed in each joint of the underactuated gripper actuator; F is the contact force, and the positive direction of these contact forces points to the object; v represents the projected velocity vector of the contact point; τ and are the vectors of torque and angular velocity of the bone joint respectively; Δθ represents the initial compression angle of the torsion spring; K represents the stiffness coefficient installed in the torsion spring.
[0014] Furthermore, after setting the operating speed and maximum operating current of the S2 driver, when the operating current of the driver after overcoming the load is less than the set maximum operating current, the actual operating speed of the driver will remain unchanged at the set value, and the actual operating current value of the driver will increase with the increase of the load; when the current value of the driver to overcome the load is greater than the set maximum operating current, the actual operating speed of the driver will gradually decrease until the driver enters a stalled state, and the actual current of the motor when it is stalled will fluctuate around the set maximum operating current value.
[0015] Further, the S3 includes: classifying the gripping situation and analyzing the corresponding structure, decoupling the motions of the three joints of the underactuated gripping actuator, and classifying the gripping mode switching situations into three categories according to the order in which the object contacts the three joints: the distal finger joint contacts the object first, the middle finger joint contacts the object first, and the proximal finger joint contacts the object first;
[0016] When the distal finger joint first contacts the object: the under-actuated grasping actuator will always maintain the parallel grasping mode, at this time the distal spring remains stationary, θ3 remains unchanged, and no load is generated when the motor moves, while the proximal spring moves, θ1 and θ2 change, and the two remain complementary. In this stage, an accurate mapping relationship is established between the driver position information and θ1 and θ2, and at this time the proximal spring will gradually relax from a taut state when the gripper is closed, so the load generated gradually decreases.
[0017] When the middle finger joint first contacts the object: the under-actuated grasping actuator maintains a parallel grasping mode before contacting the object, the distal spring remains stationary, and the proximal spring moves; after the middle finger joint contacts the object, the under-actuated grasping actuator switches to a wrapping grasping mode, the distal spring begins to move θ3, and the distal spring will gradually tighten, the load generated will gradually increase, and at this time the proximal spring will stop moving, θ1 and θ2 remain unchanged, and no load is generated. In this stage, an accurate mapping relationship is established between the driver position information and θ3.
[0018] When the proximal finger joint contacts the object first: the under-actuated grasping actuator maintains a parallel grasping mode before contacting the object, the distal spring remains stationary, and the proximal spring moves; when the proximal finger joint contacts the object, the under-actuated grasping actuator switches to the covering grasping mode, the distal spring starts to move, and is coupled with the movement of the proximal spring, θ2 and θ3 change and couple with each other, θ1 remains unchanged, this mode corresponds to the contact situation of case 1, in actual design, by increasing the stiffness coefficient of the proximal spring, so that only the distal spring moves after the under-actuated grasping actuator switches to the covering grasping mode, and is optimized to have only θ3 change, θ1 and θ2 remain unchanged, and θ i Establish an accurate mapping relationship with θ3.
[0019] Furthermore, S3 achieves accurate mapping relationships between the driver position information and θ1, θ2, and θ3 in the parallel grasping mode and the wrapping grasping mode by classifying the grasping. When the two modes are switched, θ1 changes from a variable to a fixed value. Therefore, obtaining the critical value of θ1 when the mode is switched becomes the key to closely linking the mapping relationship between the driver position information and θ1, θ2, and θ3 between the two modes.
[0020] The characteristics of the driver current changing with the load are combined with the movement of the distal spring and the proximal spring when the grasping mode is switched. The critical point of the grasping mode switching is predicted by the change of the current, and then the values of θ1, θ2, and θ3 when the mode is switched are predicted: in the parallel grasping mode, since the overall load gradually decreases, the actual current change of the motor is also small. At this time, there is an accurate mapping relationship between the driver position information and θ1 and θ2; when switching from parallel grasping to wrapping grasping mode, the overall load gradually increases and the current surges. Therefore, when the current surges, it is regarded as an under-driven grasping actuator contacting the object. At this time, there is an accurate mapping relationship between the driver position information and θ3;
[0021] The forget gate of LSTM is used to combine the information before the current time step and the information after the current time step to comprehensively judge whether the interval with large current fluctuations is forgotten. In the LSTM network model, the current information when the grasping makes θ1 form different angles is collected to form a data set. The collected information includes the actual value of the current, the first-order derivative of the current, and the actual angle of θ1. After the data collection is completed, the filtered current value and the first-order derivative of the current are input into the network. In the model training, the time point is marked according to the actual angle of θ1. When the actual angle of θ1 is smaller than the grasped object, the time point is marked as
[0022] "0", when the actual angle of θ1 is greater than the grasped object, the time point is marked as "1", that is, the data set is classified into two categories, and the model uses the mark of the time point as the output. If the output of the time point is "0"
[0023] It means that the under-actuated grasping actuator is in parallel grasping mode. If the output at a time point is "1", it means that the under-actuated grasping actuator is in wrapping grasping mode. In the actual use of this LSTM model, the current information is input in real time and the mark of the time point is output. When the output changes from "0" to "1", it means that the grasping mode is switched. At this time, the corresponding value of θ1 is the predicted value when the grasping mode is switched. Furthermore, the S4 uses a mathematical model to fit and compensate for the deviation between the θ1 predicted by the LSTM and the actual θ1, collects data on the critical value of θ1 when the mode is switched when grasping objects that form different angles with θ1 at different speeds, collects data on θ1 predicted by the LSTM network when the driver of the under-driven grasping actuator grasps different objects at different speeds, and fits the data in the xz plane and yz plane in turn, both of which are fitted using quadratic polynomials and quartic polynomials respectively, without overfitting. The obtained fitting plane and partial projections in the x and y directions are used. In actual application, the driver speed is compared with the predicted θ1 to obtain the predicted value of θ1 that needs to be compensated.
[0024] Furthermore, S5 substitutes the torque information obtained by the driver of the underactuated grasping actuator according to the processed current information and current constant, and the predicted gripper posture into formulas (1) and (2), so as to obtain the end grasping force information.
[0025] The method for roughly sensing the end force of an underactuated gripper based on driver current information of the present invention has the following advantages:
[0026] The present invention infers the gripping force of the end of the gripper by monitoring the actuator current of the gripper. This method exploits the correlation between the actuator current and the end force, thereby achieving a rough sense of the end force without the use of an additional end force sensor. The advantage of this technology lies in its simplicity and cost-effectiveness, by utilizing the existing actuator current monitoring equipment, the end force can be sensed without adding additional hardware costs. In summary, the present invention provides a simple and effective method for achieving a rough sense of the end force in an underactuated gripper, thereby providing a new solution for the safety and performance of industrial robots and automation systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 It is a schematic diagram of the structure of the underactuated grasping actuator of the present invention;
[0028] Figure 2 It is a schematic diagram of the parallel grasping mode structure of the underactuated grasping actuator of the present invention;
[0029] Figure 3 It is a schematic diagram of the structure of the underactuated grasping actuator of the present invention in the enveloping grasping mode;
[0030] Figure 4 It is a schematic flow chart of a method for roughly sensing the end force of an underactuated gripper based on driver current information of the present invention;
[0031] Figure 5 Schematic diagram of the LSTM network structure of the present invention;
[0032] Figure 6 This is a schematic diagram of the data effect result of θ1 predicted by the LSTM network of the present invention;
[0033] Figure 7 It is a schematic diagram of the fitting plane and partial projection in the x direction and the y direction of the present invention. DETAILED DESCRIPTION
[0034] In order to better understand the purpose, structure and function of the present invention, a method for roughly sensing the end force of an under-actuated gripper based on driver current information of the present invention is further described in detail below in conjunction with the accompanying drawings.
[0035] The rough sensing method of the end force of the underactuated gripper based on the driver current information of the present invention is constructed based on the underactuated gripping actuator in the paper Construction of a Multiple-DOF Underactuated Gripper with Force-Sensing via Deep Learning, such as Figure 1-3 As shown, each rotary pair of the connecting rod of the under-actuated grasping actuator is connected by a pin. To further reduce the grasping resistance, a roller bearing is installed at the pin joint. At the same time, to facilitate the configuration of the relative position of the fingers, a torsion spring and a mechanical stopper are used to ensure the relative position of their initial state. It is worth noting that the torsion springs on the left and right sides are completely independent. When grasping asymmetric objects, the grasping effect can be improved by adjusting the stiffness of the torsion springs on the left and right sides respectively.
[0036] According to the distance of each knuckle from the palm, the closest knuckle is called the proximal knuckle 1, the middle knuckle is called the middle knuckle 2, and the distal knuckle is called the distal knuckle 3. All three knuckles are driven by the outer connecting rod 4. A torsion spring is installed at the intersection of each knuckle. The torsion spring installed at the intersection of the distal knuckle and the middle knuckle is called the distal spring 5, and the torsion spring installed at the intersection of the proximal knuckle and the middle knuckle is called the proximal spring 6. The installed springs enable the underactuated grasping actuator to switch between the parallel grasping mode and the enveloping grasping mode.
[0037] When the distal finger joint contacts the object first, the three joints of the gripper will be constrained by two springs to maintain a parallel grasping mode ( Figure 2When the middle or proximal knuckle first contacts the object, the knuckles of the gripper will overcome the constraints of the springs and the gripper will switch to the envelope grasping mode ( Figure 3 As shown in the figure, however, since the gripper is driven by only one driving link, it is impossible to establish a mapping relationship between each knuckle in the envelope grasping mode and the driving link, and therefore it is impossible to obtain the posture of each knuckle in the envelope grasping mode. The present invention will use the current information of the under-driven grasping actuator driver to roughly estimate the posture of each knuckle in the envelope grasping mode.
[0038] like Figure 4 As shown, a method for roughly sensing the end force of an underactuated gripper based on driver current information of the present invention comprises the following steps:
[0039] S1: Kinematic / dynamic modeling of underactuated grasping actuator. A theoretical model is constructed based on the actual structure of the underactuated grasping actuator, with the position information and torque information of the driver of the underactuated grasping actuator as input, and the posture and coupled motion state of the end of the underactuated grasping actuator as output. The kinematic and dynamic models are constructed using the above underactuated grasping actuator, and the establishment process is as follows:
[0040]
[0041] where t represents the input torque vector from the actuator and the torsion spring mounted on the joint; ω a represents the angular velocity vector of the torsion spring installed in each joint of the underactuated gripper actuator; F is the contact force, and the positive direction of these contact forces points to the object; v represents the projected velocity vector of the contact point; τ and are the vectors of torque and angular velocity of the bone joint respectively; Δθ represents the initial compression angle of the torsion spring; K represents the stiffness coefficient installed in the torsion spring.
[0042] S2: Under-driven grab actuator driver current information analysis. This method requires that the position, speed, and current information of the driver can be read according to the encoder of the driver, and the maximum operating current, operating speed, and position of the driver can be set. At the same time, the current constant of the driver (unit N*m) can also be used to obtain the output torque of the driver, but because the current actually output by the driver is a curve with large fluctuations, a filter is used to process the current information of the driver. After setting the operating speed and maximum operating current of this type of driver, when the current running after the driver overcomes the load is less than the set maximum operating current, the actual operating speed of the driver will remain unchanged at the set value, and the actual operating current value of the driver will increase with the increase of the load; when the current value of the driver overcoming the load is greater than the set maximum operating current, the actual operating speed of the driver will gradually decrease until the driver enters the stall state (the driver speed is 0rpm), and the actual current of the motor when it is stalled will fluctuate around the set maximum operating current value.
[0043] S3: Posture prediction of underactuated grasping actuator. Since the degrees of freedom of the underactuated grasping actuator are greater than the number of actuators, the multiple joints of the actuator fingers cannot move independently of each other, and the movement of each finger joint cannot be obtained separately through the input of a single actuator. This method uses the change of current information combined with the kinematic and dynamic models of the underactuated grasping gripper to predict the posture of the fingers.
[0044] The torsion spring installed at the intersection of each finger joint enables the underactuated gripping actuator to switch between parallel gripping mode and wrapping gripping mode. However, due to the coupling relationship between θ1, θ2, and θ3, it is impossible to establish an accurate mapping relationship between the driver position information and θ1, θ2, and θ3 when the gripper moves. Therefore, by classifying the gripping situation and analyzing the corresponding structure, the movement of the three joints of the underactuated gripping actuator is decoupled, and the gripping mode switching situations are divided into three categories according to the order in which the object contacts the three joints:
[0045] 1. The distal finger joint contacts the object first: the under-actuated grasping actuator will always maintain the parallel grasping mode. At this time, the distal spring remains stationary (θ3 remains unchanged) and no load is generated when the motor moves, while the proximal spring moves (θ1 and θ2 change, and the two remain complementary). In this stage, an accurate mapping relationship can be established between the driver position information and θ1 and θ2. At this time, the proximal spring will gradually relax from the taut state when the gripper is closed, so the load generated gradually decreases.
[0046] 2. The middle finger joint contacts the object first: the under-actuated grasping actuator maintains a parallel grasping mode before contacting the object, the distal spring remains stationary, and the proximal spring moves; after the middle finger joint contacts the object, the under-actuated grasping actuator switches to the wrapped grasping mode, the distal spring starts to move (θ3 changes), and the distal spring will gradually tighten, and the load generated will gradually increase. At this time, the proximal spring will stop moving (θ1 and θ2 remain unchanged) and no longer generate load. At this stage, an accurate mapping relationship can be established between the driver position information and θ3.
[0047] 3. The proximal finger joint contacts the object first: the under-actuated grasping actuator maintains a parallel grasping mode before contacting the object, the distal spring remains stationary, and the proximal spring moves; when the proximal finger joint contacts the object, the under-actuated grasping actuator switches to the enveloping grasping mode, the distal spring starts to move, and couples with the movement of the proximal spring (θ2 and θ3 change and couple with each other, θ1 remains unchanged). This mode corresponds to the contact situation in case 1. After conducting a large number of grasping experiments, it can be found that the situation where the proximal finger contacts first is relatively rare. In actual design, by increasing the stiffness coefficient of the proximal spring, only the distal spring moves after the under-actuated grasping actuator switches to the enveloping grasping mode (optimized to only change θ3, θ1 and θ2 remain unchanged), θ i Establish an accurate mapping relationship with θ3.
[0048] By classifying the grasping, it is possible to establish accurate mapping relationships between the driver position information and θ1, θ2, and θ3 in the parallel grasping mode and the wrapping grasping mode, respectively. When the two modes are switched, θ1 changes from a variable to a fixed value. Therefore, obtaining the critical value of θ1 when the mode is switched becomes the key to closely connecting the mapping relationship between the driver position information and θ1, θ2, and θ3 between the two modes.
[0049] Combining the characteristic that the driver current changes with the load with the movement of the distal spring and the proximal spring when the grasping mode is switched, the critical point of the grasping mode switching can be predicted by the change in current, and then the values of θ1, θ2, and θ3 when the mode is switched can be predicted: in the parallel grasping mode, since the overall load gradually decreases, the actual current change of the motor is also small. At this time, there is an accurate mapping relationship between the driver position information and θ1 and θ2; when switching from parallel grasping to wrapping grasping mode, the overall load gradually increases and the current surges. Therefore, when the current surges, it is regarded as an under-driven grasping actuator contacting the object. At this time, there is an accurate mapping relationship between the driver position information and θ3.
[0050] However, due to the random fluctuation of current, when the current fluctuation is large, there may be a situation where the actual current is close to or even exceeds the mode switching current threshold even if the grasping mode of the under-actuated grasping actuator is not switched, which may affect the prediction accuracy of θ1 during mode switching, or even cause misjudgment. The forget gate of LSTM can combine the information before the current time step and the information after the current time step to comprehensively judge whether the interval with large current fluctuation is forgotten. In the LSTM network model of this experiment, the current information when grasping makes θ1 form different angles is collected to form a data set. The collected information includes the actual current value, the first-order derivative of the current, and the actual angle of θ1. After the data collection is completed, the filtered current value and the first-order derivative of the current are input into the network. In the model training, the time point is marked according to the actual angle of θ1. When the actual angle of θ1 is less than the grasped object, the time point is marked as "0", and when the actual angle of θ1 is greater than the grasped object, the time point is marked as "1", that is, the data set is classified into two categories. The model uses the mark of the time point as output. If the output of the time point is "0", it means that the underactuated grasping actuator is in the parallel grasping mode. If the output of the time point is "1", it means that the underactuated grasping actuator is in the wrapped grasping mode. In the actual use of this LSTM model, the current information is input in real time and the mark of the time point is output. When the output changes from "0" to "1", it means that the grasping mode has switched. At this time, the corresponding value of θ1 is the predicted value when the grasping mode switches. The LSTM network structure is as follows: Figure 5 shown.
[0051] In summary, the critical value of θ1 during mode switching can be predicted by the change in current.
[0052] S4: Optimization of the end force accuracy of the under-actuated grasping actuator. The θ1 directly predicted by the LSTM network model still has a certain deviation from the critical value of θ1 during the actual mode switching due to the delay in using the mean filter to process the current and the non-geometric errors of the mechanical structure (such as tooth clearance, axle pin fit, etc.). Therefore, machine learning is introduced to compensate for such uncertain errors, thereby improving the prediction accuracy. A mathematical model is used to fit and compensate the deviation between the θ1 predicted by LSTM and the actual θ1. The critical value of θ1 predicted during mode switching is collected when grasping objects that form different angles for θ1 at different speeds. A certain amount of data on θ1 predicted by the LSTM network when the driver of the under-actuated grasping actuator grasps different objects at different speeds is collected. The results are shown as follows: Figure 6The x-axis represents the motor speed, the y-axis represents the size of the grasped object, and the z-axis represents the deviation between the θ1 predicted by LSTM and the actual θ1. The data is fitted on the xz plane and the yz plane in turn. Using quadratic polynomials and quartic polynomials respectively can achieve better fitting results without overfitting. The obtained fitting plane and its partial projections in the x and y directions are shown in Figure 7 In practical applications, the drive speed and the predicted θ1 can be used to obtain the value of the predicted θ1 that needs to be compensated.
[0053] S5: Mapping the current information of the underactuated gripper and the end force. Substituting the torque information (obtained in S2) obtained by the driver of the underactuated gripper actuator according to the processed current information and current constant, and the predicted gripper posture (obtained in S4 and S5) into formulas (1) and (2), the end gripping force information can be obtained.
[0054] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.
Claims
1. A rough sensing method of the end force of an underactuated gripping actuator based on driver current information, characterized in that: The steps include: S1: Kinematics / dynamics modeling of underactuated grasping actuator: A theoretical model is constructed for the structure of the underactuated grasping actuator, and the position information and torque information of the driver of the underactuated grasping actuator are used as input, and the position of the end of the underactuated grasping actuator and the coupled motion state are used as output; S2: Under-actuated grab actuator driver current information analysis: Directly obtain the position information of the driver output according to the driver encoder, and use the driver's current constant to obtain the driver's output torque, and use the filter to process the driver's current information; S3: Posture prediction of underactuated grasping actuator: The posture of the finger is predicted by using the change of current information combined with the kinematic and dynamic models of the underactuated grasping gripper; S4: Optimization of the end force accuracy of underactuated gripping actuators: Introducing machine learning to compensate for uncertainty errors and thus improve the prediction accuracy; S5: Mapping the current information of the under-actuated gripper and the end force: Substitute the actual input torque of the under-actuated gripper actuator into the predicted gripper posture to obtain the end gripping force information.
2. The method for roughly sensing the end force of an underactuated gripping actuator based on driver current information according to claim 1 is characterized in that: The S1 uses an underactuated grasping actuator to build a kinematic and dynamic model. The establishment process is as follows: where t represents the input torque vector from the actuator and the torsion spring mounted on the joint; ω a represents the angular velocity vector of the torsion spring installed in each joint of the underactuated gripper actuator; F is the contact force, and the positive direction of these contact forces points to the object; v represents the projected velocity vector of the contact point; τ and are the vectors of torque and angular velocity of the bone joint respectively; Δθ represents the initial compression angle of the torsion spring; K represents the stiffness coefficient installed in the torsion spring.
3. The method for roughly sensing the end force of an underactuated gripping actuator based on driver current information according to claim 1, characterized in that: After the operating speed and the maximum operating current of the driver of S2 are set, when the current running after the driver overcomes the load is less than the set maximum operating current, the actual operating speed of the driver will remain unchanged at the set value, and the actual operating current value of the driver will increase with the increase of the load; When the current value of the driver to overcome the load is greater than the set maximum operating current, the actual operating speed of the driver will gradually decrease until the driver enters a stalled state, and the actual current of the motor when stalled will fluctuate around the set maximum operating current value.
4. The method for roughly sensing the end force of an underactuated gripping actuator based on driver current information according to claim 1, characterized in that: S3 includes: classifying the gripping situation and analyzing the corresponding structure, decoupling the motion of the three joints of the underactuated gripping actuator, and classifying the gripping mode switching situations into three categories according to the order in which the object contacts the three joints: the distal finger joint contacts the object first, the middle finger joint contacts the object first, and the proximal finger joint contacts the object first; When the distal finger joint first contacts the object: the under-actuated grasping actuator will always maintain the parallel grasping mode, at this time the distal spring remains stationary, θ3 remains unchanged, and no load is generated when the motor moves, while the proximal spring moves, θ1 and θ2 change, and the two remain complementary. In this stage, an accurate mapping relationship is established between the driver position information and θ1 and θ2, and at this time the proximal spring will gradually relax from a taut state when the gripper is closed, so the load generated gradually decreases. When the middle finger joint first contacts the object: the under-actuated grasping actuator maintains a parallel grasping mode before contacting the object, the distal spring remains stationary, and the proximal spring moves; after the middle finger joint contacts the object, the under-actuated grasping actuator switches to a wrapping grasping mode, the distal spring begins to move θ3, and the distal spring will gradually tighten, the load generated will gradually increase, and at this time the proximal spring will stop moving, θ1 and θ2 remain unchanged, and no load is generated. In this stage, an accurate mapping relationship is established between the driver position information and θ3. When the proximal finger joint contacts the object first: the under-actuated grasping actuator maintains a parallel grasping mode before contacting the object, the distal spring remains stationary, and the proximal spring moves; when the proximal finger joint contacts the object, the under-actuated grasping actuator switches to the covering grasping mode, the distal spring starts to move, and is coupled with the movement of the proximal spring, θ2 and θ3 change and couple with each other, θ1 remains unchanged, this mode corresponds to the contact situation of case 1, in actual design, by increasing the stiffness coefficient of the proximal spring, so that only the distal spring moves after the under-actuated grasping actuator switches to the covering grasping mode, and is optimized to have only θ3 change, θ1 and θ2 remain unchanged, and θ i Establish an accurate mapping relationship with θ3.
5. The method for roughly sensing the end force of an underactuated gripping actuator based on driver current information according to claim 1, characterized in that: S3 achieves accurate mapping relationships between the driver position information and θ1, θ2, and θ3 in the parallel grasping mode and the wrapping grasping mode by classifying the grasping. When the two modes are switched, θ1 changes from a variable to a fixed value. Therefore, obtaining the critical value of θ1 when the mode is switched becomes the key to closely linking the mapping relationship between the driver position information and θ1, θ2, and θ3 between the two modes. The characteristics of the driver current changing with the load are combined with the movement of the distal spring and the proximal spring when the grasping mode is switched. The critical point of the grasping mode switching is predicted by the change of the current, and then the values of θ1, θ2, and θ3 when the mode is switched are predicted: in the parallel grasping mode, since the overall load gradually decreases, the actual current change of the motor is also small. At this time, there is an accurate mapping relationship between the driver position information and θ1 and θ2; when switching from parallel grasping to wrapping grasping mode, the overall load gradually increases and the current surges. Therefore, when the current surges, it is regarded as an under-driven grasping actuator contacting the object. At this time, there is an accurate mapping relationship between the driver position information and θ3; The forget gate of LSTM is used to combine the information before the current time step and the information after the current time step to comprehensively judge whether the interval with large current fluctuations is forgotten. In the LSTM network model, the current information when the grasping makes θ1 form different angles is collected to form a data set. The collected information includes the actual value of the current, the first-order derivative of the current, and the actual angle of θ1. After the data collection is completed, the filtered current value and the first-order derivative of the current are input into the network. In the model training, the time point is marked according to the actual angle of θ1. When the actual angle of θ1 is smaller than the grasped object, the time point is marked as "0", when the actual angle of θ1 is greater than the grasped object, the time point is marked as "1", that is, the data set is classified into two categories, and the model uses the mark of the time point as the output. If the output of the time point is "0", it means that the under-actuated grasping actuator is in the parallel grasping mode. If the output of the time point is "1", it means that the under-actuated grasping actuator is in the wrapped grasping mode. In the actual use of this LSTM model, the current information is input in real time, and the mark of the time point is output. When the output changes from "0" to "1", it means that the grasping mode has switched. At this time, the corresponding value of θ1 is the predicted value when the grasping mode is switched.
6. The method for roughly sensing the end force of an underactuated gripping actuator based on driver current information according to claim 1, characterized in that: The S4 uses a mathematical model to fit and compensate for the deviation between the θ1 predicted by the LSTM and the actual θ1, collects data on the critical value of θ1 when grasping objects that form different angles with θ1 at different speeds, and predicts the critical value of θ1 when the mode is switched. The data of θ1 predicted by the LSTM network when the driver of the under-driven grasping actuator grasps different objects at different speeds are collected, and the data are fit in the xz plane and the yz plane in turn. The fitting is achieved using a quadratic polynomial and a quartic polynomial respectively, and overfitting is not generated. The fitting plane and partial projections in the x and y directions are obtained. In actual application, the driver speed is compared with the predicted θ1 to obtain the value of the predicted θ1 that needs to be compensated.
7. The method for roughly sensing the end force of an underactuated gripping actuator based on driver current information according to claim 2, characterized in that: S5 substitutes the torque information obtained by the driver of the underactuated grasping actuator according to the processed current information and current constant, and the predicted gripper posture into formulas (1) and (2), so as to obtain the end grasping force information.
Citation Information
Patent Citations
Method for estimating rigidity of objects grabbed by under-actuated manipulator
CN108214558A
Perceptual system for state of underactuated dexterous hand transmission system
CN108908345A
Manipulator underactuated driving structure applied to piano teaching and design method
CN110394784A
Dexterous hand with tactile feedback and variable rigidity and control method thereof
CN113146665A
KR20240007968A
