Training data acquisition apparatus for robot gripper control

Abstract

This application relates to the field of robotics, and in particular to a training data acquisition device for robot gripper control. The training data acquisition device mainly comprises four core modules: the acquisition device body, an image acquisition module (51), a motion detection module (52), and a data processing module (53). In use, the operator holds the device, aligns the mechanical finger (2) with the object to be grasped, and presses the handle (3) on the acquisition device body to control the mechanical finger (2) to close and grasp the object. During this process, the image acquisition module (51) records visual information, and the motion detection module (52) records the degree of finger opening and closing. The data processing module (53) aggregates these two types of heterogeneous data. Thus, the hand-eye coordination process of a real robot can be simulated at extremely low hardware cost, and key matching data is provided for robot imitation learning.

Description

Technical Field

[0001] This application relates to the field of robotics, and in particular to a training data acquisition device for robot gripper control. Background Technology

[0002] With the rapid development of artificial intelligence and robotics, robots are increasingly being used in industrial production, logistics sorting, and household services. Among the many skills of robots, object grasping and manipulation are among the most fundamental and core capabilities. Traditional robot grasping control usually relies on manual hard coding or model-based control methods, which often lack flexibility and versatility when dealing with objects that are irregularly shaped, of diverse materials, or in complex environments.

[0003] To address these issues, visual servoing grasping technology based on deep learning and imitation learning has gradually become mainstream. This approach requires training neural network models by collecting a large amount of human teaching data (i.e., "training data"), enabling robots to make autonomous decisions and execute grasping actions based on visual input.

[0004] In existing technologies, the collection of such training data mainly relies on remote operation using real robotic entities: operators control real robotic arms and grippers to grasp objects via teach pendants or master-slave hand devices, while simultaneously recording sensor data. Although this method provides high data fidelity, the high cost, large size, and limited mobility of industrial robotic arms or collaborative robots restrict the data collection sites, making it difficult to collect data on a large scale in diverse everyday or unstructured scenarios. Summary of the Invention

[0005] In view of this, the present invention proposes a training data acquisition device for robot gripper control, in order to solve the problems of high cost and bulky equipment in traditional robot training data acquisition.

[0006] In the training data acquisition device for robot gripper control provided by the present invention, the training data acquisition device includes an acquisition device body, an image acquisition module, a motion detection module, and a data processing module. The acquisition device body includes a frame, a pressing handle connected to the frame, and multiple mechanical fingers that are driven to open and close by the pressing handle. The image acquisition module is mounted on the frame and acquires image data of the gripping areas of the multiple mechanical fingers during the gripping action. The motion detection module is mounted on the frame and detects the motion parameters of the mechanical fingers in real time during the gripping action. The data processing module is mounted on the frame and receives and processes the image data and the motion parameters to generate training data.

[0007] Technical Benefits: By directly manipulating the mechanical structure to simulate the robot's grasping motion, a large number of training samples for robot learning can be quickly accumulated without complex teleoperation programming or expensive motion capture equipment. This achieves low-cost and high-efficiency data acquisition. Furthermore, by synchronously acquiring image data and motion parameters (mechanical finger state) on the same hardware architecture, it realistically reproduces the physical process of "hand-eye-brain" coordination, solving the problem of the disconnect between visual perception and execution in training data and ensuring the consistency of multimodal data. In addition, using physical mechanical fingers for grasping can realistically reflect the physical feedback (such as deformation and sliding) when an object comes into contact with the gripper, making it more robust than pure simulation data and able to simulate real physical characteristics.

[0008] In a preferred embodiment of the training data acquisition device provided by the present invention, the data processing module is configured to: add timestamps to the acquired image data and the detected motion parameters respectively, and align the image data of each frame with the corresponding motion parameters based on the timestamps.

[0009] Technical Benefits: Visual sensors (such as cameras) and motion sensors (such as encoders) typically have different sampling frequencies. Timestamp alignment ensures that each frame accurately corresponds to the current opening and closing state of the mechanical finger, solving the problem of asynchronous data processing. Furthermore, it provides a precise temporal correspondence for the end-to-end robot control model, avoiding oscillations or misjudgments in the actual deployment of the trained model due to latency, thus improving model training accuracy.

[0010] In a preferred embodiment of the training data acquisition device provided by the present invention, the data processing module is configured to: before associating and aligning the image data of each frame with the corresponding motion parameters, further include using a pre-stored mapping relationship to convert the original detection signal output by the motion detection module into a physical distance value or angle value reflecting the opening amplitude of the mechanical finger.

[0011] Technical benefits: Directly outputting physically meaningful parameters (such as "open 5 cm") makes the collected data independent of specific sensor models, facilitating data reuse across devices and enhancing data versatility and interpretability. Furthermore, the preprocessing step reduces the "sensor noise to physical state" mapping process that the neural network needs to learn, allowing the model to focus more on learning the grasping strategy itself and reducing the learning difficulty.

[0012] In a preferred embodiment of the training data acquisition device provided by the present invention, the data processing module is configured to store or process image data and motion parameters during the period when the mechanical finger undergoes displacement changes, based on the monitored changes in the motion parameters, as effective training data.

[0013] Technical benefits: Automatically filters out invalid frames in static states (such as when the operator is resting or not operating), significantly reducing storage space usage and effectively eliminating redundant data. Furthermore, it ensures that every frame in the generated training set contains valid action information, preventing the model from overfitting to "static" states during training, thus improving sample "density" and training efficiency.

[0014] In a preferred embodiment of the training data acquisition device provided by the present invention, the frame includes a housing assembly and a fixed handle connected to the housing assembly; a plurality of mechanical fingers are hinged to the housing assembly, and a transmission assembly is provided between the plurality of mechanical fingers and the pressing handle.

[0015] Technical benefits: The grip, similar to a "spray gun" or "pistol," is ergonomic, allowing for extended one-handed operation and reducing fatigue. Furthermore, it transforms the pressing motion of the human hand into the opening and closing of mechanical fingers, providing intuitive operation and facilitating precise control of gripping force and speed.

[0016] In a preferred embodiment of the training data acquisition device provided by the present invention, the housing assembly includes a front end plate and a rear end plate; the transmission assembly includes a guide rod, a moving plate, and a spring. The guide rod is disposed within the housing assembly and connected between the front end plate and the rear end plate; the moving plate is slidably disposed on the guide rod and is also hinged to the input ends of the plurality of mechanical fingers; the moving plate is also connected to a pressing handle, and is configured such that when the pressing handle is pressed, the moving plate drives the plurality of mechanical fingers to close; the spring is sleeved on the guide rod and is configured to cause the plurality of mechanical fingers to open under its own restoring force.

[0017] Technical benefits: The guide rod constrains the degrees of freedom of the moving plate, ensuring accurate transmission of driving force, reducing mechanical jamming, and giving the moving plate smooth and linear movement. Furthermore, the spring return force keeps the mechanical fingers in an open state by default; the operator only needs to apply force to grasp, and releases automatically, simplifying the operation process, simulating common gripper logic, and achieving automatic gripper reset (normally open logic).

[0018] In a preferred embodiment of the training data acquisition device provided by the present invention, the spring is a compression spring, and the spring is disposed between the moving plate and the rear end plate of the housing assembly.

[0019] Technical benefits: The concealed design fully utilizes the axial space within the housing, preventing exposed springs from interfering with operation or pinching the operator, resulting in a compact structure for the training data acquisition device. Furthermore, the linear reaction force of the compression spring provides a clear damping sensation to the operator's fingers, helping them perceive the current pressing depth (i.e., the degree of gripper closure) and providing force feedback.

[0020] In a preferred embodiment of the training data acquisition device provided by the present invention, the upper end plate of the housing assembly is provided with an upper pivot hole, and the lower end plate is provided with a lower pivot hole; the mechanical finger includes an input link, an intermediate link, an output link, a finger tip, and an auxiliary link. One end of the input link is hinged to the moving plate; one end of the intermediate link is hinged to the other end of the input link, and a rotating shaft is fixedly mounted on it, with both ends of the rotating shaft rotatably disposed in the upper pivot hole and the lower pivot hole, respectively; one end of the output link is hinged to the other end of the intermediate link; the finger tip is hinged to the other end of the output link; the auxiliary link is located inside the mechanical finger relative to the intermediate link, with one end hinged to the finger tip and the other end hinged to the housing assembly.

[0021] Technical benefits: Through linkage mechanisms (such as parallelogram mechanisms or specifically optimized four-bar linkages), the fingertips can move in translational or specific trajectories, ensuring stability of the contact surface between the fingertips and the object during grasping, preventing the object from slipping, and optimizing the fingertip trajectory. Furthermore, the linkage articulation method has better rigidity than a simple cantilever beam structure, can withstand greater grasping forces, and has high structural strength.

[0022] In a preferred embodiment of the training data acquisition device provided by the present invention, the mechanical finger further includes a torsion spring, which is disposed at the hinge between the intermediate connecting rod and the output connecting rod and is configured to keep the mechanical finger in an extended state under its own restoring force.

[0023] Technical benefits: When the robotic finger comes into contact with a rigid object, the torsion spring allows for a certain degree of elastic cushioning, protecting the transmission components from jamming or damage. Furthermore, in conjunction with a specific linkage design, the torsion spring enables the finger to adaptively deform when enveloping objects of different shapes, increasing the success rate of gripping.

[0024] In a preferred embodiment of the training data acquisition device provided by the present invention, a mechanical finger is provided on each of the two front ends of the housing assembly, and the input link is located inside the housing assembly, with one end of the intermediate link located inside the housing assembly and the other end located outside the housing assembly.

[0025] Technical benefits: The symmetrical design conforms to the most common two-finger parallel gripper shape in the industry, and the collected data has high universal value. In addition, hiding some of the connecting rods inside the housing reduces the overall size of the device and prevents external connecting rods from colliding and interfering in complex gripping environments (such as gripping inside a box).

[0026] In a preferred embodiment of the training data acquisition device provided by the present invention, the motion detection module includes a code disk and a reading head. The code disk is connected to one end of the rotating shaft extending from the housing assembly; the reading head is coaxially disposed opposite to the code disk and fixedly disposed relative to the housing assembly.

[0027] Technical benefits: By directly measuring the rotation angle of the shaft, the transmission backlash error that may exist when measuring the displacement of a linear push rod is avoided, achieving high-precision angle measurement. Furthermore, the code disk and reading head are typically coupled photoelectrically or magnetically, eliminating mechanical wear, extending the lifespan of the sensing system, and ensuring stable readings.

[0028] In a preferred embodiment of the training data acquisition device provided by the present invention, an end cap is formed on the outer side of the upper end of the fixed handle, the data processing module is disposed inside the end cap, and the end cap is connected to the rear end plate of the housing assembly; and the inner side of the upper end of the fixed handle is connected to the lower end plate of the housing assembly.

[0029] Technical Benefits: Placing heavier data processing modules (such as PCB boards and batteries) at the bottom of the handle lowers the device's center of gravity, making handheld operation more stable and effortless. Furthermore, utilizing the handle's internal space for wiring and encapsulation avoids exposed and tangled cables, improving the device's portability and electrical safety. Additionally, the fixed handle connects two adjacent surfaces of the housing assembly, ensuring a more secure connection between the handle and the housing.

[0030] In a preferred embodiment of the training data acquisition device provided by the present invention, the frame further includes a support plate, one end of which is connected to the housing assembly, and the other end of which is connected to the image acquisition module.

[0031] Technical benefits: The rigid connection ensures a fixed relative position between the camera and the gripper base, which is crucial for training visual servoing-based robot algorithms and creates an "eye-in-hand" perspective. Furthermore, the length and angle design of the support plate ensures that the camera fully covers the gripper's working area while preventing the gripper itself from excessively obstructing the object being grasped, thus achieving field of view (FOV) optimization. Attached Figure Description

[0032] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which will make the above and other features and advantages of the present invention more apparent to those skilled in the art. In the drawings:

[0033] Figure 1 This is a schematic diagram of the external structure of the training data acquisition device for robot gripper control in this embodiment.

[0034] Figure 2 This is a schematic diagram showing the positions of the motion detection module and the data processing module in the training data acquisition device of this embodiment.

[0035] Figure 3 This is a schematic diagram of the connection structure between the mechanical finger and the transmission component in the training data acquisition device of this embodiment.

[0036] Figure 4 This is a schematic diagram of the assembly structure of the motion detection module in the training data acquisition device of this embodiment.

[0037] The accompanying figure is labeled as follows:

[0038] 1-Rack;

[0039] 11-Housing assembly; 111-Front end plate; 112-Rear end plate; 113-Upper end plate; 114-Lower end plate;

[0040] 12-Fixed handle; 121-End cap;

[0041] 13-Support plate;

[0042] 2-Mechanical finger; 21-Input link; 22-Intermediate link; 221-Rotating shaft; 23-Output link; 24-Finger tip; 25-Auxiliary link; 26-Torsion spring;

[0043] 3-Press the handle;

[0044] 4-Transmission assembly; 41-Guide rod; 42-Moving plate; 43-Spring;

[0045] 51-Image Acquisition Module;

[0046] 52-Motion detection module; 521-Code disk; 522-Reading head;

[0047] 53-Data Processing Module. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0049] It should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "connection" and "configuration" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0050] Furthermore, it should be understood in the description of this application that the terms “center,” “upper,” “lower,” “front,” “rear,” “left,” “right,” “vertical,” “horizontal,” “top,” “bottom,” “inner,” and “outer,” etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application.

[0051] Example 1: Overall Device Architecture

[0052] Reference Figure 1 This embodiment provides a training data acquisition device for robot gripper control. This device aims to collect high-quality data for training robot grasping algorithms in a low-cost and efficient manner through manual operation.

[0053] The acquisition device mainly includes four core modules: the acquisition device body, the image acquisition module 51, the motion detection module 52, and the data processing module 53.

[0054] The data acquisition device itself forms the mechanical foundation of the entire system. It comprises a frame 1 (serving as a supporting skeleton), a push handle 3 connected to the frame 1 (serving as an input for human-machine interaction), and multiple mechanical fingers 2 (serving as end effectors for performing grasping actions). The opening and closing movements of the mechanical fingers 2 are driven by the push handle 3, meaning the operator simulates the robot's grasping commands by pressing the push handle 3.

[0055] Combination Figure 1 and Figure 2 The image acquisition module 51 is mounted on the frame 1, with its lens facing the working area of ​​the robotic finger 2. Its function is to continuously acquire image data (such as RGB video streams or depth images) containing the robotic finger 2 and the object being grasped throughout the entire process of the robotic finger 2 performing the grasping action.

[0056] The motion detection module 52 is also mounted on the frame 1 to monitor the motion state of the mechanical finger 2 in real time and output motion parameters that reflect the finger position or posture.

[0057] The data processing module 53 is integrated on the rack 1 and serves as the computing core. It is responsible for receiving image data and motion parameters and processing them to generate the final training data.

[0058] Operating Procedure: During use, the operator holds the device, aligns the mechanical finger 2 with the object to be grasped, and presses the handle 3 to control the mechanical finger 2 to close and grasp the object. During this process, the image acquisition module 51 records visual information, and the motion detection module 52 records the degree of finger opening and closing. The data processing module 53 aggregates these two types of heterogeneous data.

[0059] Technical benefits: This training data acquisition device can simulate the hand-eye coordination process of a real robot at extremely low hardware cost. The acquired data not only includes visual information, but also precise kinematic labels (i.e., "what opening and closing state the gripper is in when the image is seen"), thus providing key matching data for robot imitation learning.

[0060] Example 2: Specific Implementation of the Mechanical Structure

[0061] Combination Figure 2 and Figure 3 This embodiment describes in detail the specific mechanical structure of the data acquisition device and its transmission principle.

[0062] 1. Frame and handle assembly

[0063] The frame 1 mainly consists of a housing assembly 11 and a fixed handle 12. The housing assembly 11 serves to house the transmission mechanism, while the fixed handle 12 allows the user to hold it with one hand.

[0064] The housing assembly 11 includes a front end plate 111, a rear end plate 112, an upper end plate 113, and a lower end plate 114, forming a semi-closed or fully closed cavity structure.

[0065] Fixed handle and electronic compartment: The upper inner side of the fixed handle 12 is connected to the lower end plate 114 of the housing assembly 11. In order to make full use of space and protect the circuit, a hollow end cap 121 is formed on the upper outer side of the fixed handle 12. The data processing module 53 (such as a PCB board) mentioned above is housed inside the end cap 121. The end cap 121 is connected to the rear end plate 112 of the housing assembly 11, which makes the overall structure compact, the center of gravity reasonable, and facilitates handheld operation.

[0066] 2. Transmission components

[0067] In order to realize the function of "pressing the handle to drive the mechanical finger", a transmission component 4 is set inside the housing assembly 11.

[0068] The transmission assembly 4 includes a guide rod 41, a movable plate 42, and a spring 43. The guide rod 41 is fixedly connected between the front end plate 111 and the rear end plate 112 of the housing assembly 11, serving as a guide rail. The movable plate 42 is slidably fitted onto the guide rod 41. The pressing handle 3 is connected to the movable plate 42 (e.g., via a lever or linkage mechanism). When a person operates the pressing handle 3, it pushes the movable plate 42 to move backward along the guide rod 41.

[0069] Spring 43 is sleeved on guide rod 41; in this embodiment, compression spring 43 will be used for illustration. This compression spring 43 is disposed between movable plate 42 and rear end plate 112 of housing assembly 11.

[0070] Technical effect: When the operator presses the handle 3, the moving plate 42 moves backward against the resistance of the spring 43, driving the mechanical finger 2 to close; when the operator releases the handle, the compression spring 43 uses its own elastic restoring force to push the moving plate 42 back to its original position, thereby driving the mechanical finger 2 to complete the opening action. This design ensures smooth operation and automatic reset function.

[0071] 3. Mechanical fingers and their linkage mechanisms

[0072] Continue to refer to Figure 3 Each of the front ends of the housing assembly 11 has a mechanical finger 2 on each side, forming a grasping shape. To simulate the dexterous movements of a human hand or industrial gripper, each mechanical finger 2 employs a multi-link mechanism, specifically including:

[0073] Input linkage 21: Located inside the housing assembly 11, one end is hinged to the movable plate 42. When the movable plate 42 moves linearly, the input linkage 21 is pushed / pulled.

[0074] Intermediate connecting rod 22: One end is hinged to the other end of input connecting rod 21. A rotating shaft 221 is fixedly mounted on the intermediate connecting rod 22. The two ends of the rotating shaft 221 are respectively rotatably installed in the upper pivot hole of the upper end plate 113 and the lower pivot hole of the lower end plate 114 of the housing assembly 11. This rotating shaft 221 is the main fulcrum for the rotation of the entire finger.

[0075] Output link 23: One end is hinged to the other end of the intermediate link 22.

[0076] Finger tip 24: Hinged to the other end of the output link 23, used for direct contact with objects.

[0077] Auxiliary link 25: Located inside the mechanical finger 2 relative to the intermediate link 22, one end is hinged to the finger tip 24, and the other end is hinged to the housing assembly 11. The function of the auxiliary link 25 is to constrain the posture of the finger tip 24, so that it maintains relative translation or specific angular changes during the grasping process, and prevents slippage.

[0078] Flexible gripping (underactuated) design: A torsion spring 26 is provided at the hinge between the intermediate link 22 and the output link 23. The torsion spring 26 is configured to keep the mechanical finger 2 in an extended state under normal conditions.

[0079] Technical principle and effect: When the fingertip 24 contacts an object, if the resistance is too great, the output link 23 can rotate around the hinge point with the intermediate link 22 (compressing the torsion spring 26), thereby changing the shape of the finger to cover the object. This under-actuated design allows the acquisition device to adapt to objects of different shapes, greatly increasing the success rate of grasping and the diversity of data.

[0080] 4. Vision module installation

[0081] Continue to refer to Figure 1 The frame 1 is also equipped with a support plate 13, one end of which is fixed to the housing assembly 11 (usually the top), and the other end extends forward to mount the image acquisition module 51. This cantilever beam mounting method ensures that the camera can capture the area from above or at an angle without being obstructed by the operator's hands.

[0082] Example 3: Electronic and Electrical Systems and Data Processing Logic

[0083] This embodiment describes in detail the sensing and data processing flow of the device.

[0084] 1. Hardware implementation of motion detection

[0085] Combination Figure 2 and Figure 4To accurately measure the opening and closing angle of the mechanical finger 2, the motion detection module 52 employs a non-contact encoder solution. The motion detection module 52 includes a code disk 521 and a reading head 522. The code disk 521 is fixedly connected to one end (e.g., the upper end) of the aforementioned mechanical finger 2 rotating shaft 221 extending out of the housing assembly 11. The reading head 522 is fixedly disposed relative to the housing assembly 11 (e.g., fixed to the surface of the upper end plate 113) and is coaxially opposite to the code disk 521.

[0086] Technical effect: When the fingers open and close, the rotating shaft 221 drives the code disk 521 to rotate, and the reading head 522 reads the magnetic or optical changes of the code disk 521 and outputs a pulse signal or an absolute position signal. Directly measuring the rotation angle of the rotating shaft 221 at the base of the finger avoids the gap error caused by measuring the transmission chain (such as the handle), greatly improving the accuracy of motion parameters.

[0087] 2. Data Processing and Alignment Logic

[0088] Data processing module 53 (running on an embedded processor) executes the following key algorithms:

[0089] Timestamp Alignment: The module adds high-precision timestamps to each frame of image data and each moment's motion parameters. Since the camera sampling rate (e.g., 30fps) differs from the sensor sampling rate (e.g., 100Hz), the module uses interpolation or nearest neighbor matching based on the timestamps to align each frame with the corresponding motion parameters at that moment. This solves the time synchronization problem for multi-sensor data, ensuring that what the model "sees" and what it "acts" are in the same state at the same moment during model training.

[0090] Physical quantity mapping: Before alignment, the module uses pre-stored mapping relationships (calibration tables or formulas) to convert the raw detection signals (such as encoder pulse counts) output by the motion detection module 52 into intuitive physical distance values ​​(distance between two fingertips, in mm) or angle values ​​(in degrees). Effect: This transforms abstract sensor data into physically meaningful general data, allowing the collected data to be directly applied to real robot grippers with different drive methods but similar geometric dimensions.

[0091] 3. Data storage optimization

[0092] To reduce the storage of invalid data, the data processing module 53 features an intelligent triggering function. It monitors the rate of change of motion parameters. Only when a displacement change is detected in the mechanical finger 2 (i.e., the operator is performing a grasping or releasing action), or when the displacement change exceeds a set threshold, does the module begin storing or labeling the image and motion data for that period as valid training data. When the finger is stationary (e.g., in standby mode), recording stops or only low-frequency sampling is performed. This significantly saves storage space and reduces the data cleaning workload during subsequent model training, thereby increasing the "quality" of the data.

[0093] In summary, the specific embodiments of this invention achieve low-cost acquisition of high-quality robotic grasping data by constructing a handheld device that includes a biomimetic linkage mechanism, a precision sensing system, and intelligent data processing logic. Its mechanical structure not only ensures a comfortable operating feel and reset function but also adapts to various objects through underactuated design; its electronic and data processing systems ensure accurate synchronization and effective storage of multimodal data.

[0094] It should be understood that although this specification is described according to various embodiments, not every embodiment or implementation method contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

[0095] The above descriptions are merely illustrative embodiments of this application and are not intended to limit the scope of the embodiments of this application. Any equivalent changes, modifications, and combinations made by those skilled in the art without departing from the concept and principles of the embodiments of this application should fall within the protection scope of the embodiments of this application.

Claims

1. A training data acquisition device for robot gripper control, characterized in that, include: The main body of the acquisition device includes a frame (1), a pressing handle (3) connected to the frame (1), and a plurality of mechanical fingers (2) driven to open and close by the pressing handle (3); An image acquisition module (51) is mounted on the frame (1) and acquires image data of the grasping areas of multiple mechanical fingers (2) during the grasping action of the mechanical fingers (2). A motion detection module (52) is mounted on the frame (1) and detects the motion parameters of the mechanical finger (2) in real time during the grasping action of the mechanical finger (2). A data processing module (53) is mounted on the rack (1) and receives and processes the image data and the motion parameters to generate training data.

2. The training data acquisition device according to claim 1, characterized in that, The data processing module (53) is configured to add timestamps to the acquired image data and detected motion parameters respectively, and to associate and align the image data of each frame with the corresponding motion parameters based on the timestamps.

3. The training data acquisition device according to claim 2, characterized in that, The data processing module (53) is configured to, before associating and aligning the image data of each frame with the corresponding motion parameters, also use a pre-stored mapping relationship to convert the original detection signal output by the motion detection module (52) into a physical distance value or angle value that reflects the opening amplitude of the mechanical finger (2).

4. The training data acquisition device according to claim 2, characterized in that, The data processing module (53) is configured to store or process image data and motion parameters during the period when the mechanical finger (2) undergoes displacement changes, based on the monitored changes in the motion parameters, as effective training data.

5. The training data acquisition device according to claim 1, characterized in that, The frame (1) includes a housing assembly (11) and a fixed handle (12) connected to the housing assembly (11); a plurality of mechanical fingers (2) are hinged to the housing assembly (11), and a transmission assembly (4) is provided between the plurality of mechanical fingers (2) and the pressing handle (3).

6. The training data acquisition device according to claim 5, characterized in that, The housing assembly (11) includes a front end plate (111) and a rear end plate (112); the transmission assembly (4) includes: A guide rod (41) is disposed within the housing assembly (11) and connected between the front end plate (111) and the rear end plate (112); A movable plate (42) is slidably disposed on the guide rod (41), and the movable plate (42) is also hinged to the input ends of the plurality of mechanical fingers (2); wherein, the movable plate (42) is also connected to the pressing handle (3), and is configured such that when the pressing handle (3) is pressed, the movable plate (42) drives the plurality of mechanical fingers (2) to complete the closing action; A spring (43) is sleeved on the guide rod (41) and is configured to enable multiple mechanical fingers (2) to open under its own restoring force.

7. The training data acquisition device according to claim 6, characterized in that, The upper end plate (113) of the housing assembly (11) is provided with an upper pivot hole, and the lower end plate (114) is provided with a lower pivot hole; the mechanical finger (2) includes: The input link (21) is hinged at one end to the movable plate (42); The intermediate connecting rod (22) has one end hinged to the other end of the input connecting rod (21), and a rotating shaft (221) is fixedly installed on it. The two ends of the rotating shaft (221) are respectively rotatably installed in the upper pivot hole and the lower pivot hole. The output link (23) is hinged at one end to the other end of the intermediate link (22); The finger tip (24) is hinged to the other end of the output link (23); An auxiliary link (25) is located inside the mechanical finger (2) relative to the intermediate link (22), with one end hinged to the finger tip (24) and the other end hinged to the housing assembly (11).

8. The training data acquisition device according to claim 7, characterized in that, The mechanical finger (2) also includes: A torsion spring (26) is provided at the hinge between the intermediate link (22) and the output link (23) and is configured to keep the mechanical finger (2) in an extended state under its own restoring force.

9. The training data acquisition device according to claim 7, characterized in that, A mechanical finger (2) is provided on each of the two front ends of the housing assembly (11), and the input link (21) is located inside the housing assembly (11). One end of the intermediate link (22) is located inside the housing assembly (11), and the other end is located outside the housing assembly (11).

10. The training data acquisition device according to claim 7, characterized in that, The motion detection module (52) includes: The code disk (521) is connected to one end of the shaft (221) that extends out of the housing assembly (11); The reading head (522) is coaxially disposed opposite to the code disk (521) and fixedly disposed relative to the housing assembly (11).

Images