Method and device for picking up tennis balls by using self-driving automobile

Abstract

The invention discloses a method and device for picking up tennis balls through an automatic driving automobile, and relates to the technical field of intelligent robots and motion auxiliary equipment, and the device comprises a visual perception module, a calculation control module, a vehicle driving module, a tennis ball picking module and a classified storage module which are integrated on a chassis support. The device constructs a homography matrix by combining a visual module with an inverse perspective mapping algorithm to position three-dimensional coordinates of a tennis ball, and a calculation module executes greedy heuristic path planning and pure tracking control strategies to drive a vehicle to accurately approach a target. The picking module rolls the tennis balls in through an elastic roller suspended in a floating mode by means of mechanical deformation, the classification module controls a flow direction switching baffle to act on the basis of a wear degree score generated through deep learning, and the tennis balls are automatically distributed to a qualified or abandoned product bin. The problems that existing equipment is low in picking efficiency and lacks a sorting function are effectively solved, and automation and intellectualization of the whole operation process are achieved.

Description

Technical Field

[0001] This application relates to the field of intelligent robots and sports assistive devices, and in particular to a method and apparatus for picking up tennis balls using an autonomous vehicle. Background Technology

[0002] Tennis training typically involves numerous multi-ball drills, resulting in a large number of tennis balls scattered on the court after training. Traditional tennis ball recovery relies mainly on manual picking up or the use of manual trolley-style ball retrievers, which is labor-intensive and takes up valuable training time.

[0003] While some automated ball-retrieving robots have emerged in the market, their technological solutions still have limitations. Most existing automated ball-retrieving devices employ motion logic based on random collisions or fixed trajectories, lacking global perception and path planning capabilities for the playing surface. This leads to repeated coverage or missed areas during operation, resulting in low efficiency and poor energy utilization. Regarding the retrieval mechanism, some devices employ rigid gripping or sweeping structures that require a high degree of ground flatness and are prone to causing additional mechanical wear on the tennis ball surface.

[0004] More importantly, existing tennis ball recycling systems typically only have a single collection function and cannot distinguish the quality of the recycled balls. In actual training, if severely worn or depressurized balls are stored together with normal balls, it will affect the user experience in subsequent training sessions, forcing coaches or athletes to manually sort them again after recycling. This lack of intelligent sorting and storage capabilities limits the improvement of the automation level of tennis auxiliary equipment. Summary of the Invention

[0005] The purpose of this invention is to provide a method and apparatus for picking up tennis balls using an autonomous vehicle, which at least solves the technical problems of high labor costs, low efficiency, and lack of tennis ball quality screening function in existing tennis ball picking operations.

[0006] This invention provides the following solution:

[0007] The first aspect of this invention provides a device for retrieving tennis balls using an autonomous vehicle. This device integrates various functional modules onto an autonomous driving chassis support, and achieves connection and communication between the modules through electrical wiring and a data bus. Specifically, the device includes a vision perception module, a computing control module, a vehicle drive module, a tennis ball retrieval module, a tennis ball sorting and storage module, and a power management module.

[0008] In terms of hardware structure, the vision perception module is located on top of the chassis support. It is responsible for acquiring environmental images and converting two-dimensional pixel coordinates into three-dimensional physical coordinates in the vehicle coordinate system. A global shutter CMOS industrial camera assembly can be used to meet the image acquisition needs during movement. The computing and control module, as the core processing unit, communicates with the vision perception module, executes path planning algorithms based on the received coordinates and feature information, and generates corresponding control commands. The vehicle drive module responds to these control commands, feeding back the vehicle status through the integrated odometer sensor assembly and inertial measurement unit, and driving the vehicle to perform longitudinal movement and lateral steering.

[0009] Regarding the mechanical actuation section, the tennis ball pickup module is located at the front of the chassis support and uses mechanical deformation to draw the tennis ball into the device. The specific structure includes a pickup roller assembly mounted via a floating suspension bracket, which provides the pickup roller assembly with vertical displacement capability to adapt to ground undulations. The pickup roller assembly is driven to rotate by a drive motor assembly, and its main structure includes a central rotating shaft, support plates at both ends, and multiple elastic rods arranged at equal intervals around the circumference. These elastic rods form a cage-like structure, and the chordal distance between adjacent elastic rods is designed to be smaller than the standard diameter of the tennis ball to be picked up, thereby using the elastic deformation of the rods to clamp and draw in the tennis ball during rotation. A guide plate assembly extends to the lower tangent position of the roller to assist the tennis ball in entering the internal channel.

[0010] In the subsequent processing stage, the tennis ball sorting and storage module is located behind the picking module. It includes a gravity-fed chute-type transition conveyor, a sorting mechanism, and separate storage bins for qualified and discarded tennis balls of different qualities. The sorting mechanism is located at the end of the transition conveyor and uses an electromechanical actuator to drive a flow-direction switching baffle to rotate. Based on the sorting instructions issued by the calculation and control module, the flow-direction switching baffle switches between a first position leading to the qualified storage bin and a second position leading to the discarded storage bin, thereby achieving the physical separation of the tennis balls. The power management module provides the necessary independent operating voltage to each of the above modules through multiple DC-DC voltage conversion circuits.

[0011] A second aspect of this invention provides a method for picking up tennis balls using an autonomous vehicle. This method, applied to the aforementioned device, achieves fully automated operation through steps such as system initialization, global perception and planning, motion control and execution, mechanical picking, classification and storage, and cycle determination.

[0012] In the global perception and planning phase, the method employs an inverse perspective mapping algorithm to process visual data. This algorithm constructs a homography matrix using the camera's intrinsic parameter matrix, the camera's extrinsic parameter matrix relative to the vehicle coordinate system, and the assumption of flat ground. By solving the inverse of the homography matrix, the pixel coordinates of the image plane are mapped to three-dimensional physical coordinates in the vehicle coordinate system, and further converted to absolute coordinates in the global navigation coordinate system, forming a target point set. The path planning algorithm first calculates the geometric center of the target point set to delineate the task weight region. Then, it uses a greedy heuristic search strategy to initialize the unvisited point set, and through traversal calculation and iterative selection of the candidate point closest to the current node, it generates a path sequence containing the sequential waypoint information of all tennis target points.

[0013] In the motion control and execution phase, the method employs a pure tracking control strategy. The system selects a target forward-looking point on the generated path sequence at a preset forward-looking distance from the vehicle's current position, calculates the desired steering angle based on the Ackerman steering geometry model, and sends it to the vehicle drive module. Simultaneously, it calculates the longitudinal speed control quantity based on the remaining distance between the vehicle and the target tennis ball, enabling the vehicle to perform linear deceleration as it approaches the target, ensuring precise stopping or low-speed capture.

[0014] During the classification, storage, and recycling stages, the system incorporates deep learning technology for quality assessment. The computational control module reads the tennis ball wear score generated by a deep convolutional neural network based on the surface texture features of the tennis ball. The system compares this wear score with a preset discard threshold: if the score is greater than or equal to the threshold, the flow is directed to the qualified product storage bin via a switching baffle; if the score is less than the threshold, it is directed to the discard product storage bin.

[0015] In addition, the method includes an anomaly monitoring mechanism. The system monitors the operating current of the motor in the tennis ball retrieval module or tennis ball sorting and storage module in real time. When the monitored real-time operating current exceeds the preset stall threshold and the duration exceeds the set limit, the system determines that mechanical stall has occurred and automatically executes a reverse self-rescue action sequence, including cutting off the forward power supply, sending a reverse rotation command and continuing for a preset time, and then resuming forward rotation to release the mechanical jam.

[0016] The above solution achieves the following beneficial technical effects:

[0017] This application utilizes a visual perception module combined with an inverse perspective mapping algorithm and a greedy heuristic path planning strategy to accurately acquire the three-dimensional coordinates of tennis balls and generate a globally optimal operational path. Combined with vehicle drive control based on a pure tracking strategy, this achieves full automation and intelligent path management for tennis ball retrieval. This effectively solves the problems of low efficiency in traditional manual retrieval and insufficient coverage due to random wandering of existing equipment, improving field cleaning efficiency and reducing labor costs.

[0018] This application employs a tennis ball retrieval module with a floating suspension bracket and a cage-like structure of elastic rods, utilizing the principle of mechanical deformation to encircle the tennis ball. This design allows the retrieval mechanism to adapt to ground undulations and avoids rigid compression damage to the tennis ball through elastic buffering at the moment of contact, while also preventing damage to mechanical components from hard impacts. Compared to complex robotic arm gripping solutions, this structure significantly simplifies mechanical complexity while ensuring a high retrieval success rate, and improves the durability and maneuverability of the device.

[0019] This application integrates a deep learning-based tennis ball classification and storage module. It utilizes a convolutional neural network to score the surface wear of the tennis balls and uses an electromechanical actuator to control a flow-direction switching baffle, diverting the balls in real time to either qualified or discarded storage bins. This design endows the picking device with online quality inspection capabilities, enabling the immediate removal of unusable balls. It solves the problem of traditional devices being unable to distinguish tennis ball quality, necessitating subsequent manual secondary sorting, and further improves the convenience of tennis ball recycling. Attached Figure Description

[0020] Figure 1 This is a perspective view of the present invention;

[0021] Figure 2 This is a perspective view of the rear side of the present invention;

[0022] Figure 3 This is a system module connection diagram of the present invention.

[0023] The components include: 100. Autonomous tennis ball picking device; 10. Visual perception module; 11. Industrial camera assembly; 20. Computational control module; 30. Vehicle drive module; 31. Odometer sensor assembly; 32. Inertial measurement unit; 40. Tennis ball picking module; 41. Picking roller assembly; 411. Central rotating shaft; 412. Support end plate; 413. Elastic rod; 42. Drive motor assembly; 43. Guide shovel assembly; 44. Floating suspension bracket; 50. Tennis ball sorting and storage module; 51. Transition conveying channel; 52. Sorting execution mechanism; 521. Electromechanical actuator; 522. Rotary drive shaft; 523. Flow direction switching baffle; 53. Qualified product storage bin; 54. Waste product storage bin; 60. Power management module; 61. High energy density lithium-ion battery pack; 62. Battery management system; 63. Multi-channel DC-DC voltage conversion circuit. Detailed Implementation

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

[0025] See attached document Figures 1-3 This invention provides an autonomous tennis ball retrieval device 100, which includes: a vision perception module 10, a computing control module 20, a vehicle drive module 30, a tennis ball retrieval module 40, a tennis ball classification and storage module 50, and a power management module 60. All of these modules are integrated on the chassis support of the autonomous tennis ball retrieval device 100 and are connected and communicate with each other via electrical wiring and a data bus.

[0026] A visual perception module 10 is mounted on top of the autonomous tennis ball retrieval device 100 to acquire real-time image data of the tennis court environment. The visual perception module 10 includes at least one high-resolution industrial camera whose optical axis forms a preset angle with the horizontal plane to cover the ground area in the vehicle's direction of travel. The visual perception module 10 is configured to convert the acquired two-dimensional pixel coordinates into three-dimensional physical coordinates in the vehicle's coordinate system.

[0027] Based on the pinhole camera imaging model, the visual perception module 10 performs coordinate transformation. Assume the coordinates in the image pixel coordinate system are... The physical coordinates in the vehicle coordinate system are The transformation relationship satisfies the following equation:

[0028] ;

[0029] in: This represents the depth information of the target object in the camera coordinate system; The intrinsic parameter matrix of the camera is defined as follows:

[0030] ;

[0031] In the aforementioned intrinsic parameter matrix, and These represent the normalized focal length of the camera in the horizontal and vertical directions, respectively. and The pixel coordinates of the principal point of the image (i.e., the intersection of the optical axis and the image plane). This represents the extrinsic parameter matrix from the vehicle coordinate system to the camera coordinate system, including the rotation matrix. Translation vector Defined as:

[0032] ;

[0033] Based on the above calculations, the visual perception module 10 outputs the relative position coordinates of the tennis ball to be picked up on the vehicle's driving plane. To the calculation control module 20.

[0034] The computational control module 20 is communicatively connected to the visual perception module 10, the vehicle drive module 30, the tennis ball retrieval module 40, and the tennis ball classification and storage module 50. The computational control module 20 includes a central processing unit, memory, and input / output interfaces. The computational control module 20 is configured to receive the position coordinates and tennis ball feature information output by the visual perception module 10, execute a path planning algorithm to generate a driving path, and send motion control commands to the vehicle drive module 30.

[0035] The vehicle drive module 30 includes a drive motor, a steering servo, and a wheel system, employing an Ackermann steering geometry. The vehicle drive module 30 executes longitudinal movement and lateral steering according to instructions from the computational control module 20. The kinematic state of the vehicle drive module 30 satisfies the following set of differential equations:

[0036] ;

[0037] in: This indicates the real-time position coordinates of the center of the rear axle of the autonomous tennis ball-picking device 100. Indicates the longitudinal speed of the vehicle; Indicates the vehicle's current heading angle; Indicates the wheelbase between the front and rear axles of a vehicle; This indicates the steering angle of the front wheels. The calculation and control module 20 adjusts... and Two control variables drive the vehicle to move along the planned path to the target tennis ball position.

[0038] The tennis ball retrieval module 40 is located at the front of the chassis of the autonomous tennis ball retrieval device 100 and includes a roller mechanism composed of multiple parallel, elastic strip-shaped structures. The tennis ball retrieval module 40 is configured to use mechanical deformation to roll up tennis balls from the ground into the device. The static gap between the strip-shaped structures of the roller mechanism... Smaller than the diameter of a standard tennis ball When the tennis ball is subjected to the squeezing force of the roller mechanism When in action, it overcomes the elastic resistance of the strip structure. It enters the interior of the drum. This elastic resistance is defined as:

[0039] ;

[0040] in: Indicates the stiffness coefficient of the strip structure; This represents the displacement caused by the deformation of the strip structure.

[0041] The tennis ball sorting and storage module 50 is located at the rear end of the tennis ball picking module 40, and includes a flow guide channel, a sorting execution mechanism, and a compartmentalized storage box. The compartmentalized storage box is divided into a qualified product compartment and a waste product compartment. The tennis ball sorting and storage module 50 receives sorting instructions from the calculation and control module 20. The calculation and control module 20 determines the wear level of the tennis ball based on the texture features extracted by the visual perception module 10. Greater than or equal to the preset threshold When the sorting mechanism operates, it guides the tennis balls into the qualified goods warehouse; when the wear level of the tennis balls is determined... Less than the preset threshold At that time, the sorting mechanism operates, guiding the tennis balls into the waste bin.

[0042] The power management module 60 includes a rechargeable battery pack and a power management circuit, which are electrically connected to the vision perception module 10, the computing control module 20, the vehicle drive module 30, the tennis ball retrieval module 40, and the tennis ball sorting and storage module 50, respectively, and provide operating voltage for the above modules.

[0043] The tennis ball retrieval module 40 provided by the present invention specifically includes: a retrieval roller assembly 41, a drive motor assembly 42, a guide shovel assembly 43, and a floating suspension bracket 44.

[0044] The floating suspension bracket 44 is fixed to the front end of the chassis of the autonomous tennis ball pickup device 100 via a mechanical connector, and is configured to allow the pickup roller assembly 41 to have a preset range of displacement freedom in the vertical direction to adapt to minor undulations in the ground. The pickup roller assembly 41 is rotatably mounted on the bearing seat of the floating suspension bracket 44, and its axis of rotation is parallel to the lateral axis of the vehicle.

[0045] The pickup roller assembly 41 includes a central rotating shaft 411, support end plates 412 disposed at both ends of the central rotating shaft 411, and a plurality of circumferentially distributed elastic rods 413. The two ends of the plurality of elastic rods 413 are respectively fixedly connected to the support end plates 412 on both sides. The elastic rods 413 are parallel to each other and arranged at equal intervals along the circumferential direction, forming a cage-like cylindrical structure. The elastic rods 413 employ a preset stiffness coefficient. It is made of an elastic material selected from spring steel or high-strength modified nylon.

[0046] To effectively grasp the tennis ball and prevent it from slipping, the structural dimensions of the pickup roller assembly 41 meet specific geometric constraints. The chord length distance between two adjacent elastic members 413 is defined as... The standard diameter of the tennis ball to be picked up is Then the relationship between the two is configured to strictly satisfy:

[0047] ;

[0048] This geometric constraint ensures that the tennis ball cannot pass through the gap between adjacent elastic members 413 without external force.

[0049] The drive motor assembly 42 is mechanically coupled to the central rotating shaft 411 and configured to drive the pickup roller assembly 41 to rotate in the same or opposite direction as the vehicle's travel direction. When the pickup roller assembly 41 rotates to contact the tennis ball on the ground, the tennis ball is subjected to friction from the ground and compressive force from the elastic rod 413. The combined effect of.

[0050] During the pickup process, the elastic rod 413 undergoes radial elastic deformation, causing the distance between adjacent elastic rods 413 to increase instantaneously. The critical mechanical condition for the tennis ball to enter the pickup roller assembly 41 is the compressive force. Overcoming the elastic resistance of elastic member 413 This elastic resistance follows Hooke's law and the formula for the bending deformation of a beam, and is defined as follows:

[0051] ;

[0052] in: This represents the amount of deformation displacement of the elastic rod 413 in the radial direction. When the tennis ball completely crosses the center line of the elastic rod 413 and enters the inner cavity of the pickup roller assembly 41, the elastic rod 413 returns to its initial state under the action of the elastic restoring force, thereby capturing and confining the tennis ball within the inner cavity of the pickup roller assembly 41.

[0053] The guide shovel assembly 43 is located behind the pickup roller assembly 41 and has a curved guide structure. The front end of the guide shovel assembly 43 extends to the lower tangent position of the pickup roller assembly 41 and maintains a preset safety gap with the rotation trajectory of the elastic rod 413. When the tennis ball captured in the inner cavity rotates to a high position with the pickup roller assembly 41, the tennis ball detaches from the elastic rod 413 under the action of gravity and centrifugal force, falls into the guide groove of the guide shovel assembly 43, and slides along the guide groove into the tennis ball sorting and storage module 50 at the rear end.

[0054] Optionally, a negative pressure auxiliary suction port is also provided above the pickup roller assembly 41. This negative pressure auxiliary suction port is connected to the vehicle-mounted air pump and is configured to provide an upward pneumatic suction force at the moment the elastic rod 413 contacts the tennis ball, thereby reducing the required squeezing force. This improves the success rate of pickups on surfaces with low friction coefficients.

[0055] The tennis ball sorting and storage module 50 provided by the present invention is mechanically connected to the rear end of the tennis ball picking module 40, and specifically includes: a transition conveying channel 51, a sorting execution mechanism 52, a qualified product storage bin 53, and a waste product storage bin 54.

[0056] The transition conveyor channel 51 is located at the entrance of the tennis ball sorting and storage module 50, and its front end connects to the output end of the guide shovel assembly 43 in the tennis ball picking module 40. The transition conveyor channel 51 adopts a preset tilt angle. The gravity chute structure, the tilt angle The angle of static friction between the tennis ball and the chute material is configured to be greater than the angle between the tennis ball and the chute material to ensure that the tennis ball can automatically roll into the working area of ​​the sorting actuator 52 under the action of gravity. The cross-sectional width of the transition conveyor channel 51 is slightly larger than the diameter of the tennis ball. This forms a single-line queuing transport flow.

[0057] The sorting actuator 52 is located below the end of the transition conveyor channel 51 and includes an electromechanical actuator 521, a rotary drive shaft 522, and a flow direction switching baffle 523. The electromechanical actuator 521 is fixedly mounted on the side wall of the frame and connected to the rotary drive shaft 522 via a coupling. The flow direction switching baffle 523 is fixed on the rotary drive shaft 522 and located at the fork point of the tennis ball rolling path.

[0058] The electromechanical actuator 521 is electrically connected to the computing control module 20 and is configured to drive the flow direction switching baffle 523 to a first position or a second position in response to control commands. When the flow direction switching baffle 523 is in the first position, the baffle plane blocks the path to the waste storage bin 54 and simultaneously forms a guide ramp leading to the qualified product storage bin 53, allowing the tennis ball to slide into the qualified product storage bin 53 along the guide ramp. When the flow direction switching baffle 523 is in the second position, the baffle plane blocks the path to the qualified product storage bin 53 and simultaneously opens the entrance to the waste storage bin 54, or forms a guide ramp leading to the waste storage bin 54, allowing the tennis ball to slide into the waste storage bin 54.

[0059] The qualified product storage bin 53 and the waste product storage bin 54 are arranged side by side below or behind the sorting execution mechanism 52. The qualified product storage bin 53 has a first volume. Used to store tennis balls deemed usable; waste storage bin 54 has a second volume. It is used to store tennis balls that are deemed to be severely worn or damaged.

[0060] Non-contact liquid level sensors 55 (such as infrared beam sensors or ultrasonic ranging sensors) are installed on the upper inner walls of the qualified product storage bin 53 and the waste product storage bin 54. These non-contact liquid level sensors 55 are connected to the computing and control module 20 and configured to monitor the height of the tennis balls stacked within the bins in real time. When the monitored stacking height reaches a preset full-bin threshold... At this time, the non-contact liquid level sensor 55 sends a full-tank signal to the computing control module 20, and the computing control module 20 controls the vehicle to stop the picking operation or issues a warning signal accordingly.

[0061] To facilitate unloading, the qualified product storage bin 53 and the waste product storage bin 54 are equipped with unloading gates 56 controlled by electromagnetic locks at the bottom. When the autonomous tennis ball picking device 100 travels to the designated unloading area, the computing control module 20 issues a command to unlock the unloading gates 56, and the tennis balls in the bins are discharged by gravity.

[0062] The sorting action is triggered based on the tennis ball feature data generated by the aforementioned visual perception module 10. The system maintains a first-in, first-out queue of tennis balls to be sorted. When a tennis ball passes the photoelectric counting sensor at the end of the transition conveyor channel 51, the calculation and control module 20 reads the wear score of the current tennis ball from the head of the queue. and compare it with a preset threshold. The comparison is then used to generate a pulse width modulation signal or a level signal to drive the electromechanical actuator 521, thereby achieving specific physical sorting.

[0063] In this embodiment of the invention, in order to achieve high-precision environmental perception and motion control, the sensor system and hardware circuit adopt a layered architecture layout.

[0064] The visual perception module 10 specifically includes an industrial camera assembly 11 mounted on the front end of the top bracket of the autonomous tennis ball retrieving device 100. The industrial camera assembly 11 employs a global shutter CMOS image sensor configured to eliminate the rolling shutter effect generated during vehicle movement. A fixed pitch angle is set between the optical axis of the industrial camera assembly 11 and the horizontal plane. The pitch angle The value range is configured such that the camera's vertical field of view can cover a ground plane area of ​​0.5 meters to 10 meters in front of the vehicle. The industrial camera assembly 11 is connected to the computing control module 20 via a high-speed data interface (such as USB 3.0 or GigE interface) for transmitting raw video stream data.

[0065] To assist in precise vehicle positioning and support the solution of the aforementioned Ackermann steering kinematics model, the vehicle drive module 30 also integrates an odometer sensor assembly 31 and an inertial measurement unit 32. The odometer sensor assembly 31 includes an incremental photoelectric encoder mounted on the output shaft of the drive motor. The odometer sensor assembly 31 is configured to acquire the wheel rotation speed and cumulative rotation angle in real time, and feed the pulse signals back to the underlying controller of the vehicle drive module 30 or directly transmit them to the calculation control module 20 for calculating the vehicle's longitudinal speed. The inertial measurement unit 32 is rigidly fixed to the geometric center of the vehicle chassis and includes a three-axis accelerometer and a three-axis gyroscope. The inertial measurement unit 32 communicates with the computational control module 20 via an I2C or SPI bus and is configured to output the vehicle's angular velocity and linear acceleration data. The computational control module 20 uses this data to determine the vehicle's heading angle. Perform integral updates and drift corrections.

[0066] The computing control module 20, as the core of the hardware system, includes an embedded artificial intelligence computing platform. The hardware interfaces of the computing control module 20 include: a high-speed video interface for receiving data from the industrial camera component 11; a CAN bus interface or RS-485 serial communication interface for exchanging instructions with the underlying driver of the vehicle drive module 30; a GPIO (General Purpose Input / Output) interface for receiving signals from the non-contact liquid level sensor 55 and photoelectric counting sensor in the tennis ball sorting and storage module 50; and a PWM (Pulse Width Modulation) output interface for outputting control signals to drive the electromechanical actuator 521 in the tennis ball sorting and storage module 50.

[0067] The power management module 60 includes a high-energy-density lithium-ion battery pack 61, a battery management system 62, and a multi-channel DC-DC voltage conversion circuit 63. The lithium-ion battery pack 61 is positioned in the middle of the vehicle chassis to lower the vehicle's center of gravity. The battery management system 62 is connected in series in the battery circuit and is configured to monitor cell voltage, current, and temperature in real time, and to perform overcharge, over-discharge, and short-circuit protection. The multi-channel DC-DC voltage conversion circuit 63 is connected to the output terminal of the lithium-ion battery pack 61 and is configured to convert the battery voltage into stable DC voltages of different levels to supply different loads: the first output voltage is connected to the drive motor of the vehicle drive module 30 and the drive motor assembly 42 of the tennis ball retrieval module 40, providing power; the second output voltage is connected to the computing control module 20, providing computing power; and the third output voltage is connected to various sensors and logic control circuits, providing signal power.

[0068] The wiring of the hardware circuit follows the electromagnetic compatibility design principle. The power supply line and the low-voltage signal line are laid in separate slots or transmitted with twisted-pair shielded wires to suppress the electromagnetic interference of high-frequency noise generated by the motor operation on visual image transmission and control command issuance.

[0069] In this embodiment of the invention, the visual perception and target recognition module is integrated and runs on the embedded artificial intelligence computing platform of the computing control module 20. It is configured to process the raw video stream from the visual perception module 10 and output a set of tennis targets containing position information and attribute information.

[0070] The processing flow of the visual perception and target recognition module specifically includes an image preprocessing unit, a deep convolutional neural network inference unit, and a feature parsing unit.

[0071] The image preprocessing unit receives real-time video frame images acquired by the industrial camera assembly 11. The video frame image Defined as a three-dimensional tensor ,in Indicates the image pixel height. Indicates the image pixel width. This represents the RGB color channels. The image preprocessing unit is configured to process the input image. Perform size scaling and pixel value normalization operations to convert it into a tensor format that meets the input requirements of deep convolutional neural networks.

[0072] The deep convolutional neural network inference unit adopts a single-stage object detection model architecture. This network architecture includes a backbone feature extraction network, a neck feature fusion network, and a detection head. The backbone feature extraction network is configured to extract semantic features of the image layer by layer; the neck feature fusion network is configured to fuse feature maps of different scales to enhance the detection capability of small objects such as tennis balls; the detection head is configured to perform dense predictions on the feature maps and output the position offset of candidate bounding boxes, class confidence, and attribute scores.

[0073] The feature parsing unit is configured to perform non-maximum suppression processing on the output of the deep convolutional neural network inference unit, removing redundant bounding boxes with overlap exceeding a preset threshold, and finally generating an effective set of tennis targets. For sets The first in For each detected tennis ball target, the feature parsing unit generates a feature description vector. The vector is defined as:

[0074] ;

[0075] in: Indicates the first The geometric position information of each tennis ball in the image pixel coordinate system, specifically including the coordinates of the center point of the bounding box. and the width of the bounding box and height ,Right now ; This represents the classification confidence label, with a value range of [value range missing]. The value 1 indicates that the target is identified as a tennis ball; Indicates the first The wear rating of a tennis ball is a continuous real number taking values ​​within the closed interval [0,1]. The generation is based on the analysis of texture features of the target region by the network model. In specific implementations, the detection head of the deep convolutional neural network includes a dedicated quality assessment branch, or the system includes a cascaded lightweight classification network. This quality assessment branch or classification network is configured to extract bounding boxes. The corresponding region of interest is analyzed, including the integrity of the surface fibers, color saturation, and stain distribution characteristics. The output is then mapped using a Sigmoid activation function. value.

[0076] The feature parsing unit is further configured to score wear. Compared with the preset waste threshold Comparison operations. When When the system marks the tennis ball as qualified, the corresponding control instruction is to store it in the qualified product storage bin 53; when When the tennis ball is discarded, the system marks it as discarded, and the corresponding control instruction indicates that it should be stored in the waste storage bin 54.

[0077] Ultimately, the visual perception and target recognition module will include pixel coordinates. and wear rating Feature description vector The data is transmitted to the subsequent path planning and control module for physical coordinate calculation and triggering of sorting actions.

[0078] In this embodiment of the invention, the calculation control module 20 executes a monocular vision-based inverse perspective mapping algorithm to map the pixel coordinates of the tennis ball image output by the visual perception and target recognition module to three-dimensional physical coordinates in the vehicle coordinate system, thereby determining the spatial position of the target tennis ball relative to the autonomous driving tennis ball picking device 100.

[0079] The spatial positioning process first defines three core coordinate systems: the image pixel coordinate system. The origin is located at the top left corner of the image, and the unit is pixels; camera coordinate system. The origin is located at the camera's optical center, and the Z-axis points forward along the optical axis; and the vehicle coordinate system... The origin is located at the projection point of the vehicle's rear axle center on the ground. The X-axis points directly in front of the vehicle, the Y-axis points to the left side of the vehicle, and the Z-axis is perpendicular to the ground and pointing upwards.

[0080] The calculation and control module 20 stores a pre-calibrated camera intrinsic parameter matrix. The extrinsic parameter matrix of the camera relative to the vehicle coordinate system Intrinsic parameter matrix This describes the projection relationship of the camera from the camera coordinate system to the image pixel coordinate system, specifically expressed as:

[0081] ;

[0082] in: and These represent the normalized focal length of the camera in the horizontal and vertical directions, respectively. and This represents the pixel coordinates of the principal point of the image (i.e., the intersection of the optical axis and the image plane). Extrinsic parameter matrix. This describes the rigid body transformation relationship from the vehicle coordinate system to the camera coordinate system, including the rotation matrix. Translation vector Specifically, it can be expressed as:

[0083] ;

[0084] For the i-th tennis ball detected in the image, its pixel center coordinates are Based on the pinhole camera model, pixel coordinates and physical coordinates in the vehicle coordinate system. Satisfy the following perspective projection equations:

[0085] ;

[0086] in: It is the depth of field of the target point in the camera coordinate system; It is the height of the target point in the vehicle coordinate system.

[0087] To calculate the three-dimensional physical coordinates from the two-dimensional pixel coordinates, the calculation control module 20 introduces a ground flatness assumption constraint. Since the tennis ball is located on the ground, its height coordinates in the vehicle coordinate system are constrained to... (Ignoring the slight height deviation caused by the tennis ball's radius, or adjusting the origin of the coordinate system to the plane at the tennis ball's radius height). Based on this constraint, the above equation simplifies to:

[0088] ;

[0089] in: and These are rotation matrices. The first and second column vectors. Let the homography matrix... The matrix It is a 3×3 invertible matrix.

[0090] The calculation control module 20 solves the matrix inverse matrix An inverse perspective transformation is performed to calculate the physical coordinates of the tennis ball in the vehicle coordinate system plane. The transformation calculation process is as follows:

[0091] ;

[0092] Final physical coordinates Obtained through homogeneous coordinate normalization:

[0093] ;

[0094] Through the above calculation steps, the calculation control module 20 maps the position of each visually recognized tennis ball pixel to a longitudinal distance relative to the center of the vehicle's rear axle. and lateral distance The calculation and control module 20 further calculates the relative coordinates based on the vehicle's current global pose information (obtained by the fusion calculation of the inertial measurement unit 32 and the odometer sensor assembly 31). Convert to absolute coordinates in the global navigation coordinate system And store the coordinates of all valid tennis balls into the target point set list. This is for use by the subsequent path planning module.

[0095] In this embodiment of the invention, the calculation control module 20 internally runs a path planning algorithm unit, configured to receive the target point set output by the aforementioned spatial positioning and coordinate transformation steps. ,in This indicates the total number of valid tennis balls identified within the current field of view or working area. Indicates the first The two-dimensional position coordinates of a tennis ball in the global navigation coordinate system .

[0096] To improve the traversal efficiency of the picking operation and reduce the energy consumption of the vehicle during idle driving, the calculation and control module 20 first performs clustering determination of the area to be picked. The system is configured to calculate the target point set. geometric center The current task weight area is then defined based on this center. Geometric center The coordinates are calculated using the arithmetic mean formula:

[0097] ;

[0098] The calculation control module 20 determines the geometric center. Current location of the vehicle The Euclidean distance between them. If this distance is less than a preset operating radius threshold, then the current point set is determined to be... To immediately pick up the target group; if the distance exceeds the threshold, the system will rescan or move the vehicle towards... Directional guidance until the required working distance is met.

[0099] After determining the target point set to be picked up Subsequently, the computational control module 20 executes a local path optimization algorithm. This process is modeled as a variant of the Traveling Salesman Problem with a fixed starting point, aiming to find a path that traverses all destination points. And the total driving route length The minimum access sequence. Define the vehicle's current position as the starting point. The objective function for path optimization is expressed as:

[0100] ;

[0101] in: Represents a set A permutation sequence representing the order in which the tennis balls were picked up; The Euclidean distance norm between two coordinate points; Represents the path sequence of the first... The tennis coordinates of the visit.

[0102] Given that solving the Traveling Salesman Problem completely has NP-hard computational complexity, in order to meet the real-time requirements of the autonomous driving system, the computational control module 20 in this embodiment adopts a greedy heuristic search strategy to generate a suboptimal path sequence. The specific execution steps are as follows:

[0103] Step 1: Initialize the set of unvisited points Set the vehicle's current location as the current node. Initialize the path sequence list ;

[0104] Step 2, traverse the set All remaining candidate points ,calculate To each distance ;

[0105] Step 3, select the distance Minimum candidate point As the next target point, remove it from the set. Remove and add to the path sequence list At the same time, update the current node. ;

[0106] Step four, repeat steps two and three until a set is formed. Empty.

[0107] The calculation and control module 20 ultimately outputs a planned path sequence containing sequential waypoint information. Furthermore, the computational control module 20 is also configured to introduce non-integrity constraint checks during path planning. This applies when connecting two adjacent target points. and At that time, the system determines the minimum turning radius of the vehicle. The feasibility of a straight-line connection is verified. If the relative angle change between two points exceeds the vehicle's maximum steering capability, the calculation and control module 20 inserts an auxiliary transition point between the two points to generate a smooth reference trajectory that conforms to the vehicle's kinematic characteristics. This reference trajectory is then transmitted to the vehicle motion control module to drive the underlying motors to perform specific displacement operations.

[0108] In this embodiment of the invention, the computational control module 20 internally runs a motion control algorithm unit. This unit is configured to use the reference path sequence output by the previous-level path planning module based on the Ackerman steering geometry model and a pure tracking path tracking algorithm. This is converted into low-level control commands that can be executed by the vehicle drive module 30.

[0109] The motion control algorithm unit first establishes the kinematic state equations of the autonomous tennis ball pickup device 100. Considering that the tire lateral slip characteristics are negligible under low-speed conditions (the speed is usually low when picking up tennis balls), the system uses a nonholonomic constrained single-vehicle model to describe the vehicle motion. The position of the vehicle's rear axle center in the global coordinate system is defined as... The vehicle's heading angle is (i.e., the angle between the vehicle's longitudinal axis and the global coordinate system's X-axis), the vehicle's wheelbase is The system's state update follows the system of differential equations:

[0110] ;

[0111] in: Indicates time Longitudinal driving speed control quantity; Indicates time The front wheel steering angle control quantity. The task of the calculation control module 20 is to solve for the optimal control quantity at each moment. This makes the vehicle's actual driving trajectory approximate the reference path sequence. .

[0112] To achieve accurate path tracking, the calculation control module 20 executes a pure tracking control strategy. The core of this strategy lies in calculating a forward look-ahead distance. And select a location on the reference path that is close to the current vehicle position. for The path points are used as the target's forward viewpoints. The calculation and control module 20 calculates the vehicle's current heading vector and the distance from the vehicle's rear axle center to the target's forward viewpoint in real time. The angle between the line vectors connecting the two lines is denoted as . Based on geometric relationships, the vehicle travels from its current position to the target's forward-looking point. Required arc trajectory curvature satisfy:

[0113] ;

[0114] Based on Ackermann steering geometry, the front wheel steering angle With trajectory curvature There is a mapping relationship Therefore, the calculation control module 20 calculates the desired steering angle required at the current moment using the following formula. :

[0115] ;

[0116] In the actual control cycle, the calculation control module 20 operates at a fixed control period. (For example, 20ms) Execute the following closed-loop control steps:

[0117] Step 1: Update the vehicle's current global pose state by fusing data from the odometer sensor assembly 31 and the inertial measurement unit 32. ;

[0118] Step two, in the reference path sequence The system searches for the nearest waypoint to the vehicle's current location and then searches forward along the path based on that waypoint to find the location with a straight-line distance closest to the preset forward look-ahead distance. The point is used as the target's forward viewpoint. ;

[0119] Step 3: Substitute the target's forward view coordinates into the above formula for the desired steering angle to calculate. And convert it into a corresponding pulse width modulation signal and send it to the steering servo;

[0120] Step 4: Based on the remaining distance between the current vehicle and the target tennis ball. Calculate longitudinal speed control quantity The system is configured with a speed planning curve, when... Maintain cruise speed when the speed exceeds the deceleration threshold. ;when When the speed is below the deceleration threshold, the speed is linearly reduced to ensure that the speed is suitable for the pickup device to operate when the tennis ball is reached.

[0121] In addition, to prevent excessive wear of the steering mechanism or exceeding mechanical limits, the calculation control module 20 calculates... Apply saturation constraints:

[0122] ;

[0123] in: This is the maximum steering angle allowed by the vehicle's mechanical structure. Finally, the calculation and control module 20 will... and The communication interface sends data to the underlying controller of the vehicle drive module 30, which drives the motor and servo motor to move, thereby enabling the autonomous tennis ball picking device 100 to accurately track the planned path and accurately approach the target tennis ball.

[0124] In this embodiment of the invention, the operation process of the autonomous tennis ball retrieval device 100 is configured as a closed-loop state machine control logic dominated by the computing control module 20. This process specifically includes a system initialization stage, a global perception and planning stage, a motion control and execution stage, and a classification storage and loop judgment stage.

[0125] Step S100: System Initialization and Self-Test. After the power management module 60 is powered on, the computing control module 20 first executes a hardware self-test program. The computing control module 20 sequentially sends handshake commands to the vision perception module 10, the vehicle drive module 30, the tennis ball retrieval module 40, and the tennis ball sorting and storage module 50. If it receives ready response signals from all sub-modules, and the battery level reported by the battery management system 62 is higher than the preset operating threshold (e.g., 20%), the computing control module 20 sets the system status to standby mode and waits for the user to issue a start retrieval command through an external interface (such as a remote control or APP).

[0126] Step S200: Global Perception and Target Locking In response to the start pickup command, the computational control module 20 activates the visual perception module 10. The industrial camera assembly 11 acquires environmental images at a preset frame rate (e.g., 30fps). The computational control module 20 calls a deep convolutional neural network model to process the image. The algorithm performs inference and outputs a set of tennis ball targets. Simultaneously, it calls a coordinate transformation algorithm to map the image coordinates to a set of target points in the global navigation coordinate system. If set If the field of view is empty (i.e. no tennis ball is detected in the current field of view), the calculation control module 20 controls the vehicle to rotate in place by a preset angle (e.g., 30 degrees) to switch the field of view, and repeats the above perception steps until a valid target is detected or the rotation reaches a cumulative 360 ​​degrees.

[0127] Step S300: Path planning and task scheduling when the target point set is... When not empty, the computation control module 20 executes the region clustering and traveling salesman problem path planning algorithm. The system first calculates the point set. geometric center And generate a path that traverses all target points. Optimal path sequence The path sequence is loaded into the register of the motion control unit as the current tracking reference trajectory.

[0128] Step S400: Autonomous cruise and approximation calculation control module 20 sequentially selects from the path sequence. Extract the current target waypoint The system enters the motion control loop, and calculates the desired front wheel steering angle in real time based on the pure tracking algorithm. and longitudinal velocity The vehicle drive module 30 responds to the aforementioned control command and drives the autonomous tennis ball pickup device 100 toward the target waypoint. Movement. During this process, the visual perception module 10 continuously tracks the target tennis ball and uses Kalman filtering to smooth and correct the target's position coordinates to compensate for observation errors caused by vehicle movement.

[0129] Step S500: Mechanical pickup action is executed when the calculation control module 20 determines the Euclidean distance between the vehicle and the current target tennis ball. The effective capture radius is less than 40 for the tennis ball pickup module. When the distance is 0.5 meters (for example), the system triggers a pickup sequence:

[0130] First, the calculation and control module 20 sends a start signal to the drive motor assembly 42 of the tennis ball pickup module 40, driving the pickup roller assembly 41 to a preset speed. Rotation. This rotational speed. With the vehicle's current longitudinal speed Maintaining a linear synchronization relationship, that is This is to ensure that the relative linear velocity is suitable for the tennis ball to enter.

[0131] Second, the vehicle continues to move at low speed while maintaining its current course until it covers the target coordinates. During this process, the tennis ball on the ground comes into contact with the rotating elastic rod 413, under compressive force. Overcoming elastic resistance under action Through the gap It enters the pickup roller assembly 41 and is guided by the shovel assembly 43 into the rear transition conveyor channel 51 as it rotates.

[0132] Step S600: Real-time classification and sorting. After the tennis ball enters the transition conveyor channel 51, the system executes the classification logic. The calculation control module 20 retrieves the feature description vector generated for the tennis ball during the visual perception stage. Read the wear rating from it. .like The calculation control module 20 sends a first control signal to the sorting execution mechanism 52 of the tennis ball sorting and storage module 50, driving the flow direction switching baffle 523 to rotate to the guide position leading to the qualified product storage bin 53; if The computing control module 20 sends a second control signal, driving the flow direction switching baffle 523 to rotate to the guide position leading to the waste storage bin 54. Under the influence of gravity, the tennis balls fall into the corresponding storage bin through the sorting actuator 52, completing one picking and sorting cycle.

[0133] Step S700: Loop check and task termination to complete the current objective. After the items are picked up, the calculation and control module 20 performs a status check: First, it queries the status of the non-contact liquid level sensor 55 of the tennis ball sorting and storage module 50. If a full-load signal is received, the system forcibly interrupts the current task, plans a path back to the starting point, and enters the return unloading mode. Second, if the load is not full, it checks the path sequence. Have all the waypoints in the path been traversed? If so... If there are still waypoints remaining, the system returns to step S400 to extract the next waypoint and continue the operation; if The system returns to step S200 to perform a global perception scan again to determine if there are any undiscovered tennis balls. When no valid target is found in multiple consecutive global scans (e.g., 3 times), the calculation and control module 20 determines that the current field has been picked up, controls the vehicle to automatically drive to the user-set return point, sends a job completion notification to the user, and then enters a low-power standby state.

[0134] In this embodiment of the invention, the computing control module 20 has a built-in independently running system monitoring daemon process, configured to monitor changes in the external environment and the internal hardware status in real time, and to execute a preset interrupt handling program in response to unexpected abnormal inputs.

[0135] The special case handling logic specifically includes obstacle avoidance, mechanical jamming self-rescue, and target loss relocation.

[0136] Obstacle avoidance based on artificial potential field method:

[0137] During automatic cruise control, the visual perception module 10 is configured to not only identify the tennis ball target but also detect non-tennis ball obstacles (such as net posts, people, or other equipment) within the field of view. The computational control module 20 uses a semantic segmentation network to map the obstacle regions in the image to a set of obstacle coordinates in the vehicle coordinate system. .

[0138] When obstacle coordinates are detected Located on the current planned path Preset safety passage width During this period, the calculation and control module 20 pauses the pure tracking control algorithm and activates the local obstacle avoidance planning algorithm. In this embodiment, the artificial potential field method is used to generate the obstacle avoidance path.

[0139] The system defines the total potential energy function of the vehicle in the potential field. gravitational potential energy of the target Repulsive potential energy with obstacles sum:

[0140] ;

[0141] in: Indicates the vehicle's current position. Gravitational potential energy. With the vehicle to the current target tennis point The distance is directly proportional to:

[0142] ;

[0143] in: This is the gravitational gain coefficient; For Euclidean distance.

[0144] Repulsive potential energy Only within the influence range of the vehicle entering the obstacle It takes effect at that time, and is defined as:

[0145] ;

[0146] in: This is the repulsive force gain coefficient; This represents the shortest distance from the vehicle to the obstacle.

[0147] The calculation control module 20 calculates the negative gradient of the total potential energy function. As a virtual resultant force, the direction of this resultant force is the vehicle's obstacle avoidance direction. The calculation and control module 20 converts this resultant force direction into a new steering angle control command. The signal is sent to the vehicle drive module 30, which drives the vehicle to bypass the obstacle until the obstacle is out of its range. The system then reverted to its original path planning logic.

[0148] Stall protection and self-rescue of tennis ball picking and sorting mechanisms:

[0149] If a tennis ball gets stuck or a foreign object gets caught in the tennis ball and prevents the mechanical structure from rotating during the operation of the tennis ball picking module 40 or the tennis ball sorting and storage module 50, the system will trigger a stall protection mechanism.

[0150] The motor drive circuits of the vehicle drive module 30 and the tennis ball retrieval module 40 integrate current sampling resistors or Hall current sensors, configured to collect the motor's operating current in real time. The calculation and control module 20 is set with a stall current threshold. and duration threshold .

[0151] The monitoring logic is as follows: If detected And duration (For example, 500ms), the calculation and control module 20 determines that a mechanical stall has occurred. At this time, the system immediately executes the following reverse self-rescue action sequence: First, immediately cut off the forward drive power of the corresponding motor, allowing the motor to stop freely; Second, send a reverse rotation command to the motor, setting the reverse speed to... Continuous operation for a preset time (For example, 2 seconds) Use the reverse torque to remove the stuck tennis ball or foreign object from the mechanical gap; the third step is to resume forward rotation after the reverse rotation ends. If a stalled state is still detected after performing a preset number of reverse self-rescue operations (for example, 3 times), the calculation control module 20 will send a hardware fault alarm signal and permanently stop the operation of the motor, while controlling the vehicle to stop driving and uploading a fault code.

[0152] Dynamic target loss and relocation handling:

[0153] Due to the ball rolling or errors in the visual algorithm, vehicles may appear to be approaching the target road point. During the process, the target tennis ball suddenly disappears from the field of vision of the visual perception module 10.

[0154] In response to this situation, the computational control module 20 executes predictive tracking logic. The system utilizes an extended Kalman filter to maintain the target's state estimate. When visual measurements are interrupted, the system uses the target's position from the previous moment. and the vehicle's own motion displacement Calculate the predicted position of the target relative to the vehicle. .

[0155] If the vehicle travels to the predicted location If the tennis ball is not visually detected again within a preset neighborhood (e.g., a radius of 0.5 meters) and the sensor signal indicating that the ball has entered the pickup device is not triggered, the calculation and control module 20 determines that the target is a false alarm or has been moved. The system then executes an in-situ search strategy: the vehicle stops moving forward and performs a spin scan centered on the current position. If the tennis ball is re-identified during the scan, the target coordinates are updated and the path is replanned; if the target is still not found after rotating 360 degrees, the calculation and control module 20 removes the failed target point from the task list. Remove the vehicle from the list and immediately schedule the next target point in the list as the new travel target to avoid the vehicle getting stuck in an infinite loop at an invalid location.

Claims

1. A device for picking up tennis balls using an autonomous vehicle, characterized in that, include: The visual perception module (10) is installed on the top of the chassis bracket of the autonomous tennis ball picking device (100) and is used to collect environmental images and convert two-dimensional pixel coordinates into three-dimensional physical coordinates in the vehicle coordinate system. The calculation control module (20) communicates with the visual perception module (10) and is used to execute a path planning algorithm and generate control commands based on coordinate and feature information; The vehicle drive module (30) is used to drive the vehicle to perform longitudinal movement and lateral steering in response to the control command; A tennis ball retrieval module (40) is located at the front of the chassis support and is used to roll a tennis ball into the device through mechanical deformation. The tennis ball sorting and storage module (50) is located at the rear end of the tennis ball picking module (40) and is used to sort tennis balls into different storage areas according to the sorting instructions of the calculation control module (20). The power management module (60) is used to independently supply power to each of the above modules.

2. The device for picking up tennis balls using an autonomous vehicle according to claim 1, characterized in that, The tennis ball retrieval module (40) includes: The pickup roller assembly is mounted on the front end of the chassis support via a floating suspension bracket, which allows the pickup roller assembly to move vertically. A drive motor assembly is coupled to the central shaft of the pickup roller assembly to drive its rotation; The guide shovel assembly extends at its front end to the lower tangent position of the pickup roller assembly; The pickup roller assembly has a cage-like structure composed of multiple elastic rods arranged at equal intervals in the circumference, and the chord length distance between adjacent elastic rods is smaller than the standard diameter of the tennis ball to be picked up.

3. The device for picking up tennis balls using an autonomous vehicle according to claim 1, characterized in that, The tennis ball classification storage module (50) includes: A gravity-fed chute-type transition conveying channel is connected to the output end of the tennis ball pickup module (40); The sorting execution mechanism, located at the end of the transition conveying channel, includes an electromechanical actuator and a flow direction switching baffle controlled by the computing control module (20); Qualified product storage warehouse and waste product storage warehouse; The electromechanical actuator is used to drive the flow direction switching baffle to switch between a first position leading to the qualified product storage bin and a second position leading to the waste product storage bin.

4. The device for picking up tennis balls using an autonomous vehicle according to claim 1, characterized in that, The visual perception module (10) includes a global shutter CMOS industrial camera assembly; The vehicle drive module (30) integrates an odometer sensor assembly for collecting wheel rotation speed and an inertial measurement unit for outputting angular velocity and acceleration. The power management module (60) includes multiple DC-DC voltage conversion circuits, which output different levels of DC voltage to the vehicle drive module (30), the computing control module (20) and the sensor circuit respectively.

5. A method for picking up tennis balls using an autonomous vehicle, characterized in that, The apparatus applied to any one of claims 1 to 4 comprises the following steps: System initialization: The computing control module (20) completes hardware self-test; Global perception and planning: The visual perception module (10) acquires images, and the computational control module (20) generates a target point set and executes regional clustering and path planning algorithms to generate a path sequence; Motion control and execution: The computational control module (20) drives the vehicle drive module (30) to cruise and approach the target tennis ball based on the path sequence; Mechanical pickup: When the vehicle approaches the target tennis ball, the tennis ball pickup module (40) rotates and uses the deformation of the elastic rod to roll the tennis ball in; Classification and storage and cycle determination: The calculation control module (20) controls the tennis ball classification and storage module (50) to sort the tennis balls according to the tennis ball wear score, and decides whether to continue the operation or return to the starting point.

6. The method for picking up tennis balls using an autonomous vehicle according to claim 5, characterized in that, The global perception and planning steps include: The inverse perspective mapping algorithm is executed to construct a homography matrix using the camera intrinsic and extrinsic matrices and the assumption of flat ground. Solve for the inverse of the homography matrix to map the image pixel coordinates to three-dimensional physical coordinates in the vehicle coordinate system, and then convert them into global navigation absolute coordinates.

7. The method for picking up tennis balls using an autonomous vehicle according to claim 5, characterized in that, The path planning algorithm is specifically as follows: Calculate the geometric center of the target point set and delineate the operation weight region; A greedy heuristic search strategy is adopted to initialize the set of unvisited points, traverse and iteratively select the candidate point closest to the current node until the sequential waypoint information containing all tennis goals is generated.

8. The method for picking up tennis balls using an autonomous vehicle according to claim 5, characterized in that, The motion control and execution steps employ a pure tracking control strategy: Select a target forward-looking point at a preset forward-looking distance from the vehicle on the path sequence; The desired steering angle is calculated based on the Ackermann steering geometry model and sent to the vehicle drive module (30). The longitudinal speed control amount is calculated based on the remaining distance between the vehicle and the target tennis ball to perform linear deceleration.

9. The method for picking up tennis balls using an autonomous vehicle according to claim 5, characterized in that, The classification storage and loop determination steps include: Read the tennis ball wear rating generated by a deep convolutional neural network; If the wear score is greater than or equal to the preset discard threshold, drive the flow direction switching baffle to rotate to the position leading to the qualified product storage bin; If the wear score is less than the preset waste threshold, the flow direction switching baffle is driven to rotate to the position leading to the waste storage bin.

10. The method for picking up tennis balls using an autonomous vehicle according to claim 5, characterized in that, It also includes stall protection steps: Monitor the real-time operating current of the motor in the tennis ball picking module (40) or the tennis ball sorting and storage module (50); If the real-time operating current exceeds the preset stall threshold and the duration exceeds the time limit, it is determined to be a mechanical stall, and a self-rescue action sequence including cutting off the forward power supply, reversing the rotation, and attempting to resume forward rotation is executed.

Images