Intelligent feeding system combining machine vision and path planning

Abstract

The application relates to the technical field of intelligent manufacturing, and discloses an intelligent feeding system and method combining machine vision and path planning. The system comprises a visual acquisition unit, a visual processing unit, a path planning unit, a trajectory monitoring and re-planning unit, and a motion control unit. The visual processing unit is used for generating environment parameters containing material pose information, obstacle boundary information, and an identification confidence parameter; the path planning unit constructs a space cost field model based on the environment parameters and generates an initial motion trajectory; the trajectory monitoring and re-planning unit triggers local trajectory re-planning according to the obstacle position change amount and the cumulative change trend during the execution of the mechanical arm; and the motion control unit executes hierarchical speed regulation according to the identification confidence parameter and dynamically adjusts the safety distance in combination with the running speed. Through the above closed-loop control mechanism, real-time response to a dynamic environment and adaptive motion control are realized, and the stability and safety of the feeding operation are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and high-end equipment control technology, and in particular to an intelligent feeding system that combines machine vision and path planning. Background Technology

[0002] In automated manufacturing and intelligent production lines, the material handling system is used to pick up and transport materials, and its performance directly affects production cycle time and system stability. With the increasing demand for flexible production, traditional manual material handling methods have been gradually replaced by automated material handling systems.

[0003] Existing automated material handling systems typically employ industrial robotic arms controlled by pre-programmed procedures to perform gripping actions. Their motion trajectories are set during deployment and remain unchanged during execution. While such systems are suitable for scenarios with fixed structures and stable material positions, the pre-programmed trajectories struggle to adapt to environmental changes when material positions shift or temporary obstacles appear, potentially leading to gripping failures or collisions.

[0004] To improve adaptability, existing technologies incorporate machine vision systems to acquire material pose information through visual recognition and send the recognition results to the robotic arm control system for visual guidance and grasping. However, such systems typically only provide a static recognition result once before the action begins. The robotic arm lacks the ability to continuously perceive environmental changes and correct its trajectory during execution, and the overall control structure remains primarily open-loop control.

[0005] Furthermore, in existing systems, visual recognition results are typically used only as position input parameters and are not deeply coupled with motion control strategies. When there is uncertainty in the recognition results, the robotic arm still performs actions at a fixed operating speed, which may lead to grasping errors or safety risks.

[0006] In terms of path planning, existing methods mostly rely on geometric collision detection or discrete maps for path search, only determining whether a collision has occurred without optimizing the continuous distance relationship between the trajectory and obstacles, making it difficult to balance safety margin and motion smoothness. Furthermore, the safety distance is usually set to a fixed value, failing to consider the impact of the robotic arm's current operating speed on the level of risk.

[0007] Therefore, existing feeding systems still have room for improvement in terms of dynamic environment adaptability, perception and control coordination, and safety. It is necessary to provide an intelligent feeding system that can adjust the trajectory in real time according to environmental changes during the execution of the robotic arm and coordinate the optimization of visual recognition results and motion control strategies. Summary of the Invention

[0008] In view of this, the purpose of the present invention is to provide an intelligent feeding system that combines machine vision and path planning, so as to solve the problems of insufficient safety and stability caused by the inability of existing feeding systems to adjust the trajectory in real time according to dynamic changes in the environment during the execution of the robotic arm, and the lack of deep coupling between visual recognition results and motion control strategy.

[0009] To achieve the above objectives, the present invention provides the following technical solution: In one embodiment of the present invention, an intelligent feeding system combining machine vision and path planning is provided, including a vision acquisition unit, a vision processing unit, a path planning unit, a trajectory monitoring and replanning unit, a motion control unit, and an execution unit.

[0010] The system comprises the following components: a vision acquisition unit continuously acquires image data of the loading area; a vision processing unit performs material recognition and 3D pose calculation on the image data to generate environmental parameters including material position information, obstacle boundary information, and recognition confidence parameter P; a path planning unit constructs a spatial cost field model containing continuous distribution information of obstacle boundary distances based on the environmental parameters, and performs minimum cost trajectory search within the spatial cost field model to generate the initial motion trajectory of the robotic arm; a trajectory monitoring and replanning unit acquires two consecutive frames of environmental parameters at preset time intervals T during the robotic arm's execution, calculates the change Δd between the current position and the previous position of the obstacle, and performs local trajectory replanning when preset replanning trigger conditions are met; a motion control unit adjusts the robotic arm's motion speed in stages according to the recognition confidence parameter P; and an execution unit drives the robotic arm to complete the grasping and loading actions based on the replanned motion trajectory.

[0011] Furthermore, the trajectory monitoring and replanning unit triggers local trajectory updates based on the change amount Δd and the cumulative change amount ΣΔd, and the motion control unit adjusts the robotic arm's operating speed based on the recognition confidence parameter P, and dynamically adjusts the safety distance Ds based on the current operating speed v, thereby forming a dual-constraint closed-loop control structure that combines environmental change judgment with recognition confidence speed adjustment.

[0012] Furthermore, the change Δd is the maximum Euclidean distance change of the set of obstacle boundary points in two consecutive frames of environmental data.

[0013] Preferably, the replanning triggering condition includes any of the following conditions: Δd is greater than or equal to the first change threshold D1; the cumulative value ΣΔd of n consecutive changes is greater than or equal to the second change threshold D2; the angle between the obstacle's movement direction and the robot arm's current movement direction is less than a preset angle threshold θ0.

[0014] Furthermore, when a local trajectory replanning is triggered, only the trajectory segment within the range of the current position of the robotic arm to the preset forward distance L is recalculated, while the remaining trajectory segments remain unchanged.

[0015] In one embodiment of the present invention, the comprehensive cost function of any point x in the spatial cost field model is: C(x)=w1·Cobs(x)+w2·Csmooth(x)+w3·Ctime(x); Where Cobs(x) is the obstacle distance cost, Csmooth(x) is the trajectory curvature cost, Ctime(x) is the time cost, and w1, w2, and w3 are adjustable weight coefficients.

[0016] Preferably, the weighting coefficients are dynamically adjusted according to the operating mode, with the weight of w3 increased in high-speed operating mode and the weight of w1 increased in high-safety operating mode.

[0017] Furthermore, the preset safety distance Ds of the robotic arm is dynamically adjusted according to the current movement speed v to satisfy: Ds = D0 + k·v; Where D0 is the basic safety distance and k is the speed compensation coefficient.

[0018] Optionally, the motion control unit executes a graded speed control strategy based on the recognition confidence parameter P. When P is greater than or equal to 0.9, it runs at the rated speed Vmax. When P is in different confidence intervals, it runs at the corresponding proportional speed. When P is lower than the preset confidence threshold, the motion is paused and visual recognition is re-executed.

[0019] In one embodiment of the present invention, the visual acquisition unit includes an RGB camera and a depth camera, used to simultaneously acquire color images and depth images.

[0020] Furthermore, a feeding method based on the aforementioned intelligent feeding system is also provided, comprising: acquiring image data of the feeding area and generating environmental parameters including material pose information, a set of obstacle boundary points, and an identification confidence parameter P; constructing a spatial cost field model containing continuous distribution information of obstacle boundary distances based on the environmental parameters, and performing a minimum cost trajectory search in the spatial cost field model to generate an initial motion trajectory; acquiring two consecutive frames of environmental parameters at preset time intervals, calculating the maximum Euclidean distance change Δd of the obstacle boundary point set, and accumulating the changes several times to obtain a cumulative change ΣΔd; when the change Δd reaches a first change threshold or the cumulative change ΣΔd reaches a second change threshold, performing local replanning only on the trajectory segment within the current position of the robotic arm to the forward preset distance range; adjusting the robotic arm's movement speed in stages according to the identification confidence parameter P, and dynamically adjusting the safety distance according to the current movement speed v to complete the feeding action.

[0021] Based on the above technical solution, the intelligent feeding system combining machine vision and path planning of the present invention acquires and analyzes the environmental information of the feeding area in real time through a vision acquisition unit and a vision processing unit, generating environmental parameters including material pose information, obstacle boundary information, and recognition confidence parameter P. The path planning unit constructs a spatial cost field model containing continuous distribution information of obstacle boundary distances based on the environmental parameters, and performs minimum cost trajectory search in the spatial cost field model to generate a motion trajectory that meets the constraints of obstacle avoidance safety distance, trajectory smoothness, and time efficiency. During the execution of the robotic arm, the trajectory monitoring and replanning unit calculates the obstacle position change Δd and the cumulative change ΣΔd, and triggers local trajectory replanning when a preset change threshold condition is met. At the same time, only the trajectory segment within the range of the current position to the forward preset distance L is recalculated. The motion control unit performs graded speed regulation control according to the recognition confidence parameter P, and dynamically adjusts the safety distance Ds in combination with the current speed v of the robotic arm, so that the safety distance is coupled with the motion state. The functional units form a dual-constraint closed-loop control system based on the environmental change judgment mechanism and the recognition confidence speed regulation mechanism.

[0022] Through the above technical solutions, the present invention solves the problems of existing automated feeding systems being unable to respond in real time to dynamic environmental changes during the execution of the robotic arm, as well as the problems of insufficient safety and poor stability caused by the lack of deep coupling between visual recognition results and path planning and motion control.

[0023] Specifically: On the one hand, by introducing an environmental change Δd determination mechanism, a cumulative change threshold mechanism, and a change direction determination rule, this invention upgrades the replanning trigger condition from the traditional simple collision detection to a trend prediction and quantitative judgment mechanism, avoiding frequent replanning caused by minor disturbances, improving the continuity of trajectory execution and the stability of system operation, and enhancing the adaptability to dynamic obstacle scenarios.

[0024] On the other hand, this invention constructs a continuous distance cost field model in the path planning stage and uses a comprehensive cost function to perform weighted optimization on obstacle distance cost, trajectory curvature cost and time cost, so that the generated motion trajectory not only meets the collision-free requirement, but also has continuity, smoothness and reasonable safety margin, thereby reducing the vibration and impact of the robotic arm operation and improving motion accuracy and execution efficiency.

[0025] Furthermore, this invention directly introduces the recognition confidence parameter P output from the visual recognition stage into the motion control decision-making process. Through a graded speed control strategy, the operating speed of the robotic arm is matched with the perception reliability. When the recognition confidence decreases, the speed is automatically reduced or the movement is paused, thereby reducing the risk of grasping deviation and accidental collision caused by misrecognition and improving the safety of operation in complex lighting, occlusion or material stacking scenarios.

[0026] Furthermore, this invention constructs a speed-coupled dynamic safety distance model Ds=D0+k·v, which enables the safety distance to be adjusted in real time according to the current movement speed of the robotic arm, thereby achieving adaptive matching between the safety margin and the motion state and enhancing the system's risk control capability under high-speed operating conditions.

[0027] In summary, this invention integrates an environmental change monitoring mechanism, a continuous distance cost field planning mechanism, an identification confidence speed adjustment mechanism, and a dynamic safety distance adjustment mechanism to form a closed-loop control structure under multiple constraints. This achieves a deep fusion of real-time perception, dynamic decision-making, and adaptive execution, significantly improving the dynamic adaptability, operational stability, operational safety, and engineering practicality of the intelligent material feeding system in unstructured or semi-structured industrial environments.

[0028] Compared with existing fixed-program feeding systems or single-time vision-guided feeding systems, this invention has significant improvements in environmental dynamic response capability, control accuracy, and system robustness, and has good prospects for widespread application. Attached Figure Description

[0029] Appendix Figure 1 This is a block diagram of the overall structure of the intelligent feeding system of the present invention; Appendix Figure 2 This is a schematic diagram showing the functional module connections of the intelligent feeding system of the present invention; Appendix Figure 3This is a schematic diagram of the closed-loop control process of the intelligent feeding system of the present invention. Detailed Implementation

[0030] To make the technical solution, technical features, and beneficial effects of the present invention clearer, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the following embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit the scope of protection of the present invention. Those skilled in the art can make various modifications or equivalent substitutions to the described embodiments without departing from the spirit and substance of the present invention, and all such modifications or substitutions should fall within the scope of protection of the present invention.

[0031] This invention revolves around the technical framework of "real-time environmental perception—continuous distance optimization planning—change trend determination—adaptive motion control." Through the coordinated operation of the visual perception unit, path planning unit, trajectory monitoring and replanning unit, and motion control unit, a closed-loop control structure under multiple constraints is formed. The following will be discussed in conjunction with the appendix... Figure 1 To be continued Figure 3 The specific implementation methods of the present invention will be described in accordance with the logical order of system structure, functional modules and operation process.

[0032] I. Overall System Structure and Implementation Method (as attached) Figure 1 (As shown) As attached Figure 1 As shown, the present invention provides an intelligent material feeding system combining machine vision and path planning, including a vision acquisition unit 1, a vision processing unit 2, a path planning unit 3, a trajectory monitoring and replanning unit 4, a motion control unit 5, and an execution unit 6. These units interact with each other through an industrial communication network to form a unified control system.

[0033] During system operation, vision acquisition unit 1 first continuously acquires images of the loading area. To ensure coverage of the robotic arm's workspace, vision acquisition unit 1 is typically installed above or slightly above the robotic arm's work area, with its acquisition range covering at least the material area and the main motion trajectory area of ​​the robotic arm. The acquired image data is transmitted to vision processing unit 2 in real time.

[0034] The vision processing unit 2 processes the received image data to extract the three-dimensional pose information of the material and obstacle boundary information, and calculates the recognition confidence parameters. The vision processing unit 2 can be deployed in an industrial control host or edge computing device, and its output environmental parameters include material spatial coordinates, attitude angles, obstacle boundary point sets, and recognition confidence parameters. These environmental parameters serve as the basic input data for subsequent path planning and dynamic monitoring.

[0035] After acquiring the environmental parameters, the path planning unit 3 constructs a spatial cost field model of the robotic arm's current workspace and searches for the optimal or suboptimal motion trajectory that satisfies the constraints within the cost field model, generating the initial motion trajectory of the robotic arm from its current pose to the target grasping pose. The path planning unit 3 can be deployed on the same computing platform as the vision processing unit 2, or it can be set up independently as a real-time planning module, with its output trajectory data sent to the motion control unit 5.

[0036] The execution unit 6 includes a robotic arm body and an end effector. The robotic arm body is fixedly mounted on the work base, and the end effector is mounted at the end of the robotic arm to perform material gripping and releasing actions. The motion control unit 5 generates drive commands for each joint based on the trajectory data output by the path planning unit 3, and controls the execution unit 6 to perform actions according to the motion trajectory.

[0037] During the robotic arm's execution, the trajectory monitoring and replanning unit 4 continuously receives updated environmental parameters from the vision processing unit 2 and analyzes the trend of obstacle position changes in conjunction with the current motion state of the execution unit 6. When an environmental change is detected to meet preset conditions, the trajectory monitoring and replanning unit 4 triggers local trajectory replanning and sends the updated trajectory information to the motion control unit 5. The motion control unit 5 corrects the motion commands of the execution unit 6 based on the new trajectory data, thereby achieving real-time response to the dynamic environment.

[0038] Meanwhile, during execution, the motion control unit 5 also incorporates the recognition confidence parameters output by the vision processing unit 2 to adaptively adjust the operating speed of the robotic arm and dynamically adjust the safety distance based on the current operating speed. This forms a closed-loop collaborative structure between the perception layer, planning layer, and control layer.

[0039] Through the above structural arrangement, the vision acquisition unit 1, vision processing unit 2, path planning unit 3, trajectory monitoring and replanning unit 4, and motion control unit 5 work together to ensure that the robotic arm maintains continuous perception and dynamic response capabilities to the environmental conditions throughout the entire process of performing grasping and loading actions. The system transforms from a traditional static planning execution mode to a closed-loop control mode based on dual constraints of environmental changes and recognition reliability, thereby improving operational stability and safety in complex industrial environments.

[0040] In practical engineering deployments, the units can be centrally located in the same control host or deployed in a distributed structure. Regardless of the deployment method, the data interaction between the units maintains real-time performance and consistency to ensure the stability and response speed of the robotic arm's motion control.

[0041] Through the above overall structural arrangement, this invention provides a complete system architecture foundation for the subsequent implementation of visual recognition algorithms, the construction method of spatial cost fields, and the dynamic replanning control strategy.

[0042] II. Implementation methods of the visual perception module (as shown in the appendix) Figure 2 (As shown) As attached Figure 2 As shown, the visual perception module of the present invention includes a visual acquisition unit 1 and a visual processing unit 2, which work together to generate environmental parameters for path planning and dynamic control.

[0043] During system operation, the vision acquisition unit 1 first performs real-time image acquisition of the material loading area. In one embodiment, the vision acquisition unit 1 includes an RGB camera 11 and a depth camera 12, used to simultaneously acquire color and depth images. The RGB camera 11 is used to acquire texture and color information of the material surface, and the depth camera 12 is used to acquire spatial distance information of the material and the environment. Through a time synchronization mechanism, the color image and the depth image correspond at the same sampling time, thereby ensuring the accuracy of subsequent 3D reconstruction and pose calculation.

[0044] The acquired image data is transmitted to the vision processing unit 2. The vision processing unit 2 first preprocesses the raw image data, including noise reduction, brightness equalization, and distortion correction, to improve image quality and reduce the impact of ambient lighting changes on recognition accuracy. The preprocessed image then enters the feature extraction stage.

[0045] During feature extraction, the visual processing unit 2 segments the material region in the image and extracts material contour features and key point features. For stacked or scattered materials, edge detection algorithms or deep learning recognition algorithms can be used to identify material boundary regions, and the material contours can be spatially mapped by combining depth image information. By fusing two-dimensional image coordinates with depth data, point cloud information of the material in three-dimensional space is obtained.

[0046] After obtaining the material point cloud data, the vision processing unit 2 further calculates the three-dimensional pose information of the material. The three-dimensional pose information includes the spatial coordinates and attitude angle parameters of the material. In one embodiment, the material point cloud can be model-matched using a least-squares fitting method or a pose matching algorithm to calculate the target grasping pose. For regularly shaped materials, a template matching algorithm can be used; for irregularly shaped materials, a pose estimation algorithm based on keypoint matching can be used.

[0047] Simultaneously, the vision processing unit 2 also detects and extracts boundaries of obstacles in the loading area. By analyzing the spatial variation characteristics of the depth image, it extracts the set of boundary points of the obstacles and converts this set into a three-dimensional coordinate form. The obstacle boundary information is used to subsequently construct a spatial cost field model.

[0048] After completing material identification and obstacle extraction, the vision processing unit 2 evaluates the confidence level of the identification results. The identification confidence parameter P reflects the reliability of the current identification result, and its value can be set between 0 and 1. The confidence level can be calculated based on a comprehensive evaluation of parameters such as the output probability value of the identification algorithm, the feature matching consistency index, or the model fitting error.

[0049] Finally, the vision processing unit 2 sends the generated environmental parameter data package to the path planning unit 3 and the trajectory monitoring and replanning unit 4. The environmental parameter data package includes at least the three-dimensional pose information of the material, the set of obstacle boundary points, and the recognition confidence parameter P. This data serves as the basic input information for subsequent trajectory planning and dynamic control.

[0050] During the robotic arm's operation, vision acquisition unit 1 continuously acquires images, while vision processing unit 2 periodically updates environmental parameters, thus providing the system with continuous environmental perception capabilities. This continuous perception mechanism ensures that path planning and motion control are always based on the latest environmental information, providing data support for achieving dynamic closed-loop control.

[0051] Through the above visual perception process, this invention achieves real-time acquisition of material pose and environmental obstacle information, and outputs recognition confidence parameters, providing multi-dimensional input basis for path planning optimization and adaptive adjustment of motion control, thereby enhancing the system's adaptability and stability in complex industrial scenarios.

[0052] III. Implementation methods for path planning (as shown in the appendix) Figure 2 (As shown) As attached Figure 2 As shown, the path planning function of this invention is implemented by the path planning unit 3. After receiving the environmental parameters output by the vision processing unit 2, the path planning unit 3 models the current workspace of the robotic arm and generates a target motion trajectory that satisfies safety constraints and motion constraints.

[0053] During system operation, path planning unit 3 first acquires the current pose information of the robotic arm and the three-dimensional pose information of the material and the set of obstacle boundary points output by vision processing unit 2. Based on the above environmental parameters, path planning unit 3 performs spatial representation modeling of the reachable space of the robotic arm. In one embodiment, the spatial representation adopts a continuous distance distribution form, mapping the obstacle boundary points to a three-dimensional spatial coordinate system, and calculating the distance value from any point in the workspace to the nearest obstacle boundary, thereby constructing a spatial cost field model.

[0054] In the spatial cost field model, each spatial location point corresponds to a comprehensive cost value. This comprehensive cost value is used to reflect a comprehensive evaluation index of the location point in terms of obstacle avoidance safety, motion smoothness, and time efficiency. In one embodiment, the comprehensive cost value function is defined as: C(x)=w1·Cobs(x)+w2·Csmooth(x)+w3·Ctime(x); Where Cobs(x) represents the obstacle distance cost, the value of which is related to the distance from the point to the nearest obstacle boundary; Csmooth(x) represents the trajectory curvature cost, used to constrain the continuity and smoothness of the trajectory; Ctime(x) represents the time cost, used to constrain the overall motion efficiency; w1, w2, and w3 are weighting coefficients.

[0055] Furthermore, in one embodiment, to enable the spatial cost field model of the present invention to be concretely implemented, the calculation method of each cost term is explained as follows: (1) Obstacle distance cost Cobs(x) The obstacle distance cost can be calculated based on the Euclidean distance d(x) from spatial point x to the nearest obstacle boundary, where d(x) represents the minimum Euclidean distance from spatial point x to the nearest obstacle boundary point. In one embodiment, it can be expressed as a function: Cobs(x) = 1 / (d(x) + ε) Where ε is a constant greater than 0; or: Cobs(x) = exp(-α·d(x)) Where α is the distance attenuation coefficient.

[0056] (2) Trajectory curvature cost Csmooth(x) The trajectory curvature cost is used to characterize the smoothness of the trajectory. In one implementation: Csmooth(x) = |θ_i - θ_{i-1}| Wherein, θ_i represents the motion direction angle of the i-th trajectory segment, which is calculated based on the spatial coordinates of adjacent trajectory points; or, the trajectory curvature cost can be calculated based on the second-order difference of discrete trajectory points.

[0057] (3) Time cost Ctime(x) The time cost is used to characterize the path execution efficiency: Ctime(x) = Δs / v Where Δs is the path length between adjacent trajectory points, and v is the current operating speed of the robotic arm; or, the time cost is positively correlated with the cumulative path length.

[0058] Furthermore, in one embodiment, when performing trajectory search in the spatial cost field model, the A* algorithm, Dijkstra's algorithm, or the Fast Random Tree (RRT) algorithm can be used to perform minimum cost path search.

[0059] By employing the above method, each cost term in the comprehensive cost function has a clear calculation method, which enables those skilled in the art to implement the path planning process; the above function form and search algorithm are merely examples, and the present invention is not limited thereto.

[0060] In practical implementation, path planning unit 3 can adjust the weight coefficients according to the work mode. For example, in high-speed production mode, the weight of time cost is increased to improve execution efficiency; in high safety requirement mode, the weight of obstacle distance cost is increased to increase safety margin. Through dynamic adjustment of weights, the path planning results can adapt to the needs of different working conditions.

[0061] After constructing the spatial cost field model, path planning unit 3 performs trajectory search within this model. The trajectory search process starts with the current pose of the robotic arm and ends with the target grasping pose, searching for the path with the minimum overall cost while satisfying the kinematic constraints and joint range constraints of the robotic arm. During the search, both spatial obstacle avoidance requirements and the continuity of the trajectory in the joint space are considered.

[0062] The generated initial motion trajectory is then transmitted to the motion control unit 5 to drive the execution unit 6 to perform the grasping action. At the same time, this trajectory information is also transmitted to the trajectory monitoring and replanning unit 4 to provide baseline trajectory data for subsequent dynamic monitoring and local replanning.

[0063] During the execution of the robotic arm, if the vision processing unit 2 updates the environmental parameters, the path planning unit 3 can recalculate the cost optimization for the local trajectory segment under the trigger command of the trajectory monitoring and replanning unit 4. Since the spatial cost field model is constructed in the form of continuous distance distribution, when the local environment changes, only the cost value of the affected area needs to be updated, without having to remodel the entire workspace, thereby improving real-time response efficiency.

[0064] Through the above-described path planning implementation method, this invention achieves a transformation from traditional "collision detection-based path planning" to "continuous distance optimization-based path planning," enabling the robotic arm's motion trajectory to not only meet collision-free requirements but also maintain a reasonable safety margin in space, while balancing motion smoothness and execution efficiency. This planning method provides a stable foundation for subsequent dynamic replanning and adaptive speed control.

[0065] IV. Implementation methods for dynamic monitoring and replanning (as attached) Figure 3 (As shown) As attached Figure 3 As shown, during the process of the robotic arm performing grasping and loading actions according to the initial motion trajectory, the present invention realizes continuous monitoring and dynamic response to environmental changes through the trajectory monitoring and replanning unit 4, thereby forming a closed-loop control mechanism.

[0066] During system operation, the vision acquisition unit 1 continuously acquires image data of the loading area, and the vision processing unit 2 periodically updates the environmental parameters. The trajectory monitoring and replanning unit 4 receives two consecutive frames of environmental data at a preset time interval T, denoted as Et-1 and Et, respectively. The environmental data includes obstacle boundary point set information.

[0067] The trajectory monitoring and replanning unit 4 first performs matching analysis on the obstacle boundary point sets in the two frames of environmental data and calculates the spatial displacement difference between the corresponding boundary points. In one embodiment, the change Δd is defined as the maximum Euclidean distance change of the obstacle boundary point sets in the two frames of data. This calculation method can quantify the actual movement amplitude of the obstacle in space.

[0068] After obtaining the change Δd, the trajectory monitoring and replanning unit 4 compares it with a preset first change threshold D1. When Δd is greater than or equal to the first change threshold D1, it is determined that the current environment has changed significantly.

[0069] In another embodiment, to avoid frequent replanning triggered by minor disturbances, a cumulative change mechanism is introduced. The trajectory monitoring and replanning unit 4 accumulates the changes over several consecutive periods to obtain the cumulative change ΣΔd. When the cumulative change reaches or exceeds a second change threshold D2, a replanning operation is also triggered. In this way, slow but continuous environmental change trends can be identified.

[0070] Furthermore, in a preferred embodiment, the direction of obstacle movement is also determined. The trajectory monitoring and replanning unit 4 calculates the angle between the obstacle's movement vector and the current direction of the robotic arm's movement. When this angle is less than a preset angle threshold θ0, it indicates that the obstacle has a tendency to move closer to the robotic arm's movement path. In this case, even if the single change amount does not reach the first change threshold, replanning can be triggered.

[0071] When any of the above triggering conditions are met, the trajectory monitoring and replanning unit 4 sends a replanning request to the path planning unit 3. In this invention, to improve real-time performance and reduce computational burden, replanning is only performed on the trajectory segment within the range of the current position of the robotic arm to the forward preset distance L, while the unaffected far-end trajectory segment remains unchanged.

[0072] In its implementation, path planning unit 3 updates the local spatial cost field based on the updated environmental parameters, recalculates the cost distribution within the area affected by environmental changes, and executes a minimum cost search algorithm within this local area to generate a new local trajectory segment. The generated local trajectory segment is then smoothly joined with the unmodified portion of the original trajectory to form a new, complete motion trajectory.

[0073] The updated trajectory data is sent to the motion control unit 5, which corrects the joint control commands of the robotic arm based on the new trajectory information, so that the execution unit 6 continues to run along the updated path.

[0074] Through the aforementioned dynamic monitoring and local replanning mechanism, this invention achieves trend determination and precise response to environmental changes. Compared with the traditional method of overall replanning based solely on collision detection, this invention introduces mechanisms for determining changes, analyzing cumulative changes, and judging directional correlations. This makes the replanning trigger conditions more refined, avoiding frequent calculations caused by minor disturbances and enabling timely trajectory correction when actual risks occur in the environment.

[0075] Meanwhile, by limiting the replanning scope to a local segment, the computational complexity is significantly reduced, the real-time response capability of the system is improved, and the system is made suitable for continuous operation environments in dynamic industrial scenarios.

[0076] Therefore, this invention forms a dynamic closed-loop control process of "continuous monitoring - trend determination - local optimization - smooth splicing - continued execution" during the execution of the robotic arm, thereby enhancing the system's environmental adaptability and operational stability.

[0077] V. Implementation method for speed adjustment based on confidence level (as shown in the appendix) Figure 3 (As shown) As attached Figure 3As shown, during the process of the robotic arm performing grasping and loading actions, this invention not only dynamically monitors environmental changes, but also introduces the recognition confidence parameter P output from the visual recognition stage into the motion control decision-making process. Through the graded speed adjustment mechanism and the dynamic safety distance adjustment mechanism, the collaborative coupling of the perception layer and the control layer is realized.

[0078] During system operation, the vision processing unit 2 performs material recognition and 3D pose calculation while simultaneously assessing the reliability of the recognition results and generating a recognition confidence parameter P. The recognition confidence parameter P ranges from 0 to 1, and its value reflects the degree of matching between the current recognition result and the model or the level of prediction probability.

[0079] When the motion control unit 5 receives the motion trajectory data generated by the path planning unit 3, it also receives the identification confidence parameter P. Unlike traditional motion control methods that rely solely on trajectory information, this invention incorporates the identification confidence parameter P into the speed control decision-making process before generating joint drive commands.

[0080] In one embodiment, the motion control unit 5 executes a graded speed control strategy based on different value ranges of the recognition confidence parameter P. When P is greater than or equal to a preset high confidence threshold, the robotic arm operates at its rated speed; when P is in the medium confidence range, the robotic arm reduces its operating speed according to a preset ratio; when P is lower than a preset low confidence threshold, the robotic arm pauses its current movement and triggers a re-recognition process. This graded speed control mechanism ensures that the robotic arm's operating speed is consistent with the reliability of the visual recognition result, thereby reducing the risk of grasping deviation due to recognition errors.

[0081] During the operation of the robotic arm, the motion control unit 5 also acquires the current operating speed v in real time and adjusts the safety distance according to a preset dynamic safety distance model. In one embodiment, the dynamic safety distance satisfies the following relationship: Ds = D0 + k·v; Where D0 is the basic safety distance and k is the speed compensation coefficient. Using this formula, the safety distance increases as the robotic arm's operating speed increases and decreases as the operating speed decreases, thus achieving an adaptive match between the safety margin and the motion state.

[0082] When the trajectory monitoring and replanning unit 4 detects an environmental change and triggers local replanning, the motion control unit 5 continues to dynamically adjust based on the recognition confidence parameter P and the current speed v while updating the trajectory, so that the path correction and speed control are consistent, avoiding situations where the instantaneous speed is too high or the safe distance is insufficient due to trajectory updates.

[0083] Through the aforementioned recognition confidence adjustment and dynamic safety distance adjustment mechanisms, this invention achieves deep integration of visual perception results and motion execution processes. Compared to existing systems where recognition results are only used to provide target coordinates, this invention allows perception uncertainty to directly participate in motion control decisions, thereby improving grasping stability and operational safety in complex lighting environments, occluded scenarios, or material stacking scenarios.

[0084] Furthermore, in practical industrial applications, when the system operates in a high-cycle production mode, the high-confidence threshold can be appropriately increased to ensure sufficient recognition reliability even at high speeds. When the system operates in scenarios with high safety requirements, the low-confidence threshold can be increased, allowing the system to trigger deceleration or pause actions earlier when there is uncertainty in the recognition. Through the adjustable design of the threshold parameters, the system can adapt to different production needs.

[0085] Thus, this invention forms a multi-factor collaborative control mechanism of "identification confidence participating in speed regulation - speed affecting safety distance - safety distance participating in path execution", which enables the robotic arm to consider both identification reliability and motion risk level during execution, thereby significantly improving the overall stability and safety of the system.

[0086] VI. Implementation Method of Closed-Loop Control Mechanism (as attached) Figure 1 -Appendix Figure 3 (As shown) As attached Figure 1 To be continued Figure 3 As shown, the intelligent feeding system of the present invention forms a multi-layered collaborative closed-loop control mechanism consisting of a visual perception layer, a path planning layer, and a motion control layer in its overall structure.

[0087] In the initial stage of system operation, vision acquisition unit 1 continuously acquires images of the material loading area and transmits the image data to vision processing unit 2. Vision processing unit 2 performs recognition processing on the images, generating environmental parameters including material pose information, obstacle boundary information, and recognition confidence parameter P. These environmental parameters are simultaneously sent to path planning unit 3 and trajectory monitoring and replanning unit 4.

[0088] The path planning unit 3 constructs a spatial cost field model based on the environmental parameters and generates an initial motion trajectory that satisfies the safety distance and motion constraints. The generated motion trajectory is transmitted to the motion control unit 5, which controls the execution unit 6 to complete the grasping and loading actions according to the planned path.

[0089] During execution, the visual acquisition unit 1 continuously acquires images, and the visual processing unit 2 periodically updates environmental parameters, forming a real-time environmental perception capability. After receiving the updated environmental parameters, the trajectory monitoring and replanning unit 4 analyzes the trend of obstacle position changes, calculates the change amount Δd and the cumulative change amount ΣΔd, and determines whether local trajectory replanning is needed based on preset trigger conditions.

[0090] When environmental changes meet the triggering conditions, the trajectory monitoring and replanning unit 4 sends a replanning request to the path planning unit 3. The path planning unit 3 only performs local updates on the trajectory segment within the range from the current position to the preset forward distance. The updated trajectory is then transmitted back to the motion control unit 5, thus completing the trajectory correction.

[0091] Meanwhile, the motion control unit 5 continuously receives the identification confidence parameter P throughout the entire execution process and performs graded speed control based on its value. Simultaneously, it dynamically adjusts the safety distance Ds in conjunction with the current operating speed v to ensure that the safety margin matches the motion state.

[0092] With the above structural arrangement, the system forms the following closed-loop control path during operation: Environmental perception → Environmental parameter generation → Trajectory planning → Trajectory execution → Environmental change detection → Local trajectory correction → Adaptive speed adjustment → Continuous execution.

[0093] In this closed-loop mechanism, the environmental change determination mechanism and the confidence-based speed adjustment mechanism constitute dual constraints. The former is used to determine whether structural adjustments to the trajectory are needed, while the latter is used to control the risk level of motion execution. Together, they enable the robotic arm to maintain high operating efficiency in dynamic industrial environments while promptly reducing its operating speed or adjusting its path when risks increase.

[0094] Compared to traditional open-loop execution methods that rely solely on one-time visual positioning and static path planning, this invention enables the system to make decisions based on the latest environmental information throughout the entire grasping and loading process through continuous perception and trend judgment. Path planning is no longer a one-off action but a dynamic process that can be triggered and updated; motion control no longer simply executes trajectory commands but adaptively adjusts based on perception reliability and operating status.

[0095] Therefore, this invention achieves deep integration between the perception layer, planning layer and control layer, upgrading the system from a "static planning execution type" structure to a "multi-factor coupled dynamic closed-loop intelligent control structure", significantly improving the stability and robustness of the system in complex, semi-structured or dynamic industrial scenarios.

[0096] In practical applications, this closed-loop control mechanism can adapt to complex working conditions such as slight material movement, temporary personnel intervention, or multi-equipment collaborative operation, ensuring that the feeding action is completed continuously under safe and controllable conditions, thereby improving production efficiency and reducing downtime risks.

[0097] VII. Complete Running Example This section illustrates the complete operation of the system of the present invention through a specific industrial application scenario, further demonstrating the engineering feasibility of the technical solution of the present invention. It should be understood that the following parameters are merely examples and do not constitute a limitation on the scope of protection of the present invention.

[0098] In one embodiment, the present invention is applied to a parts loading station on an automated assembly line. The material being gripped is a regular metal structural component, approximately 120 mm × 80 mm × 30 mm in size, and is randomly scattered on the loading platform. The robotic arm is a six-degree-of-freedom industrial robot with a maximum operating speed of 1.5 m / s.

[0099] During system operation, vision acquisition unit 1 acquires images of the material loading area at a frequency of 20 Hz. The RGB camera has a resolution of 1920×1080 pixels, and the depth camera has a ranging accuracy of ±2 mm. Vision processing unit 2 processes the acquired images in real time, identifies the material outline, and calculates the 3D pose information. The recognition confidence parameter P is calculated by combining the model matching probability and the key point consistency score.

[0100] In this embodiment, the confidence level grading threshold is set as follows: The high confidence threshold is 0.9; The confidence threshold is 0.7; The low confidence threshold is 0.5.

[0101] When the recognition confidence level P ≥ 0.9, the robotic arm operates at the rated speed of 1.2 m / s; When 0.7 ≤ P < 0.9, the robotic arm's operating speed drops to 0.8 m / s; When 0.5 ≤ P < 0.7, the robotic arm's operating speed drops to 0.4 m / s; When P < 0.5, the robotic arm pauses its movement and re-executes the recognition.

[0102] In the path planning phase, path planning unit 3 constructs a spatial cost field model. The weights in the comprehensive cost function are set as follows: w1=0.5 (obstacle distance cost weight); w2=0.3 (trajectory smoothing cost weight); w3=0.2 (time cost weight).

[0103] The basic safety distance D0 is set to 80 mm, and the speed compensation coefficient k is set to 0.04. When the robotic arm is running at 1.2 m / s, the dynamic safety distance Ds is calculated as follows: Ds=80 mm+0.04×1200 mm / s=128 mm.

[0104] When the operating speed is reduced to 0.4 m / s, the safety distance is adjusted accordingly: Ds=80 mm+0.04×400 mm / s=96 mm.

[0105] During the robotic arm's grasping process, the trajectory monitoring and replanning unit 4 collects environmental update data at 50 ms intervals. The first change threshold D1 is set to 15 mm, the cumulative change threshold D2 is set to 30 mm, the cumulative number of times n is set to 3, and the direction determination threshold θ0 is set to 30°.

[0106] When the maximum displacement Δd of the obstacle boundary point is detected to be 18 mm at a certain moment, it exceeds the first change threshold D1, and the system immediately triggers local trajectory replanning; when Δd is detected to be 8 mm, 9 mm and 10 mm three times in a row, although it does not exceed D1 in a single time, the cumulative change ΣΔd reaches 27 mm and exceeds 30 mm in the next cycle, and replanning is also triggered.

[0107] Replanning only recalculates the trajectory segment within a 300 mm range in front of the robot arm's current position; the rest of the trajectory remains unchanged. The updated trajectory is smoothly spliced ​​with the original trajectory to ensure motion continuity.

[0108] During an actual operation, slight vibration of the conveyor belt caused a slow shift in the material's position. The system detected this trend through a cumulative change mechanism and triggered a local trajectory correction. Simultaneously, the confidence level dropped to 0.75 due to material edge obstruction, and the motion control unit automatically reduced the operating speed to 0.8 m / s and adjusted the safety distance accordingly. Ultimately, the robotic arm successfully completed the grasping action without collision or grasping deviation.

[0109] As can be seen from the above complete operational examples, this invention can be implemented in dynamic environments: Continuously monitor environmental changes; Trend identification and quantitative analysis; Local trajectory optimization; Reliability identification is used in speed regulation; Speed ​​and safety distance are coupled and adjusted.

[0110] This operating mechanism enables the system to maintain high production efficiency while effectively improving security and stability in dynamic scenarios.

[0111] VIII. Industrial Application Scenarios and Examples To further illustrate the applicability of the intelligent feeding system of the present invention in different industrial scenarios, the actual application of the present invention is described below in conjunction with several typical application environments. It should be understood that the following embodiments are only used to illustrate the application of the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0112] In one embodiment, the present invention is applied to the loading station of an automotive parts stamping production line. The stamped parts are stacked in a semi-random state at the end of the conveyor belt, and their positions are randomly shifted due to conveyor belt vibration and manual replenishment. In this scenario, a vision acquisition unit continuously acquires images of the stamped parts, and a vision processing unit calculates their three-dimensional pose and boundary information in real time. A path planning unit generates a grasping trajectory based on a constructed spatial cost field model. During the movement of the robotic arm, when a slight swaying of the conveyor belt is detected causing an overall shift in the material, the trajectory monitoring and replanning unit triggers local replanning based on the change Δd and the cumulative trend, causing the robotic arm's grasping path to adjust accordingly, thereby avoiding grasping failure due to target offset. A confidence-based speed adjustment mechanism automatically reduces the operating speed in environments with strong light reflection on metal surfaces, improving grasping stability.

[0113] In another embodiment, the present invention is applied to a precision material handling scenario in an electronic component assembly line. In this scenario, the materials are small in size and densely packed, requiring high precision and safety margins in the gripping process. The system increases the obstacle distance cost weight to maintain a large safety distance between adjacent components in the planned trajectory. When the recognition confidence level slightly decreases, the motion control unit automatically reduces the robotic arm's operating speed and dynamically narrows the safety distance adjustment range, thereby ensuring accuracy while avoiding accidental contact with nearby components. This scenario demonstrates the stable operation capability of the present invention in a high-precision, high-density environment.

[0114] In another embodiment, the present invention is applied to a logistics warehousing and sorting system. Materials to be picked up may be temporarily placed manually or transferred collaboratively by other robots, and the environment involves personnel movement and multi-device interaction. The system continuously monitors changes in obstacle boundaries to determine the trend of dynamic obstacles entering the robotic arm's movement area. When a person is detected entering a preset safe area, the trajectory monitoring and replanning unit triggers local replanning or directly suspends movement; simultaneously, it automatically reduces the operating speed when the confidence level decreases due to occlusion, thereby improving system operational safety. This embodiment illustrates the advantages of the present invention in a semi-open dynamic environment.

[0115] In multi-robot collaborative operation scenarios, this invention also demonstrates excellent adaptability. When adjacent robots enter a shared workspace, the vision system treats them as dynamic obstacles and extracts boundary information. Through change determination and orientation angle analysis mechanisms, the system can identify potential path conflict trends in advance and achieve avoidance through local trajectory adjustments. The dynamic safety distance model ensures a reasonable match of safety margins at different operating speeds, thereby guaranteeing stability during collaborative operation.

[0116] As can be seen from the above-mentioned industrial application scenarios, the closed-loop control mechanism constructed by this invention, which consists of "real-time environmental perception - continuous cost field optimization - trend determination - local trajectory replanning - speed adjustment based on recognition confidence - dynamic safety distance adjustment", does not depend on a specific industry or single working condition. It can be applied to various complex industrial environments such as scattered material handling, high-precision assembly, logistics sorting, and multi-device collaboration.

[0117] Therefore, this invention has strong versatility and scalability. It can achieve adaptive operation by adjusting parameters under different production cycle requirements and safety level requirements, and has good prospects for industrial promotion.

[0118] IX. System Hardware and Software Deployment Examples To further illustrate the engineering implementation of the intelligent feeding system of the present invention in a real industrial environment, the hardware and software deployment structure of the system is described below. It should be understood that this embodiment is merely illustrative and does not constitute a limitation on the scope of protection of the present invention.

[0119] In one embodiment, the system of the present invention adopts a layered hardware and software architecture, including a perception layer hardware device, a computing processing layer hardware platform, and a motion execution layer control device.

[0120] In terms of the perception layer hardware, the vision acquisition unit includes an industrial-grade RGB camera and a depth camera. The camera is mounted above the robotic arm's working area via a fixed bracket and connected to the computing layer devices through an industrial Ethernet interface. The camera supports a hardware time synchronization mechanism to ensure that color and depth images are acquired within the same sampling period, thereby improving the accuracy of 3D reconstruction.

[0121] At the computational processing layer, the vision processing unit and path planning unit can be deployed in an industrial control host or a high-performance edge computing device. The computing device includes a multi-core processor and a graphics processing unit (GPU) to support the real-time operation of image recognition algorithms and spatial cost field calculations. In one embodiment, the vision processing algorithm runs in a GPU-accelerated environment to improve the processing speed of image segmentation and pose estimation; the path planning algorithm runs in a real-time control thread to ensure the temporal determinism of trajectory generation.

[0122] The trajectory monitoring and replanning unit can share the computing platform with the path planning unit, or it can be deployed independently on the real-time control server. To ensure real-time performance, environmental change determination and local replanning calculations are run using high-priority threads to avoid delays caused by other tasks consuming resources.

[0123] At the motion execution layer, the motion control unit is deployed inside the robot controller or connected to the robot controller via a high-speed communication interface. This communication interface can be EtherCAT, Profinet, or other real-time industrial communication protocols. The motion control unit converts the planned trajectory data into drive commands for each joint and sends control signals to the robot arm actuator.

[0124] In terms of software structure, the system of this invention can adopt a modular software architecture design. The vision processing module, path planning module, trajectory monitoring module, and motion control module operate as independent functional modules and communicate with each other through data interfaces. In one embodiment, the modules share data through a publish-subscribe mechanism, thereby ensuring low coupling and high scalability between modules.

[0125] To ensure system stability, an exception handling mechanism can be set at the software layer. When visual recognition fails or the recognition confidence level continuously falls below the threshold, the system automatically enters a safe mode; when the communication delay exceeds a preset threshold, the motion control unit pauses the robotic arm's movements and performs a status check.

[0126] In industrial applications, the system supports both centralized and distributed deployment modes. In centralized deployment mode, all computing modules are integrated into the same industrial control host, which is suitable for small and medium-sized production lines. In distributed deployment mode, vision processing, path planning, and motion control modules are deployed on independent devices and interconnected through an industrial network, which is suitable for large-scale or multi-robot collaborative scenarios.

[0127] Through the aforementioned hardware and software deployment methods, this invention enables the coordinated operation of visual recognition, path planning, and dynamic replanning algorithms while ensuring real-time performance. Simultaneously, the modular software structure and standardized industrial communication interfaces provide the system with excellent scalability and maintainability, facilitating integrated applications in various industrial scenarios.

[0128] Therefore, this invention is not only innovative at the algorithm and control logic level, but also has a clear and feasible implementation path at the system engineering level, which can meet the comprehensive requirements of stability, real-time performance and security in actual industrial sites.

[0129] In summary, this invention achieves continuous environmental perception, continuous distance optimization planning, and adaptive motion control for the material feeding process by constructing a multi-layered collaborative structure that includes a visual perception unit, a path planning unit, a trajectory monitoring and replanning unit, and a motion control unit.

[0130] Specifically, this invention upgrades trajectory replanning from traditional collision-triggered control to trend analysis control by introducing a judgment mechanism based on the obstacle position change Δd and its cumulative trend, thereby improving the response accuracy to dynamic environmental changes. By constructing a spatial cost field model with continuous distance distribution and performing weighted optimization search within this model, the motion trajectory maintains a reasonable safety margin and smoothness while meeting the collision-free requirement. By introducing the recognition confidence parameter into motion control decision-making, deep coupling between perception results and execution control is achieved, matching motion speed with recognition reliability. Through a speed-coupled dynamic safety distance model, the safety margin can be adaptively adjusted according to the robotic arm's operating state.

[0131] The aforementioned mechanisms do not exist in isolation, but rather operate collaboratively within a closed-loop control framework. This allows the system to simultaneously consider multiple factors during execution, such as environmental change trends, perception reliability, motion efficiency, and safety margins, thereby achieving the fusion of real-time decision-making and dynamic execution.

[0132] Therefore, compared with existing material feeding systems that rely solely on static visual positioning or single path planning, this invention has significantly improved dynamic environment adaptability, trajectory continuous optimization capability, and safety control stability, enabling efficient, stable, and safe automated material feeding operations in complex, semi-structured, or multi-device collaborative industrial scenarios.

[0133] Those skilled in the art will understand that any equivalent substitutions or improvements made to the above embodiments without departing from the spirit and essence of the present invention should fall within the protection scope of the present invention.

Claims

1. An intelligent feeding system combining machine vision and path planning, characterized in that, include: The vision acquisition unit is used to continuously acquire image data of the material loading area; The vision processing unit is used to perform material recognition and three-dimensional pose calculation on image data, and generate environmental parameters including material position information, obstacle boundary information and recognition confidence parameter P; The path planning unit is used to construct a spatial cost field model containing continuous distribution information of obstacle boundary distances based on the environmental parameters, and to perform minimum cost trajectory search in the spatial cost field model to generate the initial motion trajectory of the robotic arm. The trajectory monitoring and replanning unit is used to collect two consecutive frames of environmental parameters at a preset time interval T during the execution of the robotic arm, calculate the maximum Euclidean distance change Δd of the obstacle boundary point set in the two consecutive frames of environmental data, and perform local trajectory replanning when the preset replanning trigger condition is met. The motion control unit is used to adjust the movement speed of the robotic arm in stages according to the recognition confidence parameter P; The execution unit is used to drive the robotic arm to complete the grasping and loading actions according to the replanned motion trajectory; The trajectory monitoring and replanning unit triggers local trajectory updates based on the change amount Δd and the cumulative change amount ΣΔd. The motion control unit adjusts the robotic arm's running speed based on the recognition confidence parameter P and dynamically adjusts the safety distance Ds based on the current running speed v, thereby forming a dual-constraint closed-loop control structure that combines environmental change judgment with recognition confidence speed adjustment. The replanning trigger condition includes any of the following conditions: Δd ≥ first change threshold D1; The cumulative value of n consecutive changes, ΣΔd, is greater than or equal to the second change threshold, D2. The comprehensive cost function for any point x in the spatial cost field model is: C(x)=w1·Cobs(x)+w2·Csmooth(x)+w3·Ctime(x); in: Cobs(x) represents the obstacle distance cost; Csmooth(x) is the trajectory curvature cost; Ctime(x) represents the time cost; w1, w2, and w3 are adjustable weighting coefficients; The motion control unit executes a graded speed control strategy based on the recognition confidence parameter P: When P ≥ 0.9, operate at the rated speed Vmax; When 0.7 ≤ P < 0.9, it runs at the first proportional speed; When 0.5 ≤ P < 0.7, it runs at the second proportional speed; When P < 0.5, pause the motion and re-execute visual recognition.

2. The system according to claim 1, characterized in that: When a local trajectory replanning is triggered, only the trajectory segment within the range of the current position of the robotic arm to the preset forward distance L is recalculated, while the remaining trajectory segments remain unchanged.

3. The system according to claim 1, characterized in that: The weighting coefficients are dynamically adjusted according to the work mode: In high-speed operation mode, increase the weight of w3; In high-safety operation mode, increase the weight of w1.

4. The system according to claim 1, characterized in that: The preset safety distance Ds of the robotic arm is dynamically adjusted according to the current movement speed v, satisfying: Ds = D0 + k·v; in: D0 is the basic safety distance; k is the speed compensation coefficient.

5. The system according to claim 1, characterized in that: The visual acquisition unit includes an RGB camera and a depth camera, used to simultaneously acquire color images and depth images.

6. A feeding method based on the system according to any one of claims 1-5, characterized in that, include: S1: Collect image data of the loading area and generate environmental parameters including material pose information, obstacle boundary point set and recognition confidence parameter P; S2: Construct a spatial cost field model containing continuous distribution information of obstacle boundary distances based on the environmental parameters, and perform minimum cost trajectory search in the spatial cost field model to generate the initial motion trajectory of the robotic arm; S3: Collect environmental parameters for two consecutive frames at a preset time interval, calculate the maximum Euclidean distance change Δd of the obstacle boundary point set, and accumulate the cumulative change ΣΔd by accumulating the changes over several consecutive times. S4: When the change amount Δd is greater than or equal to the first change threshold, or the cumulative change amount ΣΔd is greater than or equal to the second change threshold, local replanning is performed only on the trajectory segment within the range from the current position of the robotic arm to the forward preset distance. S5: Adjust the robotic arm's movement speed in stages according to the recognition confidence parameter P, and dynamically adjust the safety distance Ds according to the current movement speed v before completing the loading action.

Images