A remotely operated path-holding speed planning and control method, system, and storage medium

By forming a local path queue and using sampling interval and collinearity threshold for filtering, a scalar path speed is generated. Combined with corner speed limiting and terminal braking capability, the problems of path fidelity and smooth continuity in teleoperation are solved, and efficient teleoperation control is achieved.

CN123103649APending Publication Date: 2026-11-17LINGMAI TIANGONG (SHENZHEN) TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611337940.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-31
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing teleoperation technologies cannot simultaneously meet the requirements of path fidelity, responsiveness, and smooth continuity, and suffer from problems such as lateral offset, path deformation, abrupt pauses, and controller bounce.

Method used

By synthesizing discrete target point flows into a local path queue and using sampling interval threshold and collinearity threshold for dual screening, a scalar path velocity is generated. Combined with target velocity feedforward and lag catch-up terms, corner speed limits and terminal stop constraints are implemented, and distributed smoothing is performed using window superimposed smoothing layers.

Benefits of technology

It reduces the risk of lateral offset and path deformation, ensures fast and smooth response, avoids single-frame accidental stops and endpoint drift, and improves the path-keeping capability of teleoperation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123103649A_ABST
    Figure CN123103649A_ABST
Patent Text Reader

Abstract

This invention relates to the field of robot teleoperation control technology, and discloses a teleoperation path-maintaining speed planning and control method, system, and storage medium. The method acquires real-time target input, constructs and simplifies a local path queue, calculates the remaining path length and turning angle, generates a motion-constrained scalar path velocity based on target speed feedforward, lag catch-up, turning angle speed limiting, and terminal braking, and performs arc-length integration along the local path to output robot control commands. This invention can reduce the risks of lateral offset, velocity spikes, and endpoint drift while maintaining the shape of the input path.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the technical field of robot motion control and teleoperation interaction, and more specifically, to a teleoperation path-keeping speed planning and control method, system, and storage medium. Background Technology

[0002] In robot teleoperation and real-time human-machine interaction, operator input typically arrives in discrete frames. Each frame's target position, pose, or joint target must be converted into executable motion commands within a very short control cycle to accurately transmit the master's operational intent to the slave's execution object. To obtain a smooth, continuous output trajectory that conforms to kinematic constraints, various real-time motion command generation methods have been developed. Position filtering methods apply exponential moving averages, moving averages, or low-pass filters to the target position to suppress high-frequency noise and random jitter, thus smoothing the output trajectory. Error feedback methods, based on the target position... The error between the current output position and the target position generates a correction vector, which is then added to the output command to achieve continuous tracking of the target position. Point-to-point online planning methods use the latest target as the new endpoint in each control cycle and replan the entire trajectory using a trajectory planner with limited velocity, acceleration, or jerk to ensure that the output meets kinematic constraints. Fixed-delay buffering methods distribute motion across several future frames and smooth the output using window functions or buffer queues to reduce velocity spikes. Emergency stop or threshold stop methods quickly reduce the velocity to zero when the target is detected to be stationary or the input is lost to prevent the output object from continuing to move. These methods have been widely used in various teleoperation systems and real-time control scenarios, becoming the mainstream means of generating real-time motion commands.

[0003] However, the aforementioned technologies all have inherent defects in practical applications, making it difficult to simultaneously meet multiple requirements such as path fidelity, responsiveness, and smooth continuity. While position filtering methods are simple to implement, the output point may not fall on the path actually drawn by the operator, easily resulting in corner cutting or straightening at corners, leading to poor path fidelity. Error feedback methods can achieve target tracking, but when the target changes direction rapidly, the correction vector may cross the path, forming large triangle shortcuts, additional lateral displacement, or peak speeds, thus altering the path shape intended by the operator. In point-to-point online planning methods, when the operator continuously draws the path, the planner tends to chase the latest endpoint, easily losing intermediate path details. Fixed delay caching methods, if only caching position errors or correction vectors, may still change the path shape and cause drift or inaccurate stopping at the path end. Emergency stop or threshold stop methods are prone to abrupt pauses, inconsistent states, and controller bounce, especially when the underlying servo controller, inverse kinematics solver, filter, and upper-level planner are all present. Summary of the Invention

[0004] The purpose of this invention is to provide a remotely operated path-keeping speed planning and control method, system and storage medium. By synthesizing discrete target point flows into a local path queue and using sampling interval threshold and collinearity threshold for dual screening, the output state always advances along the path intended by the operator, reducing the risk of lateral deviation and path deformation.

[0005] This invention is implemented as follows: a teleoperated path-keeping speed planning and control method, applied to controlling a robot, specifically includes the following steps: S11: Obtain the real-time target input of the current control cycle. The real-time target input includes a position vector, a joint vector, or an end pose composed of position and attitude, and uses Euclidean distance, weighted Euclidean distance, or a weighted sum of position distance and relative rotation angle as the path distance, respectively. S12: Based on the real-time target input, a local target path is formed. When the distance between the current target input and the end point of the path queue or the current output point exceeds the preset sampling interval threshold, the current target input is added to the path queue as a candidate path point, and adjacent path segments that are approximately collinear are simplified and merged. S13: Calculate the remaining path length from the current output state along the local target path to the end of the path, and the turning angle between the current path segment and the next path segment; S14: Generate a scalar path speed based on the target input speed, remaining path length, and turning angle, wherein the scalar path speed is obtained by superimposing the target speed feedforward term and the lag catch-up term, and a speed limit constraint is applied based on the turning angle. S15: Integrate the arc length along the local target path based on the scalar path velocity, and map the integrated output state into robot joint commands, end-effector pose commands, or virtual object motion commands; wherein, the output state is constrained to advance along the local target path, and the smoothness is generated by the velocity planning of the scalar path velocity.

[0006] Further, in S12, forming a local target path based on the real-time target input includes: When the path queue is empty, the current output state is used as the distance reference; when the path queue is not empty, the end point of the path queue is used as the distance reference; when the distance between the current target input and the corresponding distance reference exceeds the preset sampling interval threshold, the current target input is used as a candidate path point; when the path queue is empty, the candidate path point is added to the path queue as the first path point; when there is only one path point in the path queue, the candidate path point is appended as the second path point. When the path queue contains at least two path points, if the angle between the candidate path point and the adjacent path segment formed by the end of the existing path is less than a preset collinearity threshold, the end point of the path queue is updated with the candidate path point to extend the last path segment; otherwise, the candidate path point is added to the path queue to preserve the corner shape. When the length of the path queue exceeds the maximum length limit, the first path point that has been crossed by the current output state is deleted. If the length still exceeds the maximum length limit after deletion, unconsumed path points are simplified within the preset path error range, and the path end point and path points greater than the corner retention threshold are preserved. If the limit still cannot be met, the queue is entered into an overload state, the reception of new path points is suspended, and the output state is controlled to safely decelerate.

[0007] Furthermore, in S14, a scalar path velocity is generated based on the target input velocity, remaining path length, and turning angle, including: The target speed feedforward term is obtained by multiplying the target input speed by a preset feedforward coefficient, and the lag catch-up term is obtained by multiplying the remaining path length by a preset lag catch-up coefficient. The lag catch-up term is set with an upper limit truncation to limit the contribution of the maximum lag catch-up speed. The target speed feedforward term and the lag catch-up term are superimposed to generate a following speed reference. When the target input speed is higher than the preset high speed threshold, the high speed response mode is activated, increasing the upper limit of dynamic speed and acceleration allowed in this cycle.

[0008] Furthermore, speed limits are imposed based on the turning angle, specifically including: The turning speed limit is generated based on the current turning angle; for non-collinear turning angles retained in the path queue, when the directions of adjacent path segments are not continuous, the allowable speed at the turning point is set to zero, the output state reaches the turning point and completes the speed zeroing before accelerating again along the next path segment; for cases using curvature continuous transition segments, the non-zero turning speed limit is determined based on the turning angle, the curvature of the transition segment and the maximum normal acceleration.

[0009] The braking distance is calculated based on the current scalar path speed and the system deceleration capability. When the distance from the output state to the current corner point is less than or equal to the sum of the braking distance and the preset buffer distance, corner speed limiting is enabled. Corner speed limiting is not triggered when the corner angle is less than the preset corner dead zone threshold.

[0010] Furthermore, the generation of scalar path velocity also includes real-time stationary determination and terminal braking, including: The target step size is set as the distance between the target input of the current control cycle and the target input of the previous control cycle. When the target step size is less than the preset static threshold for multiple consecutive frames, the target input is determined to be truly static, so as to avoid misjudgment caused by single-frame jitter or instantaneous pause. After determining that the target is truly stationary, the terminal's braking speed is calculated based on the remaining path length and the system's deceleration capability, and the scalar path speed is constrained to not exceed the terminal's braking speed; when the remaining path length is less than the preset terminal threshold, the output state is snapped to the path terminal and the speed is cleared to zero.

[0011] Furthermore, after generating the scalar path velocity, the process also includes performing motion constraint updates on the scalar path velocity, including: When the scalar path velocity before the update is less than the scalar path velocity target, the scalar path velocity before the update is added to the product of the system acceleration capability and the control cycle duration to obtain the updated velocity, and the velocity target is used as the upper limit for truncation; when the scalar path velocity before the update is greater than or equal to the velocity target, the product of the system deceleration capability and the control cycle duration is subtracted, and the velocity target is used as the lower limit for truncation. Based on the acceleration and deceleration limited update, an acceleration constraint is introduced and acceleration write-back is performed. The difference between the acceleration of the current cycle and the acceleration of the previous cycle is calculated. When the absolute value of the acceleration difference is greater than the product of the preset acceleration upper limit and the control cycle duration, the change of the acceleration of the current cycle relative to the acceleration of the previous cycle is truncated to the range of the product, and the acceleration of the current cycle is updated with the truncated acceleration of the current cycle.

[0012] Furthermore, prior to the path integral in S15, a distributed smooth release of the path arc length increment is performed using a window overlay smoothing layer, specifically including: Set the cache length M (M≥5) to control the time window span for speed smoothing, and assign weights to each cache slot using a Hann window function, a sine window function, or a binomial window function; each weight satisfies 0≤w j <1. If the sum of weights is 1 and the weights of at least two cache slots are greater than 0, the path arc length increment generated in this period is multiplied by the weight of each cache slot and then accumulated to the corresponding cache slot. The output arc length increment of the current control cycle is the current value of the first buffer slot in the buffer queue. After being retrieved, the buffer queue is shifted one position to the left and the last buffer slot is cleared to zero, so that the newly added arc length increment is gradually released in subsequent cycles to reduce speed spikes. When terminal snapping, emergency stop, communication interruption, safety pause or path reset occurs, the buffer vector is cleared, the output arc length increment of the current cycle is set to zero, and the scalar velocity state and acceleration state are reset.

[0013] Compared with the prior art, the remotely operated path-holding speed planning and control method, system, and storage medium provided by the present invention have the following beneficial effects: 1. By synthesizing discrete target point flows into a local path queue and using sampling interval threshold and collinearity threshold for dual screening, the output state always advances along the path intended by the operator, reducing the risk of lateral deviation and path deformation. At the same time, the synergistic superposition of target speed feedforward and lag catch-up not only ensures rapid follow-up response on straight sections, but also actively intervenes at curves through corner adaptive speed limit and early braking window, solving the problems of path deviation and peak speed caused by high-speed cornering. The combination of continuous multi-frame static confirmation and terminal stop constraint eliminates single-frame false stop and endpoint drift. In addition, the speed, acceleration and jerk constraints in the entire control link all act on the path arc length direction, so that the scalar path speed and its acceleration change meet the preset motion constraints and reduce sudden changes in motion commands. 2. By introducing maximum length limits and collinearity simplification strategies in path queue maintenance, the balance between computational overhead and path shape preservation is effectively controlled, avoiding real-time performance degradation caused by excessive accumulation of path points. The adjustable design of the minimum angular velocity ratio and curve exponent in the corner speed limiting function allows the system to configure deceleration curves differently according to the dynamic characteristics of different robots and task scenarios. In fine operation mode, the deceleration sensitivity can be increased to ensure trajectory accuracy, while in fast movement mode, the sensitivity is reduced to maintain motion efficiency. The joint configurability of the stationary threshold and the number of consecutive frames in the real stationary determination makes it compatible with input devices with different noise characteristics, such as mice, gamepads, VR controllers, and optical trackers. The window overlay smoothing layer distributes the arc length increment within the time window, further suppressing speed spikes and mechanical shocks. The parameter adaptive module can dynamically adjust control parameters according to the remaining path change trend, following error, and robot limit status, ensuring that the system has automatic catching-up capability when the path continues to grow and active speed limiting protection when approaching joint boundaries.

[0014] A remotely operated path-keeping speed planning and control system, configured to perform the above-described method, the system comprising: The target input module is used to receive real-time target position, target pose, joint target, end-effector pose or multi-source fusion target, and provides timestamp and status marker; The path queue module is used to maintain local target paths based on sampling interval thresholds, collinearity thresholds, and maximum queue length, and to perform lightweight simplification on adjacent path segments that are approximately collinear. The input state estimation module is used to calculate the target velocity, target increment, number of consecutive static frames, and whether the target is truly stationary. The path geometry analysis module is used to calculate the remaining path length, current turning angle, distance to the turning point, path segment direction, and path terminal status. The scalar velocity planning module is used to generate scalar path velocities based on target velocity feedforward, lag catch-up, dynamic speed limit, corner speed limit, and terminal stop speed, and to update the scalar path velocities by performing at least one of the following motion constraints: acceleration limit, deceleration limit, acceleration limit, deceleration limit, or jerk limit. The path integration module is used to integrate the scalar path velocity along the local target path, traversing the reached path points and outputting the current position or status. The window overlay smoothing module is used to perform distributed overlay smoothing of velocity increments or path arc length increments using sine window functions, Hann window functions, binomial window functions, or triangular window functions. The robot mapping module is used to map the position or state output by the path integration module into robot joint commands, end-effector pose commands, velocity commands, servo commands, or simulation object commands. The parameter adaptive module is used to dynamically adjust the parameters of each module based on the following error, corner radius, task mode, robot limit, or operator speed.

[0015] Specifically, the scalar velocity planning module includes: The dynamic speed limit unit is used to dynamically adjust the allowable speed limit for the current cycle based on the target input speed, the system's maximum safe speed, robot joint limits, or the task mode. The motion constraint unit is used to update the scalar path velocity by performing at least one of the following motion constraints: acceleration constraint, deceleration constraint, acceleration constraint, deceleration constraint, or jerk constraint, to ensure that the output satisfies the kinematic constraints. The high-speed response unit is used to activate the high-speed response mode when the target input speed is higher than the preset high-speed threshold, increase the upper limit of dynamic speed and acceleration allowed in this cycle, and restore to the basic parameters after the target input speed decreases; The scalar velocity planning module adaptively adjusts the feedforward coefficient and the lag catch-up coefficient according to the changing trend of the remaining path length: when the remaining path length continues to increase, the lag catch-up coefficient is increased to enhance the catching-up capability; when the remaining path length continues to decrease, the lag catch-up coefficient is decreased to avoid overshoot.

[0016] A remotely operated path-keeping speed planning and control storage medium stores a computer program that, when executed by a processor, implements the above-described method. Attached Figure Description

[0017] Figure 1 This is a schematic flowchart of the remote operation path-keeping speed planning and control method proposed in this invention. Figure 2This is a simplified diagram illustrating the local target path construction and collinearity in a remotely operated path-maintaining velocity planning and control method proposed in this invention. Figure 3 This is a schematic diagram illustrating the integral progression of the output state along the local path arc length in a remotely operated path-maintaining velocity planning and control method proposed in this invention. Figure 4 This is a schematic diagram of the adaptive speed limit and early braking window in a remotely operated path-holding speed planning and control method proposed in this invention; Figure 5 This is a schematic diagram of the real stationary determination and terminal braking constraint in a remotely operated path-keeping speed planning and control method proposed in this invention. Figure 6 This is a schematic diagram of the window overlay smoothing layer structure in the remote operation path-keeping speed planning and control method proposed in this invention; Figure 7 This is a schematic diagram of the structure of a remotely operated path-keeping speed planning and control system proposed in this invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] The implementation of the present invention will be described in detail below with reference to specific embodiments.

[0020] In the accompanying drawings of this embodiment, the same or similar reference numerals correspond to the same or similar components. In the description of this invention, it should be understood that if terms such as "upper," "lower," "left," and "right" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting this invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0021] Reference Figure 1-6 As shown, a teleoperated path-keeping speed planning and control method, applied to controlling a robot, specifically includes the following steps: S11: Obtain the real-time target input for the current control cycle. The real-time target input includes position vector, joint vector, or end pose composed of position and attitude, and uses Euclidean distance, weighted Euclidean distance, or weighted sum of position distance and relative rotation angle as the path distance, respectively. In this embodiment, the target input module receives target data output from a mouse, gamepad, VR controller, optical tracker, inertial tracker, force feedback device, or multi-source fusion device, and provides timestamps and status markers. S12: Form a local target path based on the real-time target input. When the distance between the current target input and the end point of the path queue or the current output point exceeds the preset sampling interval threshold, add the current target input as a candidate path point to the path queue, and simplify and merge adjacent path segments that are approximately collinear. In this embodiment, the path queue module samples the target point flow at a fixed control period. When the target step size is greater than the sampling interval, path point appending is triggered. At the same time, a maximum length limit is set for the path queue. When the queue length exceeds the upper limit, the first path point that has been crossed by the current output state is deleted to prevent the path from growing indefinitely. Through dual screening of sampling interval threshold and collinearity threshold, high-frequency noise and overly dense points are filtered out, while the true corner shape is preserved. Specifically, the simplification and merging of approximately collinear adjacent path segments includes: Let the control period be k, and the target input position be... x∈ k tar R d The output position is x∈ k cmd R d Where d can be 1, 2, 3 or higher dimension, the local path queue is represented as an ordered sequence in chronological order, and the elements of the following formula are arranged in index order:

[0022] P k The local target path queue in the kth control cycle is an ordered sequence of multiple path points arranged in chronological order. p k,i : The position vector of the i-th path point in the path queue (can be 1-dimensional, 2-dimensional, 3-dimensional or higher-dimensional); N k The total number of path points in the path queue during the k-th control cycle; When the distance between a new target point and the end point of the path queue or the current output point is greater than the sampling interval threshold sp, the target point is taken as a candidate path point; when the path queue is empty, the current output point is used as the distance reference, and when the path queue is not empty, the end point of the path queue is used as the distance reference.

[0023]

[0024] x k tar : The position vector of the target input in the kth control cycle; xk cmd : The position vector of the output state in the kth control cycle; Euclidean distance norm, representing the spatial distance between two vectors; s p : Preset sampling interval threshold, used to determine whether to add the current target point to the path queue; If N k If N = 0, then the candidate path point that satisfies the sampling interval condition will be added to the path queue as the first path point; if N k =1, and the distance between the candidate path point and the existing path point is greater than the sampling interval threshold sp, then the candidate path point is appended as the second path point; if N k If ≥2, then the direction of the previous segment is recorded as: ; When the angle between a and b is less than the collinearity threshold θ c In this case, the system extends the last segment of the path using the current target point, instead of adding a new path point:

[0025] When the included angle is not less than the threshold, the system adds the current target point to the path queue to preserve the corner shape. In this way, the local path queue can filter high-frequency noise and overly dense points without straightening the real corner. a: The direction vector of the last path segment in the path queue, pointing from the second-to-last path point to the end point; b: The direction vector from the end point of the path queue to the current target point; p k,Nk The last path point (end point) in the path queue; p k,Nk-1 The second-to-last path point in the path queue; ∠(a,b): The angle between direction vectors a and b; θ c : A preset collinearity threshold is used to determine whether candidate path points are approximately collinear with existing path directions; ←: Assignment operator, indicating that the value on the right is used to update the variable on the left; To prevent paths from growing indefinitely, the path queue is set to a maximum length N. max ;When Nk>N max When the current output state has already crossed the first path point in the queue, delete the path point at the head of the queue. If the maximum length limit is still exceeded after deletion, simplify the unconsumed path points within the preset path error range, and retain the path endpoint and path points greater than the corner retention threshold. If the limit still cannot be met, enter the queue overload state, suspend the reception of new path points and control the output state to safely slow down. S13: Calculate the remaining path length from the current output state along the local target path to the end of the path, and the turning angle between the current path segment and the next path segment; In this embodiment, the path geometry analysis module calculates the remaining path length by summing the remaining distance of the current segment and the lengths of the subsequent segments based on the geometric relationship between the current output position and the first segment of the path queue. It also calculates the angle between the direction of the current path segment and the direction of the next path segment as the turning angle, and the distance from the output state to the current turning point. These geometric quantities are used to determine whether it is necessary to slow down in advance before reaching the corner. Specifically, the remaining path length is the length calculated by the system from the current output point to the end of the path before the output point moves forward along the path:

[0026] When the path queue contains at least two waypoints, the turning angle between the current path segment and the next path segment can be represented as:

[0027] The distance from the output point to the current corner point is:

[0028] If the path queue has fewer than two points, then we can let θ k =0, and set the corner speed limit constraint to the dynamic speed limit. The above path geometry is used to determine whether it is necessary to slow down in advance before reaching the corner, rather than correcting the output position after the corner has been passed. L k : The remaining path length from the current output state to the end point of the path in the kth control cycle; p k,1 The first path point in the path queue (the head of the queue, i.e., the nearest corner point); θ k : The turning angle between the current path segment and the next path segment in the kth control cycle; p k,2 The second path point in the path queue; d k corner : The distance from the current output state to the current turning point (the path point at the head of the queue) in the kth control cycle; S14: Generate a scalar path speed based on the target input speed, remaining path length, and turning angle. The scalar path speed is obtained by superimposing the target speed feedforward term and the lag catch-up term, and a speed limit constraint is applied based on the turning angle. In this embodiment, the input state estimation module calculates the target input velocity based on the target input increment of adjacent frames. The scalar velocity planning module superimposes the target velocity feedforward term with the lag catch-up term based on the remaining path length to generate a following velocity reference. When the target input velocity is higher than a preset high-speed threshold, a high-speed response mode is activated, increasing the allowed dynamic velocity and acceleration limits for the current cycle. This mode only changes the velocity planning constraint parameters without pruning the local target path. While generating the scalar path velocity, the system calculates the corner speed limit value based on the current corner angle using a corner speed limit function that includes the minimum angular velocity ratio and the corner speed limit curve exponent. The larger the corner angle, the lower the allowed angular velocity; the smaller the corner angle, the closer the allowed angular velocity is to the dynamic velocity limit. For corner angles less than a preset corner dead zone threshold, no cornering is triggered. Speed ​​limiting; the system calculates the braking distance based on the current scalar path speed and the system deceleration capability. When the distance from the output state to the current corner point is less than or equal to the sum of the braking distance and the preset buffer distance, corner speed limiting is enabled to reduce the speed of the output state to the allowable angular velocity before reaching the corner. The system sets the target step size as the distance between the target input of the current control cycle and the target input of the previous control cycle. When the target step size is less than the preset stationary threshold for multiple consecutive frames, the target input is determined to be truly stationary to avoid misjudgment caused by single-frame jitter or instantaneous pause. After determining that the target is truly stationary, the terminal can stop speed is calculated based on the remaining path length and the system deceleration capability, and the scalar path speed is constrained to not exceed the terminal can stop speed. When the remaining path length is less than the preset terminal threshold, the output state is snapped to the path terminal and the speed is cleared to zero. Specifically, the scalar path velocity is calculated by superimposing the target velocity feedforward term and the lag catch-up term, including: The system calculates the target velocity based on the adjacent frame increments of the target input:

[0029] To accommodate path remnant elimination, this invention superimposes the target velocity feedforward term and the lag catch-up term to obtain a following velocity reference:

[0030] Among them, K ff K is the target velocity feedforward coefficient. lag L is the lag catch-up coefficient. cap The upper limit of the remaining path length is used to improve the output speed when the path lag is long, but through L cap Limit its maximum impact and avoid excessive speed due to long-distance pursuit; The system can activate a high-speed response mode based on the target input speed, when v k tar Above the high speed threshold v fast At that time, increase the maximum allowable speed and acceleration for the current cycle:

[0031] When v k tar ≤ fast At that time, let v k lim =min(v max v abs ), a k acc =a acc a k dec =a dec This ensures that the dynamic speed limit in both the basic mode and the high-speed response mode does not exceed the absolute safe speed limit; the high-speed response mode only changes the speed planning constraints and does not prune local paths. The speed limit constraint based on the turning angle is as follows: When the output point approaches a path corner, maintaining excessive speed will result in excessive corner detours, overshooting the path point, or forming an excessively large arc. The allowable angular velocity is calculated based on the turning angle; the larger the turning angle, the lower the allowable speed. Smaller angles or near-straight lines result in less or no speed reduction. The turning angle speed limit function is:

[0032] Where, r min This applies only to implementations where path corners have been converted into curvature-continuous transition segments; for directionally discontinuous corners retained in the path queue, let r min =0 and set the permissible speed at the corner point to 0; p is the exponent of the corner speed limit curve; to avoid unnecessary speed reduction triggered by slight changes in direction, the corner dead zone θd can be set:

[0033] The system calculates the braking distance based on the current speed and deceleration capability:

[0034] When the distance from the output point to the corner satisfies:

[0035] The system then reduced the current speed limit to v. k corner Among them, B s The buffer distance can be determined by the sampling interval, control cycle, current speed or safety factor. This mechanism enables the system to slow down in time before entering the corner and keep up with the train when it is still far from the corner, without slowing down too early. For discontinuous turns in the path queue, the system reduces its speed to zero after entering the braking window. After the output state reaches the turn point and the speed returns to zero, it accelerates again along the next path segment. Only when a curvature-continuous transition segment is generated is it allowed to pass through the turn at a non-zero speed constrained by curvature and maximum normal acceleration.

[0036] Specifically, after determining that the target is truly stationary, the terminal braking speed is calculated based on the remaining path length and the system's deceleration capability, and the scalar path speed is constrained to not exceed the terminal braking speed. The calculation includes: Real-time input often contains slight jitter or short pauses. If the target increment in a single frame is less than the threshold and the system is judged to be stationary, the system is prone to stopping falsely. If the system is never judged to be stationary, the output point may continue to drift after the target stops. This invention uses continuous frame stationary confirmation. Let the target step size be:

[0037] A static counter is defined as:

[0038] When C k still ≥N s If the target is determined to be truly stationary, the system then calculates the braking speed based on the remaining path length:

[0039] The speed reference is set as the minimum of the following upper limits of constraints, i.e., the following equation holds true:

[0040] When the remaining path length is less than the terminal threshold Lϵ, the system will attach the output point to the end of the path and reset the speed. This process can make the output object smoothly reach the end point after the operator has really stopped, reducing the drift and low-speed oscillation of the end point. Explanation of parameters in S14: v k tar : Target input speed in the kth control cycle; x k-1 tar : The position vector of the target input in the previous control cycle; Δt: Control cycle duration (time interval between two adjacent frames); v k follow : The reference value of the following speed in the kth control cycle; K ff : Target velocity feedforward coefficient, used to adjust the contribution weight of the feedforward term; Klag The lag catch-up coefficient is used to adjust the contribution weight of the lag catch-up term. L cap The maximum remaining path length is used to limit the maximum impact of lagging catch-up items; v fast : Preset high-speed threshold to trigger high-speed response mode; v k lim : The maximum dynamic speed allowed in the kth control cycle; β v Speed ​​amplification factor (greater than 1), used to increase the speed limit in high-speed response mode; v max Base speed limit; v abs : The absolute safe speed limit; the speed limit shall not exceed this value. a k acc The system acceleration capability during the k-th control cycle; a k dec The system deceleration capability during the k-th control cycle; β a : Acceleration amplification factor (greater than 1), used to improve acceleration and deceleration capabilities in high-speed response mode; a acc Basic acceleration capability; a dec Basic deceleration capability; v k corner : The turning speed limit value (allowed angular velocity) for the kth control cycle; r min Minimum angular velocity ratio (between 0 and 1) determines the minimum permissible speed at the maximum turning angle; p: Angle speed limit curve index (greater than 0), controls the curvature of the angle speed limit curve; θ k / 2: Half the turning angle; θ d : Preset corner dead zone threshold. When the corner is smaller than this value, it is considered an approximate straight line and corner speed limit is not triggered; D k brake : Braking distance in the kth control cycle, which is the minimum distance required to reduce the speed from the current value to the corner speed limit; v k : The current scalar path speed in the kth control cycle; B s: Preset buffer distance, used to compensate for braking distance calculation errors and system response delay; δ k tar : Target step size of the kth control cycle (displacement between target inputs in adjacent frames); C k still The static counter value in the kth control cycle; C k-1 still The static counter value of the previous control cycle; ϵ s : A preset static threshold is set. When the target step size is less than this value, the target is considered to be in a static state. v k stop : The terminal stop speed in the kth control cycle, that is, the maximum speed at which the system can safely stop with deceleration within the remaining path length; v k ref : The final speed reference value in the kth control cycle; S15: Integrate the arc length along the local target path based on the scalar path velocity, and map the integrated output state into robot joint commands, end-effector pose commands, or virtual object motion commands; wherein, the output state is constrained to advance along the local target path, and the smoothness is generated by the velocity planning of the scalar path velocity. In this embodiment, the path integration module consumes arc length increments starting from the first segment of the path queue. If the arc length increment is less than the remaining distance of the current segment, it advances the output state along the direction of the current segment. If the arc length increment is greater than the remaining distance of the current segment, the output state reaches and pops the first path point in the queue, then continues to consume the remaining arc length along the next path segment. When performing motion constraint updates on the scalar path velocity, if the scalar path velocity before the update is less than the scalar path velocity target, the velocity before the update is added to the product of the system acceleration capability and the control cycle duration as the updated velocity, and truncated with the velocity target as the upper limit. When the scalar path velocity is greater than or equal to the velocity target, the product of the system deceleration capability and the control cycle duration is subtracted, and the velocity target is used as the lower limit for truncation. Based on the constrained acceleration and deceleration update, jerk constraints are introduced, and acceleration write-back is performed. The difference between the current cycle acceleration and the previous cycle acceleration is calculated. When the absolute value of the acceleration difference is greater than the product of the preset jerk upper limit and the control cycle duration, the change in current cycle acceleration relative to the previous cycle acceleration is truncated to the range of this product, and the velocity is updated with the truncated current cycle acceleration, thereby ensuring that the scalar path velocity and its acceleration change meet the preset operational requirements. Dynamic constraints are imposed to reduce abrupt changes in motion commands. Before path integration, a smoothing layer is overlaid with a window to distribute and smoothly release the path arc length increment. The buffer length M (M≥5) is set, and a weight is assigned to each buffer slot using a Hann window function, a sine window function, or a binomial window function. The sum of the weights of all buffer slots is 1. The path arc length increment generated in this cycle is multiplied by the weight of each buffer slot and then accumulated to the corresponding buffer slot. The output arc length increment of the current control cycle is the current value of the first buffer slot in the buffer queue. After being retrieved, the buffer queue is shifted one position to the left and the last buffer slot is cleared to zero, so that the newly added arc length increment is gradually released in subsequent cycles to reduce speed spikes. The robot mapping module maps the output state of the path integration module to robot joint commands, end-effector pose commands, speed commands, servo commands, or simulation object commands. The parameter adaptation module dynamically adjusts the parameters of each module according to the following error, corner radius, task mode, robot limit, or operator speed. When the robot feedback information shows that the joint is close to the limit, the speed is close to the upper limit, or there is a risk of collision, the dynamic speed upper limit of the scalar path speed is reduced or the path advancement is paused until the safety conditions are restored. When a terminal snap-in, emergency stop, communication interruption, safety pause, or path reset occurs, the system clears the cache vector B of the window overlay smoothing layer, and sets the output arc length increment Δs for the current cycle. k out =0, and reset the scalar velocity state and acceleration state; after the safety conditions are restored, the path arc length increment is regenerated from the zero buffer state.

[0041] The cache length is M≥5 to ensure that symmetric windows such as Hann windows have at least two non-zero weights and form an effective distributed release; the weight of each cache slot satisfies 0≤w.j <1. The sum of weights is 1 and the weights of at least two cache slots are greater than 0; the larger the M value, the smaller the proportion of motion released in a single cycle, the more significant the smoothing effect, but the greater the system response latency; the smaller the M value, the faster the response, but the weaker the smoothing effect. The cache length M is configured according to the system control cycle, the desired smoothness, and the allowable response latency. A larger M value is selected in precision assembly scenarios, and a smaller M value is selected in high-speed following scenarios, provided that M≥5 and the above weight conditions are met.

[0042] Specifically, the arc length is integrated along the local target path based on the scalar path velocity, and the integrated output state is mapped to robot joint commands, end-effector pose commands, or virtual object motion commands. The arc length integral output includes: Using a one-dimensional scalar path velocity v k This represents the speed at which the output point moves along the local path. The speed target can be expressed as:

[0043] In the non-stationary state, v can be omitted. k stop Alternatively, we can set it to infinity. Scalar velocity updates can employ different motion constraints. The update method with limited acceleration and deceleration is as follows:

[0044] Further constraints such as jerk constraints, velocity curve basis functions, S-curves, or other online trajectory planning constraints can be added. The key is that velocity planning acts on the path arc length direction, rather than directly filtering the position error vector. The arc length along the path during this period is:

[0045] The path integration module consumes Δs starting from the first segment of the queue. k If Δs k If the distance is less than the distance to the next path point, then update the output along the current segment direction:

[0046] If Δs k If the distance is greater than the remaining distance of the current segment, the output point reaches the head path point of the queue, pops the path point, and continues to move along the next path segment with the remaining arc length. In this way, the output point is always located near the local path or its end, and will not deviate from the operator's input trajectory due to lateral error correction. Specifically, before path integration, a distributed smoothing release of the path arc length increment is performed using a window-overlay smoothing layer, including: Distributed smoothing is performed on the speed increment or path arc length increment. Let the cache length be M (M≥5) and the weight of the j-th cache slot be w.j Each weight satisfies 0 ≤ w j <1. The sum of weights is 1 and the weights of at least two cache slots are greater than 0:

[0047] Hann window weight representation:

[0048] Alternatively, a sinusoidal window, binomial window, or other weighting functions with continuous beginning and end can be used, assuming the path arc length increment generated by the velocity state layer in this cycle is δ. sk The cache vector is Then it can be executed:

[0049] The buffer is then shifted to the left and the tail slot is cleared. This layer allows the new motion to be released gradually, further reducing the velocity spike. Unlike traditional position buffering, this invention preferably applies the buffer to the path arc length, velocity increment, or motion along the path direction, so that the output still advances along the target path. When a terminal snap, emergency stop, communication interruption, safety pause, or path reset occurs, the buffer vector B is cleared and Δs is set. k out =0, to prevent residual motion in the cache from being released again after stopping or resuming.

[0050] Explanation of parameters in S15: v k+1 : Scalar path velocity after update (next control cycle); v k goal : Scalar path velocity target; v k : Scalar path velocity before update (current control cycle); Δs k : The arc length increment along the path during the kth control cycle; x k+1 cmd : Output state position vector for the next control cycle; w j The weight coefficient of the j-th cache slot; M: Cache length of the window overlay smoothing layer; r: The summation index variable, which takes values ​​within the same range as j; δ sk : The path arc length increment generated by the velocity state layer in this cycle; B: Cache vector, containing M cache slots B0, B1, ..., BM-1; B j: The current value of the j-th cache slot; Δs k out The actual arc length increment after the overlay layers in this cycle.

[0051] Reference Figure 7 As shown, a teleoperated path-holding speed planning and control system is configured to execute the above-described method. The system includes: a target input module for receiving real-time target position, target attitude, joint target, end-effector pose, or multi-source fusion target, and providing timestamps and status markers; a path queue module for maintaining local target paths based on sampling interval thresholds, collinearity thresholds, and maximum queue length, and performing lightweight simplification on adjacent path segments that are approximately collinear; an input state estimation module for calculating target speed, target increment, number of consecutive stationary frames, and whether the target is truly stationary; a path geometry analysis module for calculating the remaining path length, current turning angle, distance to the turning point, path segment direction, and path terminal state; and a scalar speed planning module for calculating target speed feedforward, lag catch-up, dynamic speed limit, turning angle speed limit, and terminal braking speed. The system generates a scalar path velocity and updates it with at least one of the following motion constraints: acceleration limit, deceleration limit, acceleration limit, deceleration limit, or jerk limit. A path integration module integrates the scalar path velocity along the local target path, traversing reached path points and outputting the current position or state. A window overlay smoothing module uses a sinusoidal window function, Hann window function, binomial window function, or triangular window function to perform distributed overlay smoothing on the velocity increment or path arc length increment. A robot mapping module maps the position or state output by the path integration module to robot joint commands, end-effector pose commands, velocity commands, servo commands, or simulation object commands. A parameter adaptation module dynamically adjusts the parameters of each module based on following error, corner radius, task mode, robot limits, or operator speed. In this embodiment, the scalar velocity planning module specifically includes a dynamic velocity upper limit unit, a motion constraint unit, and a high-speed response unit. The dynamic velocity upper limit unit is used to dynamically adjust the allowable velocity upper limit for the current cycle based on the target input velocity, the system's maximum safe velocity, robot joint limits, or task mode. The motion constraint unit is used to update the scalar path velocity with at least one of the following motion constraints: acceleration limit, deceleration limit, acceleration limit, deceleration limit, or jerk limit, to ensure that the output meets the kinematic constraints. The high-speed response unit is used to activate the high-speed response mode when the target input velocity is higher than a preset high-speed threshold, increasing the allowable dynamic velocity upper limit and acceleration upper limit for the current cycle, and restoring them to the basic parameters after the target input velocity decreases. The speed planning module adaptively adjusts the feedforward coefficient and the lag catch-up coefficient based on the changing trend of the remaining path length. When the remaining path length continues to increase, the lag catch-up coefficient is increased to enhance the catching-up capability; when the remaining path length continues to decrease, the lag catch-up coefficient is decreased to avoid overshoot. The buffer length M of the window overlay smoothing module is configured according to the system control cycle, the desired smoothness, and the allowable response latency. Each module is deployed in the same process, multiple processes, embedded controller, robot host computer, VR host, simulator, or edge computing device. The modules are connected to each other through function calls, shared memory, message queues, ROS communication, DDS communication, TCP communication, UDP communication, serial communication, or CAN communication. In this embodiment, when the present invention is used in the robot joint space, x can be... k tar Replace with joint target vector q k tar , will x k cmd Replace with joint command vector q k cmd At this point, the path distance can be defined by the weighted joint space norm:

[0052] When this invention is used in the pose space of a robot's end effector, the state can be represented as:

[0053] Where p k R is the end position. k For the end-effector attitude, the position part can directly use a path queue; the attitude part can use quaternion spherical interpolation, rotation vector interpolation, Lie algebra mapping, or other attitude interpolation methods synchronized with the arc length parameter. The distance between end-effector states can be defined as:

[0054] Where Θ(·) represents the relative rotation angle, λ RAs attitude weights, this can be extended from two-dimensional mouse trajectories to three-dimensional position, six-DOF pose, and multi-joint robot control; For the end-effector pose space, the path segment direction is determined by a weighted direction consisting of the position increment and the rotation vector increment of the relative rotation. The angle between adjacent weighted directions is used as the pose path rotation angle. The attitude interpolation is synchronized with the path arc length parameter and is constrained by angular velocity, angular acceleration and angular jerk respectively.

[0055] Explanation of the parameters in the above formula: q a ,q b Two joint vectors, such as the target joint vector and the current joint command vector; W q : Weighted norm symbol; W q Joint space weight matrix (diagonal matrix), with the weight coefficients of each joint arranged on the diagonal; (⋅) T Matrix transpose operation; y k : The overall representation of the state at the end of the k-th control cycle; p k : The position vector at the end of the k-th control cycle; R k : The attitude matrix at the end of the k-th control cycle; d(y a ,y b ): The combined distance between two terminal states; p a ,p b : The position vectors of the two terminal states; R a ,R b : The attitude matrix of the two end states; λ R : Attitude weighting coefficient, used to balance the contributions of position distance and attitude distance; Θ(⋅): Relative rotation angle function, representing the rotation angle between two rotation matrices.

[0056] Figure 1The top-down flow structure in the diagram illustrates the complete processing chain from real-time target input to robot or virtual object command output, including core components such as target state estimation, local path queue construction, path geometry analysis, scalar velocity planning, velocity constraint update, and path integral output. Specifically, the target state estimation module calculates the target velocity and stationary candidate information based on the real-time target input; the local path queue construction module filters the target point flow and maintains the path by combining sampling interval thresholds and collinearity thresholds; the path geometry analysis module outputs the remaining path length, current turning angle, and distance to the turning point; the scalar velocity planning module generates a velocity target by integrating feedforward velocity, lag catch-up, turning speed limit, and terminal braking, and is supplemented by a parameter adaptation and safety constraint module that dynamically adjusts control parameters based on limits, errors, and communication status; finally, the robot or virtual object motion command is output through velocity constraint update and path integral. Figure 2 The diagram illustrates the process of forming candidate path points from the real-time target point stream after filtering by a sampling interval threshold. The system then calculates the angle between the candidate point and the end direction of an existing path and compares it with a collinearity threshold. When the angle is less than the collinearity threshold, it indicates that the candidate point is approximately aligned with the end direction of an existing path. In this case, the system updates the end point of the path queue with the current target point to extend the last path segment and avoid adding redundant nodes. When the angle is not less than the collinearity threshold, it indicates a significant change in path direction. The system appends the candidate point as a new path point to the end of the queue to preserve the corner shape, thus retaining the true corner semantics while filtering high-frequency noise and overly dense points. Figure 3 The diagram illustrates two branching scenarios where the system calculates the arc length increment for the current cycle based on the scalar path velocity and consumes data starting from the first segment of the path queue. When the arc length increment is less than the remaining distance from the output state to the first path point, the output state advances the corresponding arc length along the current segment direction. When the arc length increment is greater than the remaining distance, the output state reaches the first path point, the system pops the path point, and continues to advance along the next path segment using the remaining arc length. This achieves the effect of advancing the output state across points and always remaining on the local target path. Figure 4 The system calculates the permissible angular velocity based on the current turning angle and the braking distance based on the current scalar path speed and deceleration capability. It then superimposes the braking distance with a preset buffer distance to form an early braking window. When the distance from the output state to the current turning point enters this braking window, the system activates corner speed limiting to reduce the speed limit to the permissible angular velocity, allowing the output state to decelerate before reaching the turning point. When the output state is still far from the turning point, speed limiting is not triggered to maintain responsiveness. Figure 5The diagram illustrates the system's mechanism for determining true stillness by having the target step size less than a stillness threshold for multiple consecutive frames, thus avoiding misjudgment in a single frame. After determining true stillness, the system calculates the terminal's braking speed based on the remaining path length and deceleration capability, constraining the scalar path speed below this speed to ensure smooth deceleration of the output state within the remaining path. When the remaining path length further falls below a preset terminal threshold, Figure 5 The hollow dots in the diagram only represent candidate positions x_cand(k+1) when entering the adsorption threshold. The system sets the actual output state to the path terminal p. end And then the speed is reset to zero to achieve precise braking; Figure 6 The diagram shows the processing flow of the path arc length increment or speed increment generated by the speed state layer being weighted by the window weight generation module and accumulated into multiple buffer slots; each control cycle takes the increment from the first buffer slot as the output for this week, then the entire buffer queue is shifted to the left and the tail slot is cleared, so that the new motion is gradually released in a gradual manner within the time window to reduce speed spikes, while ensuring that the output still advances along the local target path; Figure 7 The diagram illustrates the system architecture where operator input is routed through a path-keeping motion overlay to two parallel processing channels: the joint space link and the end-effector space link. The joint space link outputs joint commands to the servo controller via a weighted joint space path queue and scalar velocity planning. The end-effector space link drives the actuator through a position path queue and attitude interpolation, and then drives the actuator after inverse kinematics solution. The safety constraint and parameter adaptation module continuously monitors joint limits, end-effector errors, inverse kinematics status, and communication status to achieve full-condition safety protection.

[0057] A remotely operated path-holding speed planning and control storage medium, on which a computer program is stored, which implements the above method when executed by a processor. In this embodiment, the storage medium includes any one of ROM, RAM, magnetic disk, optical disk, USB flash drive, or solid-state drive; when the computer program is executed by the processor, the real-time target input is constructed into a local target path, so that the output state advances along the local target path, and smoothness is generated by the acceleration, deceleration, corner speed limit, lag catch-up, and terminal braking of the scalar path speed; when the computer program is executed by the processor, the static threshold and the continuous static frame number threshold are also adaptively adjusted according to the noise characteristics of the input device, and the speed limit, acceleration limit, and deceleration capability parameters are adaptively adjusted according to the robot model and task mode.

[0058] In this embodiment, each module is executed by a processor and can achieve closed-loop operation through sensor feedback and a control bus.

[0059] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A teleoperation path-keeping velocity planning control method, characterized by, When applied to robot control, the specific steps include: S11: Obtain the real-time target input of the current control cycle. The real-time target input includes a position vector, a joint vector, or an end pose composed of position and attitude, and uses Euclidean distance, weighted Euclidean distance, or a weighted sum of position distance and relative rotation angle as the path distance, respectively. S12: Based on the real-time target input, a local target path is formed. When the distance between the current target input and the end point of the path queue or the current output point exceeds the preset sampling interval threshold, the current target input is added to the path queue as a candidate path point, and adjacent path segments that are approximately collinear are simplified and merged. S13: Calculate the remaining path length from the current output state along the local target path to the end of the path, and the turning angle between the current path segment and the next path segment; S14: Generate a scalar path speed based on the target input speed, remaining path length, and turning angle, wherein the scalar path speed is obtained by superimposing the target speed feedforward term and the lag catch-up term, and a speed limit constraint is applied based on the turning angle. S15: Integrate the arc length along the local target path based on the scalar path velocity, and map the integrated output state into robot joint commands, end-effector pose commands, or virtual object motion commands; wherein, the output state is constrained to advance along the local target path, and the smoothness is generated by the velocity planning of the scalar path velocity.

2. A teleoperation path-keeping velocity planning control method according to claim 1, characterized by, In S12, forming a local target path based on the real-time target input includes: When the path queue is empty, the current output state is used as the distance reference; when the path queue is not empty, the end point of the path queue is used as the distance reference; when the distance between the current target input and the corresponding distance reference exceeds the preset sampling interval threshold, the current target input is used as a candidate path point; when the path queue is empty, the candidate path point is added to the path queue as the first path point; when there is only one path point in the path queue, the candidate path point is appended as the second path point. When the path queue contains at least two path points, if the angle between the candidate path point and the adjacent path segment formed by the end of the existing path is less than a preset collinearity threshold, the end point of the path queue is updated with the candidate path point to extend the last path segment; otherwise, the candidate path point is added to the path queue to preserve the corner shape. When the length of the path queue exceeds the maximum length limit, the first path point that has been crossed by the current output state is deleted. If the length still exceeds the maximum length limit after deletion, unconsumed path points are simplified within the preset path error range, and the path end point and path points greater than the corner retention threshold are preserved. If the limit still cannot be met, the queue is entered into an overload state, the reception of new path points is suspended, and the output state is controlled to safely decelerate.

3. The remotely operated path-keeping speed planning and control method as described in claim 2, characterized in that, In S14, a scalar path velocity is generated based on the target input velocity, remaining path length, and turning angle, including: The target speed feedforward term is obtained by multiplying the target input speed by a preset feedforward coefficient, and the lag catch-up term is obtained by multiplying the remaining path length by a preset lag catch-up coefficient. The lag catch-up term is set with an upper limit truncation to limit the contribution of the maximum lag catch-up speed. The target speed feedforward term and the lag catch-up term are superimposed to generate a following speed reference. When the target input speed is higher than the preset high speed threshold, the high speed response mode is activated, increasing the upper limit of dynamic speed and acceleration allowed in this cycle.

4. The teleoperated path-keeping speed planning and control method as described in claim 3, characterized in that, Speed ​​limits are imposed based on the turning angle, specifically including: The turning speed limit is generated based on the current turning angle; for non-collinear turning angles retained in the path queue, when the directions of adjacent path segments are not continuous, the allowable speed at the turning point is set to zero, the output state reaches the turning point and completes the speed zeroing before accelerating again along the next path segment; for cases using curvature continuous transition segments, the non-zero turning speed limit is determined based on the turning angle, the curvature of the transition segment and the maximum normal acceleration. The braking distance is calculated based on the current scalar path speed and the system deceleration capability. When the distance from the output state to the current corner point is less than or equal to the sum of the braking distance and the preset buffer distance, corner speed limiting is enabled. Corner speed limiting is not triggered when the corner angle is less than the preset corner dead zone threshold.

5. The remotely operated path-keeping speed planning and control method as described in claim 4, characterized in that, The generation of scalar path velocity also includes real-time stationary determination and terminal braking, including: The target step size is set as the distance between the target input of the current control cycle and the target input of the previous control cycle. When the target step size is less than the preset static threshold for multiple consecutive frames, the target input is determined to be truly static. After determining that the target is truly stationary, the terminal's braking speed is calculated based on the remaining path length and the system's deceleration capability, and the scalar path speed is constrained to not exceed the terminal's braking speed; when the remaining path length is less than the preset terminal threshold, the output state is snapped to the path terminal and the speed is cleared to zero.

6. The teleoperated path-holding speed planning and control method as described in claim 5, characterized in that, After generating the scalar path velocity, the process also includes updating the motion constraints on the scalar path velocity, including: When the scalar path velocity before the update is less than the scalar path velocity target, the scalar path velocity before the update is added to the product of the system acceleration capability and the control cycle duration to obtain the updated velocity, and the velocity target is used as the upper limit for truncation; when the scalar path velocity before the update is greater than or equal to the velocity target, the product of the system deceleration capability and the control cycle duration is subtracted, and the velocity target is used as the lower limit for truncation. Based on the acceleration and deceleration limited update, an acceleration constraint is introduced and acceleration write-back is performed. The difference between the acceleration of the current cycle and the acceleration of the previous cycle is calculated. When the absolute value of the acceleration difference is greater than the product of the preset acceleration upper limit and the control cycle duration, the change of the acceleration of the current cycle relative to the acceleration of the previous cycle is truncated to the range of the product, and the speed is updated with the truncated acceleration of the current cycle.

7. The remotely operated path-keeping speed planning and control method as described in claim 6, characterized in that, Before the path integral in S15, a distributed smooth release of the path arc length increment is also included using a window overlay smoothing layer, specifically including: The cache length M is set to control the time window span for speed smoothing. A Hann window function, a sine window function, or a binomial window function is used to assign weights to each cache slot. Each weight satisfies the condition that the sum of the weights is 1 and the weights of at least two cache slots are greater than 0. The path arc length increment generated in this period is multiplied by the weights of each cache slot and then accumulated to the corresponding cache slot. The output arc length increment of the current control cycle is the current value of the first buffer slot in the buffer queue. After being retrieved, the buffer queue is shifted one position to the left and the last buffer slot is cleared to zero, so that the newly added arc length increment is gradually released in subsequent cycles to reduce speed spikes. When terminal snapping, emergency stop, communication interruption, safety pause or path reset occurs, the buffer vector is cleared, the output arc length increment of the current cycle is set to zero, and the scalar velocity state and acceleration state are reset.

8. A remotely operated path-keeping speed planning and control system, characterized in that, The system, configured to perform the method of any one of claims 1-7, comprises: The target input module is used to receive real-time target position, target pose, joint target, end-effector pose or multi-source fusion target, and provides timestamp and status marker; The path queue module is used to maintain local target paths based on sampling interval thresholds, collinearity thresholds, and maximum queue length, and to perform lightweight simplification on adjacent path segments that are approximately collinear. The input state estimation module is used to calculate the target velocity, target increment, number of consecutive static frames, and whether the target is truly stationary. The path geometry analysis module is used to calculate the remaining path length, current turning angle, distance to the turning point, path segment direction, and path terminal status. The scalar velocity planning module is used to generate scalar path velocities based on target velocity feedforward, lag catch-up, dynamic speed limit, corner speed limit, and terminal stop speed, and to update the scalar path velocities by performing at least one of the following motion constraints: acceleration limit, deceleration limit, acceleration limit, deceleration limit, or jerk limit. The path integration module is used to integrate the scalar path velocity along the local target path, traversing the reached path points and outputting the current position or status. The window overlay smoothing module is used to perform distributed overlay smoothing of velocity increments or path arc length increments using sine window functions, Hann window functions, binomial window functions, or triangular window functions. The robot mapping module is used to map the position or state output by the path integration module into robot joint commands, end-effector pose commands, velocity commands, servo commands, or simulation object commands. The parameter adaptive module is used to dynamically adjust the parameters of each module based on the following error, corner radius, task mode, robot limit, or operator speed.

9. A remotely operated path-keeping speed planning and control system as described in claim 8, characterized in that, The scalar velocity planning module includes: The dynamic speed limit unit is used to dynamically adjust the allowable speed limit for the current cycle based on the target input speed, the system's maximum safe speed, robot joint limits, or the task mode. The motion constraint unit is used to update the scalar path velocity by performing at least one of the following motion constraints: acceleration constraint, deceleration constraint, acceleration constraint, deceleration constraint, or jerk constraint, to ensure that the output satisfies the kinematic constraints. The high-speed response unit is used to activate the high-speed response mode when the target input speed is higher than the preset high-speed threshold, increase the upper limit of dynamic speed and acceleration allowed in this cycle, and restore to the basic parameters after the target input speed decreases; The scalar velocity planning module adaptively adjusts the feedforward coefficient and the lag catch-up coefficient according to the changing trend of the remaining path length: when the remaining path length continues to increase, the lag catch-up coefficient is increased to enhance the catching-up capability; when the remaining path length continues to decrease, the lag catch-up coefficient is decreased to avoid overshoot.

10. A remotely operated path-holding speed planning and control storage medium, wherein a computer program is stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.