A robot trajectory planning method, device and medium based on multi-objective optimization

Abstract

The application discloses a kind of robot trajectory planning method, equipment and medium based on multi-objective optimization, it is related to process control technical field, comprising: according to risk caliber package, risk contraction is executed to robot passable feasible region, obtain passable boundary segment, and carry out circumnavigation topological marking merger and corridor boundary cutting solidification, generate topological corridor candidate set;In topological corridor candidate set Distribution candidate trajectory, and carry out constraint check and corridor projection recovery, obtain feasible candidate trajectory set, while carrying out non-dominated reservation and cross-corridor convergence, generate cross-corridor Pareto candidate archive;Interlock effective time window boundary and risk door control threshold consistency verification are carried out to cross-corridor Pareto candidate archive, and selectable candidate is selected, while solidifying control cycle instruction, generates trajectory issuing instruction set.The application realizes multi-objective solution set rules screening, ensures that instruction meets process control timing constraint, guarantees issuing reliability.

Description

Technical Field

[0001] This invention relates to the field of process control technology, and in particular to a robot trajectory planning method, device and medium based on multi-objective optimization. Background Technology

[0002] In the fields of industrial automation and intelligent manufacturing, robot trajectory planning, as a core component of process control systems, directly impacts production line efficiency, operational accuracy, and system safety. With the deepening implementation of Industry 4.0, modern manufacturing environments are becoming increasingly complex, leading to greater demands for multi-robot collaboration, human-robot integration, and operations in dynamic and uncertain environments. Traditional trajectory planning techniques primarily rely on geometric path search and kinematic constraint solving, aiming to find a collision-free path from the starting point to the target point, and then parameterizing it over time to meet dynamic constraints such as velocity and acceleration. High-speed planning algorithms based on sampling and optimization-based numerical methods have been widely applied, demonstrating strong capabilities in handling high-dimensional state spaces and nonlinear constraints.

[0003] While existing technologies perform well in static environments, they still fall short in highly dynamic industrial process control scenarios. First, current methods often separate perception, evaluation, and planning, lacking deep spatiotemporal alignment of multi-source heterogeneous data such as workstation interlocks, dynamic obstacle prediction, and robot states. This results in delayed responses to sudden risks, making it difficult to guarantee the real-time performance and robustness of process control. Second, traditional multi-objective optimization algorithms often ignore the dynamic shrinkage characteristics of safety boundaries and topological connectivity constraints, easily generating redundant or infeasible trajectories. They lack effective mechanisms to ensure that trajectories simultaneously satisfy optimization objectives and safety gating thresholds and interlocking time windows in process control, thus weakening the reliability of command issuance. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a robot trajectory planning method based on multi-objective optimization to solve the problems of dynamic response lag and unreliable command issuance.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a robot trajectory planning method based on multi-objective optimization, comprising: collecting robot state, workstation interlocks, and environmental observations, and performing timestamp alignment and short-term prediction entry registration to generate a scene alignment collection table; inputting the scene alignment collection table into a risk scale model, dividing effective safety boundaries, and extracting risk scale entries to generate a risk calibrator package; performing risk contraction on the robot's traversable feasible domain according to the risk calibrator package, obtaining traversable boundary segments, and performing detour topology marking and merging with corridor boundary trimming and solidification to generate a topology corridor candidate set; allocating candidate trajectories in the topology corridor candidate set, and performing constraint verification and corridor projection recovery to obtain a feasible candidate trajectory set, while performing non-dominated retention and cross-corridor convergence to generate a cross-corridor Pareto candidate file; performing interlock effective time window boundary and risk gating threshold consistency verification on the cross-corridor Pareto candidate file, selecting deployable candidates, and solidifying control cycle instructions to generate a trajectory deployment instruction set.

[0008] As a preferred embodiment of the robot trajectory planning method based on multi-objective optimization described in this invention, the steps for collecting robot state, workstation interlocks, and environmental observations are as follows:

[0009] The robot's pose, joint angle, velocity, acceleration, and drive margin are read according to the process control sampling caliber, and the acquisition time stamp and control cycle number are fixed to generate robot state entries.

[0010] Read the interlock input signal, interlock output enable signal and interlock reason code from the process control interlock interface, and perform debouncing and solidify the interlock effective time window to generate workstation interlock entries;

[0011] Static obstacle boundaries and dynamic obstacle observation sequences are extracted from environmental sensors, and boundary closure verification, duplicate observation merging, and missing measurement slot filling are performed to generate environmental observation entries.

[0012] As a preferred embodiment of the robot trajectory planning method based on multi-objective optimization described in this invention, the steps for generating the scene alignment acquisition table are as follows:

[0013] Establish a unified clock grid for process control, and map robot status entries, workstation interlock entries, and environmental observation entries to the same control cycle number to generate short-term prediction entries for robot traversable domain boundaries and dynamic obstacles.

[0014] Assign a prediction effective time window to the short-term prediction items of dynamic obstacles, obtain the set of aligned prediction items, write it into the row-level alignment key, and generate a scene alignment acquisition table.

[0015] As a preferred embodiment of the robot trajectory planning method based on multi-objective optimization described in this invention, the steps for generating the risk calibrator package are as follows:

[0016] The risk scaling model includes an input embedding layer, an uncertainty mapping layer, a risk aggregation layer, and a risk gating layer;

[0017] In the input embedding layer, the robot status entries, workstation interlock entries, environmental observation entries, and dynamic obstacle short-term prediction entries in the scene alignment acquisition table are rearranged and encoded into a unified channel sequence.

[0018] In the uncertainty mapping layer, the observation uncertainty evidence field is extracted from the unified channel sequence and aligned and spliced ​​with the workstation interlock reason code to output the risk scale candidate entries;

[0019] In the risk aggregation layer, continuous segments of the risk scale candidate entries are merged into the risk scale aggregation, the effective safety boundary is defined, and the effective safety boundary entry set is output.

[0020] At the risk gating layer, risk scale entries and risk gating threshold entries are extracted from the set of effective security boundary entries and written into the risk scale level, and the sequence of high-risk boundaries and risk scale entries is output.

[0021] The effective security boundary entry set, high-risk boundary, risk scale entry sequence, and interlock effective time window boundary are encapsulated, and the correspondence between the boundary number and the corridor clipping index is registered as the mapping entry field to generate the risk caliber package.

[0022] As a preferred embodiment of the robot trajectory planning method based on multi-objective optimization described in this invention, the steps for generating the candidate set of topological corridors are as follows:

[0023] Based on the risk caliber package, the robot's traversable feasible domain boundary is effectively corrected within the interlock effective time window, and inward risk contraction and connectivity segmentation are performed on high-risk boundaries to generate a set of traversable boundary fragments.

[0024] In the set of passable boundary segments, the starting and ending areas are solidified, the bottleneck number sequence and detour direction information are registered, and combined into detour topology markers to generate a set of corridor skeleton segments;

[0025] The outer envelope boundary is expanded along the corridor skeleton fragment set, and clipping, tightening and continuous splicing are performed. At the same time, the key point sequence of the corridor boundary and the effective time window of the corridor are solidified to generate a candidate set of topological corridors.

[0026] As a preferred embodiment of the robot trajectory planning method based on multi-objective optimization described in this invention, the steps for generating cross-corridor Pareto candidate profiles are as follows:

[0027] Within the effective time window of the corridor in the candidate set of the topological corridor, candidate trajectory skeletons are allocated, candidate trajectory control point sequences are obtained, and discretized into sampling point sequences to generate a set of candidate trajectory entries in the corridor.

[0028] Verify the inclusion of the interlock effective time window boundary, robot state item boundary and corridor boundary in the candidate trajectory item set within the corridor, and perform corridor backtracking and time scale stretching write-back to generate a feasible candidate trajectory set;

[0029] Extract a set of multi-objective entries from the feasible candidate trajectory set, and perform non-dominated preservation and cross-corridor diversity maintenance convergence to generate a cross-corridor Pareto candidate profile.

[0030] As a preferred embodiment of the robot trajectory planning method based on multi-objective optimization described in this invention, the steps for verifying the consistency between the interlocking effective time window boundary and the risk gating threshold of the cross-corridor Pareto candidate files are as follows:

[0031] Based on the cross-corridor Pareto candidate archive, the interlock effective time window boundary is checked, continuous violation segments are merged, and interlock violation candidate trajectories are eliminated to generate a qualified interlock candidate set.

[0032] Based on the interlocked qualified candidate set, the risk scale entries and risk gate threshold entries are checked for threshold hits, and risk violation segments are merged. At the same time, high-risk candidate trajectories are removed to generate a risk qualified candidate set.

[0033] As a preferred embodiment of the robot trajectory planning method based on multi-objective optimization described in this invention, the steps for generating the trajectory issuance instruction set are as follows:

[0034] Extract process control cycle items and risk scale level summary items from the risk qualified candidate set, perform priority sorting and secondary boundary consistency check, and generate a candidate list for issuance;

[0035] Expand the sampling point sequence corresponding to the candidate list and write the target pose and target joint angle instruction fields and the effective boundary field to generate the trajectory delivery instruction set.

[0036] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the robot trajectory planning method based on multi-objective optimization as described in the first aspect of the present invention.

[0037] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the robot trajectory planning method based on multi-objective optimization as described in the first aspect of the present invention.

[0038] The beneficial effects of this invention are as follows: by aligning multi-source data in time and space and modeling risk scales, dynamic safety boundary quantification is achieved, and process control interlocking logic is integrated to improve planning robustness; by cross-corridor Pareto convergence and consistency verification, multi-objective solution set compliance screening is achieved to ensure that instructions meet process control timing constraints and guarantee the reliability of issuance. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of a robot trajectory planning method based on multi-objective optimization.

[0041] Figure 2 A time-series separation comparison diagram of risk scale in the neighborhood of risk gating threshold.

[0042] Figure 3 Spatial distribution map of Pareto candidate archives across corridors.

[0043] Figure 4 A funnel diagram for step-by-step screening of candidate trajectory risks and interlocking dual gating. Detailed Implementation

[0044] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0045] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0046] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0047] Reference Figures 1-4 This is one embodiment of the present invention, which provides a robot trajectory planning method based on multi-objective optimization, including the following steps:

[0048] S1: Collect robot status, workstation interlocks and environmental observations, and perform timestamp alignment and short-term prediction entry registration to generate a scene alignment collection table;

[0049] S1.1: Read pose, joint angle, velocity, acceleration and drive margin according to the process control sampling caliber, and solidify the acquisition time stamp and control cycle number to generate robot state entries;

[0050] Furthermore, according to the process control sampling caliber, the pose, joint angle, velocity, acceleration, and drive margin are read within each control cycle number. The pose and joint angle are checked for dimensionality and out-of-bounds values ​​are clipped and registered. The velocity and acceleration are checked for arrival order and rearranged according to out-of-order conditions, and the sequence mark before the rearrangement is registered. The drive margin is checked for non-negativity and abnormal entries are marked with quality. The acquisition time mark is fixed for each reading record and bound to the control cycle number. The records that have passed the check within the same control cycle number are encapsulated into robot status entries by field alignment.

[0051] S1.2: Read the interlock input signal, interlock output enable signal and interlock reason code from the process control interlock interface, and perform debouncing and solidify the interlock effective time window to generate workstation interlock entries;

[0052] Furthermore, the process control interlock interface samples the interlock input signal, interlock output enable signal, and interlock cause code according to the control cycle number. The sampling record is written to the acquisition time stamp and bound to the control cycle number. When the interlock input signal and interlock output enable signal exhibit short pulse flips, the flipped segment is backfilled according to the stable value of the adjacent control cycle number and a backfill mark is registered. Repeated flipped segments within the same control cycle number are merged according to their time proximity and one valid switch is retained. The interlock cause code is verified one by one according to the valid code table. Missing interlock cause codes are written to the missing test mark and the summary of the most recent valid interlock cause code is recorded. The enable and disable switching positions of the interlock output enable signal are located by the status change of the adjacent control cycle number. The control cycle number corresponding to the switching position is registered as the start or end point of the interlock effective time window. Interlock effective time windows with the same interlock cause code and adjacent time windows are spliced ​​and merged and the window number is registered. The interlock input signal, interlock output enable signal, interlock cause code, and interlock effective time window are encapsulated to generate a workstation interlock entry.

[0053] S1.3: Extract static obstacle boundaries and dynamic obstacle observation sequences from environmental sensors, and perform boundary closure verification, duplicate observation merging, and missing measurement gap filling to generate environmental observation entries;

[0054] Furthermore, static obstacle boundaries and dynamic obstacle observation sequences are extracted from environmental sensors according to control cycle numbers. The acquisition time stamp and control cycle number are fixed for each observation. Boundary closure verification is performed on static obstacle boundaries. The boundary closure verification checks the consistency of the distance between the first and last points and the boundary direction, and performs closure supplementation registration for unclosed boundaries. Repeated observations are merged on dynamic obstacle observation sequences. Repeated observations are merged according to target identification and spatial proximity conditions, and entries with higher confidence are retained. Missing measurement placeholders are filled on dynamic obstacle observation sequences. Missing measurement placeholders are written to the missing positions of continuous control cycle numbers and missing measurement span markers are registered. The static obstacle boundaries and dynamic obstacle observation sequences, together with closure verification markers, merging markers and missing measurement markers, are encapsulated into environmental observation entries.

[0055] S1.4: Establish a unified clock grid for process control, and map robot status entries, workstation interlock entries, and environmental observation entries to the same control cycle number to generate short-term prediction entries for robot traversable feasible domain boundaries and dynamic obstacles.

[0056] Furthermore, a unified clock grid for process control is established, and the grid step size and control cycle number increment rules are solidified. The acquisition time stamps of robot status entries, workstation interlock entries, and environmental observation entries are mapped to the corresponding control cycle numbers, and multiple records within the same control cycle number are merged and bound to the sequence number based on their adjacent execution times. According to the static obstacle boundary and the occupied boundary of the dynamic obstacle observation sequence in the environmental observation entries, the outer boundary of the passable area is generated according to the control cycle number and the boundary number is registered and encapsulated as the robot's passable feasible domain boundary. For the dynamic obstacle observation sequence, the displacement change of adjacent control cycle numbers is extracted according to the target identifier and the prediction step number is extrapolated. The prediction occupied boundary and control cycle number are written for each prediction step to generate a short-term prediction entry for dynamic obstacles.

[0057] S1.5: Assign a prediction effective time window to the short-term prediction items of dynamic obstacles, obtain the set of aligned prediction items, write it into the row-level alignment key, and generate a scene alignment acquisition table.

[0058] Furthermore, for short-term prediction entries of dynamic obstacles, a prediction effective time window is assigned according to the prediction step sequence number, and the start and end points of the prediction effective time window are written into the entry time window field. The start point of the prediction effective time window is taken from the grid start time marker of the corresponding control cycle sequence number, and the end point of the prediction effective time window is taken from the grid end time marker covered by the prediction step sequence number. The short-term prediction entries of dynamic obstacles are aligned with the robot's feasible domain boundary according to the control cycle sequence number and merged into an aligned prediction entry set. The aligned prediction entry set performs target identification sorting on the multi-target prediction occupied boundary within the same control cycle sequence number and registers the sorting index. A row-level alignment key is generated for each control cycle sequence number, and the row-level alignment key is bound to the robot status entry, workstation interlock entry, environmental observation entry, robot feasible domain boundary, and aligned prediction entry set. The row-level alignment key is written into the control cycle sequence number and the acquisition time marker summary, and encapsulated to generate a scene alignment acquisition table.

[0059] S2: Input the scene alignment acquisition table into the risk scale model, divide the effective safety boundary, extract the risk scale entries, and generate a risk caliber package;

[0060] S2.1: The risk scaling model includes an input embedding layer, an uncertainty mapping layer, a risk aggregation layer, and a risk gating layer;

[0061] Furthermore, the input embedding layer expands the scene-aligned acquisition table into risk scale candidate entries according to the process control unified clock grid, performs channel rearrangement, time stamp alignment, juxtaposition of missing and quality evidence fields, and window position encoding, and outputs a unified channel sequence; the uncertainty mapping layer extracts the observation uncertainty evidence field from the unified channel sequence, completes scale normalization, evidence aggregation summary and confidence weight writing according to the mapping entry field, and outputs an uncertainty scale entry set; the risk aggregation layer performs continuous segment merging, interlock tightening mark consistency check and risk scale aggregation on the uncertainty scale entry set according to the boundary number index, and outputs a valid safety boundary entry set; the risk gating layer performs threshold hit determination, gating weight modulation and hierarchical mark registration on the valid safety boundary entry set based on the risk gating threshold entries, forming a risk scale model including the input embedding layer, uncertainty mapping layer, risk aggregation layer and risk gating layer.

[0062] S2.2: In the input embedding layer, the robot state entries, workstation interlock entries, environmental observation entries, and dynamic obstacle short-term prediction entries in the scene alignment acquisition table are rearranged and encoded into a unified channel sequence;

[0063] Furthermore, the scene alignment acquisition table is written into the input embedding layer of the risk scale model, and the control cycle number is expanded according to the row-level alignment key order. The robot status entry, workstation interlock entry, environmental observation entry, and dynamic obstacle short-term prediction entry are extracted row by row, and a channel type label and boundary number index are bound to each record. The dynamic obstacle short-term prediction entries are arranged in ascending order of prediction step number. Within the same prediction step number, they are arranged in a fixed order according to the target identifier and the sorting index is registered to avoid output order drift within the same cycle. The robot status entry and workstation interlock entry are written into the control cycle number position encoding field, the environmental observation entry is written into the boundary closure check mark and the missing measurement occupancy mark, and the dynamic obstacle short-term prediction entry is written into the prediction effective time window field. Within the same control cycle number, they are spliced ​​into a channel segment according to the order of the channel type label. The continuity of the channel segments across control cycle numbers is checked by incrementing the control cycle number. For missing control cycle numbers, a occupancy segment is inserted and a missing measurement mark is registered. The result is a unified channel sequence.

[0064] S2.3: In the uncertainty mapping layer, extract the observation uncertainty evidence field from the unified channel sequence, and align and splice it with the workstation interlock reason code to output the risk scale candidate entries;

[0065] Furthermore, the unified channel sequence is written into the uncertainty mapping layer of the risk scale model and the records are expanded according to the control cycle number and boundary number index. The environmental observation items and dynamic obstacle short-term prediction items are located, and the observation uncertainty evidence field is extracted. When the observation uncertainty evidence field is missing, a missing measurement mark is written. When there is an abnormal jump in the observation uncertainty evidence field, a quality mark is written. When the boundary number index is repeated for the same control cycle number, the repeated records are sorted according to the quality mark and the record with the higher quality mark is retained. The remaining records are written with a merge mark. The workstation interlocking reason code in the workstation interlocking item is aligned to the boundary number index record with the same control cycle number. When multiple workstation interlocking reason codes exist at the same time, they are selected according to the preset priority order and written into the reason code summary. The observation uncertainty evidence field, reason code summary, control cycle number and boundary number index are encapsulated into a risk scale candidate item and concatenated in the order of control cycle number to output the risk scale candidate item.

[0066] It should be noted that the workstation interlock cause code is used to identify the specific interlock cause category that triggers the switching between the interlock input signal and the interlock output allow signal, and serves as the basis for generating the interlock tightening mark and risk gate threshold entries in the risk scale candidate entries.

[0067] S2.4: In the risk aggregation layer, continuous segments of the candidate entries for the risk scale are merged and the risk scale is aggregated to divide the effective safety boundary and output the set of effective safety boundary entries.

[0068] Furthermore, candidate entries for the risk scale are written into the risk aggregation layer of the risk scale model and recorded by control cycle number. When the same boundary number index appears consecutively in adjacent control cycle numbers, they are merged into continuous segments, and the control cycle number at the start and end of the segment is recorded. The observation uncertainty evidence fields within the continuous segments are aggregated into a single scale by boundary number index, and the scale value corresponding to the uncertainty summary field is calculated. The scale value corresponding to the uncertainty summary field is written into the uncertainty summary field. The workstation interlock reason code summary within the continuous segments is checked for consistency according to priority, and an interlock tightening mark is written when tightening conditions occur. Based on the uncertainty summary field and the interlock tightening mark, effective safety boundary correction entries are generated and bound to the boundary number index. The effective time window field is fixed with the time stamp corresponding to the control cycle number at the start and end of the segment. The effective safety boundary correction entries are merged and encapsulated into a set of effective safety boundary entries by boundary number index.

[0069] The formula for calculating the scale value corresponding to the uncertainty summary field is:

[0070] ;

[0071] in, Indicates the control cycle number With boundary number index The corresponding uncertainty summary field corresponds to the scale value. This indicates a trimming operator that restricts the calculated value within the parentheses to the range of 0 to 100. This represents a gating function that maps the input to the interval between 0 and 1 and keeps it monotonic. The field indicating evidence of observational uncertainty in the control period number is Boundary number index is The first Each component scale value, Indicates the first Sensitivity coefficients of each component The coupling coefficient represents the interlock severity scale. This indicates that the workstation interlock reason code is in the control cycle sequence number. The interlock severity scale value obtained from the mapping.

[0072] It should be noted that the sensitivity coefficient is obtained by offline calibration based on the correspondence between the changes in the uncertainty evidence field components observed in historical operating conditions and the frequency of hits of the risk scale level (example range: 0.5–2.0).

[0073] Coupling coefficient of interlock severity scale: After mapping the workstation interlock cause code hierarchy table to the interlock severity scale, the tightening range of risk gate threshold items is obtained by offline calibration according to the effective period of the interlock tightening mark (example range: 0–1.5).

[0074] S2.5: In the risk gating layer, extract risk scale entries and risk gating threshold entries from the set of effective safety boundary entries, write them into the risk scale level, and output the high-risk boundary and risk scale entry sequence.

[0075] Furthermore, the effective safety boundary entry set is written into the risk gating layer of the risk scale model and expanded according to the boundary number index. For each effective safety boundary correction entry, the uncertainty summary field, interlock tightening flag, and effective time window field are read to generate a risk gating threshold entry. The risk gating threshold entry is written into the boundary number index, threshold caliber flag, and effective time window boundary. For the effective safety boundary correction entries, gating hit verification is performed according to the risk gating threshold entry, and the hit record is written into the risk scale entry. The risk scale entry is written into the control cycle number range, boundary number index, and hit flag. At the same time, the risk scale level is written according to the hit level (e.g., low, medium, relatively high, high). Boundary number indices that reach the high risk scale level are merged into high-risk boundaries and a high-risk boundary number list is registered. The risk scale entries are concatenated into a risk scale entry sequence according to the control cycle number and output together with the risk gating threshold entry.

[0076] It should be noted that the following are examples of rules for writing risk scale levels: When the scale value corresponding to the uncertainty summary field falls between 0 and 20, the risk scale level is written as low; when the scale value corresponding to the uncertainty summary field falls between 20 and 40, the risk scale level is written as medium; when the scale value corresponding to the uncertainty summary field falls between 40 and 60, the risk scale level is written as relatively high; when the scale value corresponding to the uncertainty summary field falls between 60 and 80, the risk scale level is written as high; when the scale value corresponding to the uncertainty summary field reaches 80 or above, the risk scale level is written as high and a high-risk boundary marker is added.

[0077] The risk gating threshold entries are used to index the uncertainty gating levels by boundary number and solidify the uncertainty gating levels by the effective time window boundary. The example value range can be set to 4 threshold levels (e.g., 20, 40, 60, 80, corresponding to low, medium, relatively high, and high), and the scale value corresponding to the uncertainty summary field is mapped to the risk scale level by the threshold level.

[0078] Figure 2The diagram illustrates the temporal changes of risk scale entries output by different scheme groups under the same process control clock grid during representative operations, with the risk gating threshold entries indicated by dashed lines as a reference for judgment. The overview diagram observes the fluctuation patterns of risk scale entries throughout the entire time period, the dense areas of peak occurrences, and the threshold hit trend. The local window enclosed by the red dashed rectangle is used to compare the degree of separation and the continuity of rise / fall near the threshold of multiple curves within a short period, thus visually demonstrating the quantification capability of the risk scale model after spatiotemporal alignment of multi-source data and aggregation of uncertainty evidence fields. The enlarged local diagram further reveals that different schemes exhibit differences in their hit performance on the risk gating threshold within the same window. This can be used to explain how the risk scale entry sequence, when truncated and hierarchically written in the gating layer, supports subsequent high-risk boundary marking and risk caliber encapsulation, thereby providing traceable evidence for risk contraction and corridor pruning.

[0079] Among them, the G0 baseline group: under the same process control unified clock grid, the scene alignment acquisition table is directly mapped to the risk scale entry sequence according to a fixed caliber. The risk scale model is not enabled to form a control by extracting and aggregating the evidence fields of observation uncertainty and aligning and splicing the workstation interlocking cause code.

[0080] G1 Risk Scale Ablation Group: The risk scale model retains the extraction and aggregation of observation uncertainty evidence fields and writes them into the corresponding scale values ​​of the uncertainty summary field to generate risk scale entries, but does not introduce the interlock severity scale coupling of the workstation interlock cause code mapping to modulate the tightening of the gate threshold entries to highlight the difference of "risk scale modeling only".

[0081] G2 Interlock Verification Ablation Group: It retains the workstation interlock cause code summary alignment, interlock tightening mark and risk gating threshold hit verification link to output risk scale entries, but only performs coarse-grained processing on the observation uncertainty evidence field without sufficient aggregation and summarization, so that curve separation is mainly driven by interlock verification and gating tightening.

[0082] The G3 complete solution group simultaneously enables the entire chain of unified channel sequence rearrangement, observation uncertainty evidence field extraction and aggregation, workstation interlocking cause code alignment and splicing, and risk gating layer threshold entry generation / hit writing, thus demonstrating more stable hit continuity and separation within the threshold neighborhood.

[0083] S2.6: Encapsulate the effective security boundary entry set, high-risk boundary, risk scale entry sequence, and interlock effective time window boundary, and register the correspondence between the boundary number and the corridor clipping index as the mapping entry field to generate the risk caliber package.

[0084] Furthermore, the effective safety boundary entry set, high-risk boundary, risk scale entry sequence, and interlock effective time window boundary are checked for consistency according to the boundary number index and control cycle sequence range. Missing boundary number indices are marked as missing and duplicate boundary number indices are deduplicated and merged. The interlock effective time window boundary is written into the effective field and the corrected effective safety boundary entry is written into the boundary caliber field. High-risk boundaries are written into the high-risk caliber field and the risk scale entry sequence is written into the risk caliber field. Based on the positional relationship of the boundary number index in the corridor boundary key point sequence, a corridor clipping index is generated and the correspondence between the boundary number index and the corridor clipping index is registered as the mapping entry field. The reachability of the mapping entry field is checked and an invalid mapping mark is written. The risk caliber package is then encapsulated.

[0085] S3: Based on the risk caliber package, perform risk contraction on the robot's traversable feasible domain, obtain traversable boundary segments, and perform detour topology marking and merging with corridor boundary trimming and solidification to generate a topology corridor candidate set;

[0086] S3.1: Based on the risk caliber package, perform effective safety boundary correction on the robot's traversable feasible domain boundary within the interlock effective time window, and perform inward risk contraction and connected segment segmentation on high-risk boundaries to generate a traversable boundary fragment set;

[0087] Furthermore, based on the interlocking effective time window boundary of the risk caliber package, the control cycle number range is limited, and the robot's traversable feasible domain boundary is captured simultaneously. The boundary number index is used to match the robot's traversable feasible domain boundary with the effective safety boundary entry set one by one. The boundary number index of the effective safety boundary correction entry is written into the correction scale mark, and the corresponding boundary is offset inward according to the correction scale. The boundary number index of the missing boundary is written into the retention mark to keep the original boundary. The boundary number index of the high-risk boundary triggers inward risk contraction and registers the risk scale level summary. The contracted robot traversable feasible domain boundary is checked for boundary closure consistency and self-intersection points and breakpoints are located. The breakpoints are numbered according to the boundary point sequence. Based on the breakpoint sequence number, the boundary is divided into several continuous boundary segments, and each segment is written with a segment number and a boundary number index set. The continuous boundary segments are encapsulated to form a traversable boundary segment set.

[0088] It should be noted that inward risk contraction refers to, at the boundary number index where a high-risk boundary is hit, translating the corresponding set of boundary points along the inner normal of the robot's traversable feasible domain boundary to the interior of the traversable area and writing back the coordinates of the contracted boundary points, while simultaneously registering the contraction source and risk scale level summary.

[0089] S3.2: Solidify the starting and ending areas in the set of passable boundary segments, register the bottleneck number sequence and detour direction information, and combine them into detour topology markers to generate a set of corridor skeleton segments;

[0090] Furthermore, in the set of passable boundary segments, the coordinates of the boundary endpoints are extracted for each segment number, and endpoint attribution is checked according to the starting region constraint and the ending region constraint. The endpoint attribution check writes the starting region mark or the ending region mark for endpoints within the distance threshold, and writes the free mark for the missing endpoints. The starting region mark and the ending region mark are solidified as the starting region and the ending region. Connecting path segments are constructed in the set of passable boundary segments according to the segment number order, and the minimum gap segment is located. The segment number corresponding to the minimum gap segment is registered as the bottleneck number, and a bottleneck number sequence is formed according to the traversal order of the path segments. For each connecting path segment, the detour relative lateral relationship is extracted near the boundary number index corresponding to the static obstacle boundary and the dynamic obstacle observation sequence, and the detour direction information is written. The bottleneck number sequence and the detour direction information are spliced ​​and encapsulated into a detour topology mark. The connecting path segments are merged into segment groups according to the consistency of the detour topology mark, and the segment groups are written into the group number and segment number set to generate the corridor skeleton segment group set.

[0091] S3.3: Expand the outer envelope boundary along the corridor skeleton fragment set, and perform clipping, tightening and continuity splicing. At the same time, solidify the corridor boundary key point sequence and the effective time window of the corridor to generate a candidate set of topological corridors.

[0092] Furthermore, along the corridor skeleton segment set, the segment number set is expanded according to the group number and the boundary point set is extracted. The boundary point set is expanded outward along the normal direction of the boundary point tangent to form the outer envelope boundary and the outward expansion distance mark is registered. The boundary points of the outer envelope boundary are aligned with the boundary number index to map the entry field to locate the corridor clipping index. When the corridor clipping index hits a high-risk boundary, the corresponding boundary segment is clipped inward and written with a tightening mark. When the corridor clipping index does not hit a high-risk boundary, adjacent boundary segments are spliced ​​according to the principle of endpoint proximity and written with the splicing sequence number. After the outer envelope boundary is clipped, tightened and spliced ​​with continuity, the boundary closure consistency check is performed and the inflection point is extracted. The inflection point is solidified into the corridor boundary key point sequence according to the boundary point order and the group number and boundary number index set are registered. The effective time window of the corridor is solidified according to the control cycle sequence range covered by the interlock effective time window boundary and the group number and written into the topology corridor candidate set.

[0093] It should be noted that when the corridor clipping index hits a high-risk boundary, the boundary segment is clipped into the corridor along the inner normal of the corresponding boundary segment of the outer envelope boundary and the clipped boundary point set is written back. At the same time, a tightening mark is written to solidify the tightening position and range.

[0094] S4: Assign candidate trajectories to the topological corridor candidate set, perform constraint verification and corridor projection recovery to obtain a set of feasible candidate trajectories, and simultaneously perform non-dominated retention and cross-corridor convergence to generate a cross-corridor Pareto candidate profile.

[0095] S4.1: Within the effective time window of the corridor in the candidate set of the topological corridor, perform candidate trajectory skeleton allocation, obtain the candidate trajectory control point sequence, and discretize it into a sampling point sequence to generate a candidate trajectory entry set in the corridor;

[0096] Furthermore, within the candidate set of topological corridors, entries covering the control cycle number range are filtered according to the effective time window of the corridor. For each candidate set entry, the key point sequence of the corridor boundary is extracted. Based on the key point sequence of the corridor boundary, multiple candidate trajectory skeletons are generated, and each candidate trajectory skeleton is assigned a candidate trajectory number and a corridor identifier. Control points are laid out along the candidate trajectory skeleton at fixed arc length intervals, and the control points are projected onto the area enclosed by the key point sequence of the corridor boundary to obtain the candidate trajectory control point sequence and register the control point number. The candidate trajectory control point sequence is discretized into a sampling point sequence according to the control cycle number of the unified clock grid of the process control and written into the sampling point number and the control cycle number. At the same time, the sampling point sequence is written into the effective time window boundary of the corridor. The candidate trajectory number, corridor identifier, candidate trajectory control point sequence and sampling point sequence are encapsulated to generate a set of candidate trajectory entries within the corridor.

[0097] S4.2: Verify the inclusion of the interlock effective time window boundary, robot state item boundary and corridor boundary in the candidate trajectory item set in the corridor, and perform corridor pushback and time scale stretch writeback to generate a feasible candidate trajectory set;

[0098] Furthermore, the candidate trajectory entry set within the corridor is expanded into a sampling point sequence according to the candidate trajectory number. The control cycle number of the sampling point sequence is aligned item by item with the interlock effective time window boundary. Sampling points falling into interlock prohibited segments are marked with an interlock violation flag and merged into the control cycle number range of the violation segments. The sampling point sequence is synchronously aligned with robot state entries. The upper limit of velocity, upper limit of acceleration, and lower limit of drive margin are extracted from the robot state entries and solidified as the boundary verification caliber of the robot state entries. When a sampling point exceeds the boundary verification caliber of the robot state entries, an over-limit flag is written and the over-limit category is registered. The sampling point coordinates and the key point sequence of the corridor boundary are used to check the inclusion of the area point by point. The sampling point that exceeds the boundary is located at the nearest boundary point index and written as an out-of-bounds marker. The sampling point that exceeds the boundary is pushed back to the corridor boundary according to the nearest boundary point index and the sampling point position is written back. The sampling point that exceeds the limit is reduced by stretching the control cycle number mapping of the sampling point sequence and the control cycle number mapping is written back. The interlock violation sampling point is rearranged in the local segment within the interlock effective time window boundary and the effective boundary field is written back. The violation marker is cleared and the candidate trajectory numbers that pass the review are merged and encapsulated to generate a feasible candidate trajectory set.

[0099] It should be noted that the backtracking within the corridor and the time-scale stretching backtracking refer to the backtracking of out-of-bounds sampling points along the inner normal of the key point sequence of the corridor boundary according to the nearest boundary point index and the backtracking of the sampling point position. At the same time, the control cycle number mapping of the out-of-bounds sampling points is stretched to extend the control cycle span between sampling points and the control cycle number mapping is backtracked.

[0100] S4.3: Extract a set of multi-objective entries from the feasible candidate trajectory set, and perform non-dominated preservation and cross-corridor diversity maintenance convergence to generate a cross-corridor Pareto candidate profile.

[0101] Furthermore, the feasible candidate trajectory set is traversed by candidate trajectory number. The sampling point sequence and corridor identifier are used to generate a multi-objective entry set and bind the candidate trajectory number index. The multi-objective entry set is grouped by corridor identifier. The candidate trajectories within each group are compared pairwise to determine their relative merits and demerits and written into the dominance elimination mark. The candidate trajectory numbers that do not hit the dominance elimination mark are merged into the non-dominated retention set within the corridor. The non-dominated retention set within the corridor enters the cross-corridor diversity maintenance and sorting process. The number of retentions for each corridor identifier is controlled according to the proportion of corridor identifiers. At the same time, candidate trajectory numbers covering different bottleneck number sequences are preferentially retained according to the bottleneck number sequence coverage. The candidate trajectory numbers under the crowded corridor identifiers are sorted and pruned according to the difference of the multi-objective entry set. The retained candidate trajectory numbers, together with the candidate trajectory control point sequence pointer, sampling point sequence pointer, and multi-objective entry set, are encapsulated to generate a cross-corridor Pareto candidate file.

[0102] It should be noted that the multi-target entry set is generated by binding the sampling point sequence and corridor identifier with the candidate trajectory number index. It includes at least process control beat entries, energy consumption entries, smoothing entries, effective safety gap entries, and risk scale level summary entries. Among them, the process control beat entries are solidified by the control cycle number range and effective boundary field covered by the sampling point sequence; the energy consumption entries and smoothing entries are solidified by the convergence of changes in velocity, acceleration, and drive margin in the sampling point sequence; the effective safety gap entries are solidified by the inclusion check of the sampling point sequence and the corridor boundary key point sequence and the convergence of the nearest boundary point index; and the risk scale level summary entries are solidified by aligning the risk scale entry sequence with the sampling point sequence according to the control cycle number and merging the hit segments.

[0103] Figure 3The distribution structure of multi-objective entries in the cross-corridor Pareto candidate archive is displayed in scatter plot form. The horizontal axis represents the time cost of multi-objective entries, and the vertical axis represents the risk summary of multi-objective entries. Different colors correspond to different topological corridor numbers, which is used to reflect the convergence effect of "Pareto parallel programming of topological corridor decomposition". The point cloud in each corridor reflects the multi-objective trade-off form of the feasible candidate trajectory set generated in that corridor within the effective time window after non-dominated preservation. The overall point cloud formed by cross-corridor superposition is used to illustrate the necessity of cross-corridor diversity maintenance convergence: on the one hand, it avoids the homogenization of solution sets due to overcrowding of single corridor candidates; on the other hand, it preserves representative candidates under different detour topology markings, which facilitates the structured archiving of "feasible region - corridor - candidate trajectory - multi-objective entry" before subsequent consistency verification. This figure can intuitively support the role of "cross-corridor Pareto convergence" in the selection and coverage guarantee of multi-objective solutions.

[0104] S5: Perform consistency verification between the interlock effective time window boundary and the risk gating threshold for cross-corridor Pareto candidate files, select candidates that can be issued, solidify control cycle instructions, and generate trajectory issuance instruction set.

[0105] S5.1: Based on the cross-corridor Pareto candidate archive, check the interlock effective time window boundary, merge continuous violation segments, and remove interlock violation candidate trajectories to generate a qualified interlock candidate set;

[0106] Furthermore, the cross-corridor Pareto candidate file expands the sampling point sequence according to the candidate trajectory number and reads the interlock effective time window boundary in the effective boundary field. The control cycle number of the sampling point sequence is checked against the interlock effective time window boundary item by item. Sampling points that fall into the interlock prohibited segment are written with an interlock violation mark and the interlock cause code summary is recorded. The interlock violation marks are continuously merged into continuous violation segments according to the control cycle number and the control cycle number of the segment start point and the control cycle number of the segment end point are fixed. The candidate trajectory number with continuous violation segments is written into the interlock elimination list and removed from the cross-corridor Pareto candidate file. The candidate trajectory number that does not enter the interlock elimination list retains the corridor identifier, sampling point sequence pointer, multi-target entry set and corridor effective time window and is merged and encapsulated into an interlock qualified candidate set.

[0107] It should be noted that the interlock violation candidate trajectory refers to the candidate trajectory corresponding to the candidate trajectory number of the continuous violation segment within the interlock prohibition segment defined by the boundary of the interlock effective time window of the sampling point sequence, where there is an interlock violation mark and the segments are merged.

[0108] S5.2: Based on the interlocked qualified candidate set, the risk scale entries and risk gate threshold entries are checked for threshold hits, and risk violation segments are merged. At the same time, high-risk candidate trajectories are removed to generate a risk qualified candidate set.

[0109] Furthermore, the interlocked qualified candidate set is expanded into a sampling point sequence and a multi-target item set according to the candidate trajectory number, and the corridor identifier and corridor effective time window are read. The risk calibrator package provides a risk scale item sequence and risk gating threshold items. The risk scale items and risk gating threshold items are aligned with the control cycle number according to the boundary number index. The risk scale level is mapped to a level number and compared with the threshold number of the risk gating threshold item. The control cycle number that meets the requirements is written into the hit mark and a risk violation mark is generated. The risk violation marks are continuously merged into risk violation segments according to the control cycle number, and the control cycle number of the segment start point and the control cycle number of the segment end point are fixed. When the risk violation segment overlaps with the corridor effective time window, an overlap mark is written. The candidate trajectory number that reaches the high risk scale level and has a risk violation segment is written into the risk elimination list and removed from the interlocked qualified candidate set. The candidate trajectory number is retained and merged and encapsulated with the corridor identifier, multi-target item set and sampling point sequence pointer to generate a risk qualified candidate set.

[0110] It should be noted that the threshold hit check is used to align risk scale entries and risk gate threshold entries by boundary number index and control cycle sequence number. The risk scale level is mapped to a level sequence number in four order, and the risk gate threshold entry is mapped to a threshold sequence number in four threshold levels. A hit mark is written when the level sequence number is greater than or equal to the threshold sequence number, and a miss mark is written when the level sequence number is less than the threshold sequence number. The risk scale level and threshold level are uniformly classified as low, medium, higher, and high, and correspond to the sequence numbers 1, 2, 3, and 4 (Example: if the risk scale level is higher, corresponding to sequence number 3, and the risk gate threshold entry requires a threshold level of medium, corresponding to sequence number 2, a hit mark is written; if the risk scale level is medium, corresponding to sequence number 2, and the risk gate threshold entry requires a threshold level of high, corresponding to sequence number 4, a miss mark is written).

[0111] S5.3: Extract process control cycle items and risk scale level summary items from the risk qualified candidate set, perform priority sorting and secondary boundary consistency check, and generate a candidate list for issuance;

[0112] Furthermore, the risk-qualified candidate set is expanded into a multi-target entry set according to the candidate trajectory number and the sampling point sequence pointer is located. The control cycle number range covered by the sampling point sequence is used to generate process control beat entries and written into the process control beat entry field. The risk scale level summary entries and process control beat entries are jointly written into the candidate sorting entries and attached with corridor identifiers and sampling point sequence pointers. The candidate sorting entries are arranged from low to high priority according to the risk scale level summary entries. Candidate sorting entries with the same risk scale level summary entries are arranged from small to large according to the process control beat entries. Parallel candidates are written with parallel clipping marks according to the coverage of corridor identifiers. The candidate trajectory numbers of the first-ranked candidates are checked for secondary boundary consistency. The range of the sampling point sequence control cycle number is checked item by item with the effective boundary field and overlapped with the interlocked effective time window boundary and the corridor effective time window. The verification is passed and the candidate trajectory number is written into the issuance order number and encapsulated to generate the issuance candidate list.

[0113] Figure 4 The bar chart summarizes the number of candidate trajectories screened at each level across all control cycles, providing the total number of candidate trajectories, the number of qualified interlocked candidates, the number of qualified risk candidates, and the final size of the issued candidate list. This illustrates the reduction and retention relationships within the "compliance screening" chain throughout the entire process. By displaying different scheme groups side-by-side, it becomes clear that: consistency verification at the interlock effective time window boundary prioritizes filtering out trajectories that do not meet process control timing constraints at the "number of qualified interlocked candidates" level; after risk scale entries and risk gate threshold entries participate in threshold hit verification, they further converge at the "number of qualified risk candidates" level, eliminating high-risk candidates and thus keeping the final issueable candidates within the executable range. This funnel-shaped comparison serves as evidence of "multi-objective solution set compliance screening": it not only demonstrates that screening does indeed occur at traceable hierarchical nodes but also shows that the final issuance size is jointly determined by the dual constraints of interlocks and risks, supporting the reliability of instruction issuance and the security of process control.

[0114] S5.4: Expand the sampling point sequence corresponding to the candidate list and write the target pose and target joint angle instruction fields and the effective boundary field to generate the trajectory delivery instruction set.

[0115] Furthermore, the candidate list is issued, and the candidate trajectory number is determined by the issuance sequence number. The sampling point sequence pointer is located. The sampling point sequence is expanded incrementally according to the control cycle number, and the continuity of the control cycle number is checked. When a gap occurs, a placeholder instruction is written into the gap control cycle number, and a missing measurement mark is registered. The pose and joint angle of the sampling point corresponding to each control cycle number are extracted. The pose is written into the target pose instruction field, and the joint angle is written into the target joint angle instruction field. The range of control cycle numbers covered by the sampling point sequence is written into the effective boundary field. The effective boundary field is used to solidify the overlap boundary between the interlock effective time window boundary and the corridor effective time window. The trajectory start and stop positions are located by the beginning and end of the control cycle number and start / stop marks and safety stop marks are written into them. The candidate trajectory number, control cycle number, target pose instruction field, target joint angle instruction field, and effective boundary field are encapsulated according to the control cycle number to generate the trajectory issuance instruction set.

[0116] This embodiment also provides a computer device applicable to robot trajectory planning methods based on multi-objective optimization, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the robot trajectory planning method based on multi-objective optimization as proposed in the above embodiment.

[0117] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0118] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the robot trajectory planning method based on multi-objective optimization as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0119] In summary, this invention achieves dynamic safety boundary quantification through multi-source data spatiotemporal alignment and risk scale modeling, and enhances planning robustness by integrating process control interlocking logic; it also achieves multi-objective solution set compliance screening through cross-corridor Pareto convergence and consistency verification, ensuring that instructions meet process control timing constraints and guaranteeing issuance reliability.

[0120] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A robot trajectory planning method based on multi-objective optimization, characterized in that, include: Collect robot status, workstation interlocks, and environmental observations, and perform timestamp alignment and short-term prediction entry registration to generate a scene alignment collection table; Input the scene alignment data collection table into the risk scale model, delineate the effective safety boundaries, extract risk scale entries, and generate a risk calibrator package. The steps are as follows: The risk scaling model includes an input embedding layer, an uncertainty mapping layer, a risk aggregation layer, and a risk gating layer; In the input embedding layer, the robot status entries, workstation interlock entries, environmental observation entries, and dynamic obstacle short-term prediction entries in the scene alignment acquisition table are rearranged and encoded into a unified channel sequence. In the uncertainty mapping layer, the observation uncertainty evidence field is extracted from the unified channel sequence and aligned and spliced ​​with the workstation interlock reason code to output the risk scale candidate entries; In the risk aggregation layer, continuous segments of the risk scale candidate entries are merged into the risk scale aggregation, the effective safety boundary is defined, and the effective safety boundary entry set is output. At the risk gating layer, risk scale entries and risk gating threshold entries are extracted from the set of effective security boundary entries and written into the risk scale level, and the sequence of high-risk boundaries and risk scale entries is output. The effective security boundary entry set, high-risk boundary, risk scale entry sequence, and interlock effective time window boundary are encapsulated, and the correspondence between the boundary number and the corridor clipping index is registered as the mapping entry field to generate the risk caliber package. Based on the risk caliber package, risk contraction is performed on the robot's traversable feasible domain to obtain traversable boundary segments, and detour topology marking and merging are performed and corridor boundary clipping is fixed to generate a topology corridor candidate set; Candidate trajectories are assigned to the candidate set of topological corridors, and constraint checks and projection reclamation within the corridors are performed to obtain a set of feasible candidate trajectories. At the same time, non-dominated retention and cross-corridor convergence are performed to generate cross-corridor Pareto candidate profiles. The consistency between the interlocking effective time window boundary and the risk gating threshold is verified for the cross-corridor Pareto candidate files, and candidates that can be issued are selected. At the same time, the control cycle instructions are solidified and the trajectory issuance instruction set is generated.

2. The robot trajectory planning method based on multi-objective optimization as described in claim 1, characterized in that, The steps for collecting robot status, workstation interlocks, and environmental observations are as follows: The robot's pose, joint angle, velocity, acceleration, and drive margin are read according to the process control sampling caliber, and the acquisition time stamp and control cycle number are fixed to generate robot state entries. Read the interlock input signal, interlock output enable signal and interlock reason code from the process control interlock interface, and perform debouncing and solidify the interlock effective time window to generate workstation interlock entries; Static obstacle boundaries and dynamic obstacle observation sequences are extracted from environmental sensors, and boundary closure verification, duplicate observation merging, and missing measurement slot filling are performed to generate environmental observation entries.

3. The robot trajectory planning method based on multi-objective optimization as described in claim 1 or 2, characterized in that, The steps for generating the scene alignment acquisition table are as follows: Establish a unified clock grid for process control, and map robot status entries, workstation interlock entries, and environmental observation entries to the same control cycle number to generate short-term prediction entries for robot traversable domain boundaries and dynamic obstacles. Assign a prediction effective time window to the short-term prediction items of dynamic obstacles, obtain the set of aligned prediction items, write it into the row-level alignment key, and generate a scene alignment acquisition table.

4. The robot trajectory planning method based on multi-objective optimization as described in claim 1, characterized in that, The steps for generating the candidate set of topological corridors are as follows: Based on the risk caliber package, the robot's traversable feasible domain boundary is effectively corrected within the interlock effective time window, and inward risk contraction and connectivity segmentation are performed on high-risk boundaries to generate a set of traversable boundary fragments. In the set of passable boundary segments, the starting and ending areas are solidified, the bottleneck number sequence and detour direction information are registered, and combined into detour topology markers to generate a set of corridor skeleton segments; The outer envelope boundary is expanded along the corridor skeleton fragment set, and clipping, tightening and continuous splicing are performed. At the same time, the key point sequence of the corridor boundary and the effective time window of the corridor are solidified to generate a candidate set of topological corridors.

5. The robot trajectory planning method based on multi-objective optimization as described in claim 4, characterized in that, The steps for generating cross-corridor Pareto candidate profiles are as follows: Within the effective time window of the corridor in the candidate set of the topological corridor, candidate trajectory skeletons are allocated, candidate trajectory control point sequences are obtained, and discretized into sampling point sequences to generate a set of candidate trajectory entries in the corridor. Verify the inclusion of the interlock effective time window boundary, robot state item boundary and corridor boundary in the candidate trajectory item set within the corridor, and perform corridor backtracking and time scale stretching write-back to generate a feasible candidate trajectory set; Extract a set of multi-objective entries from the feasible candidate trajectory set, and perform non-dominated preservation and cross-corridor diversity maintenance convergence to generate a cross-corridor Pareto candidate profile.

6. The robot trajectory planning method based on multi-objective optimization as described in claim 5, characterized in that, The steps for verifying the consistency between the interlocking effective time window boundary and the risk gating threshold of the cross-corridor Pareto candidate files are as follows: Based on the cross-corridor Pareto candidate archive, the interlock effective time window boundary is checked, continuous violation segments are merged, and interlock violation candidate trajectories are eliminated to generate a qualified interlock candidate set. Based on the interlocked qualified candidate set, the risk scale entries and risk gate threshold entries are checked for threshold hits, and risk violation segments are merged. At the same time, high-risk candidate trajectories are removed to generate a risk qualified candidate set.

7. The robot trajectory planning method based on multi-objective optimization as described in claim 6, characterized in that, The steps for generating the trajectory and issuing the instruction set are as follows: Extract process control cycle items and risk scale level summary items from the risk qualified candidate set, perform priority sorting and secondary boundary consistency check, and generate a candidate list for issuance; Expand the sampling point sequence corresponding to the candidate list and write the target pose and target joint angle instruction fields and the effective boundary field to generate the trajectory delivery instruction set.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the robot trajectory planning method based on multi-objective optimization as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the robot trajectory planning method based on multi-objective optimization as described in any one of claims 1 to 7.

Images