SLAM system, deployment method and device for humanoid robot in lithium battery factory

Abstract

The application provides a SLAM system, a deployment method and equipment for a humanoid robot in a lithium battery factory, the deployment method comprising: firstly acquiring a scene label of the humanoid robot and collecting multi-sensor data, and generating a degradation vector of the SLAM after quality perception calculation of a state vector; then matching a candidate parameter group from a preset station level parameter template library in combination with the scene label, the degradation vector and a task stage; then performing end-side priority safety arbitration and parameter adaptive deployment by the humanoid robot in combination with a real-time state; finally, performing station level positioning confirmation before a mechanical arm executes autonomous operation, if a positioning confidence is greater than or equal to a safety operation threshold, the mechanical arm is allowed to execute autonomous operation, if the positioning confidence is less than the safety operation threshold, the mechanical arm is prohibited to execute autonomous operation and enters a degradation mode. The application overcomes the positioning degradation problem of the humanoid robot in a complex production line environment, and ensures the safety of the humanoid robot operation.

Description

Technical Field

[0001] This invention relates to the field of lithium battery production technology, and in particular to a method, system and equipment for deploying a SLAM system for humanoid robots in lithium battery factories. Background Technology

[0002] With the rapid development of intelligent manufacturing and embodied intelligence technology, humanoid robots are gradually penetrating the industrial manufacturing field, especially entering lithium battery PACK factories to perform complex tasks such as applying adhesive, closing covers, sealing / potting, loading materials, electrical testing, and material handling.

[0003] Unlike traditional AGVs / AMRs, humanoid robots must not only have the ability to move and position, but also provide a stable spatial reference for the operation of the upper body and robotic arm. Positioning errors not only affect the navigation trajectory, but also the relative positional relationship between the end effector and the workpiece, fixture, test probe, glue valve, screw hole, box cover edge and workstation button.

[0004] Therefore, SLAM systems in PACK factory scenarios cannot only pursue the minimum global trajectory error, but also need to switch parameters around the task stage and workstation-level safety. For example, during the production line channel movement stage, more emphasis can be placed on robustness and real-time performance; during the approach to the workstation stage, the weight of local maps and visual labels needs to be increased; before the robotic arm operation, multi-source pose consistency confirmation is required; and when personnel or AGVs enter the safety zone, conservative mode or remote operation mode needs to be triggered.

[0005] Existing humanoid robot SLAM systems still have the following shortcomings when dealing with PACK factory scenarios: First, the parameters of existing SLAM algorithms are fixed and it is difficult to adapt to complex and ever-changing degradation scenarios; Second, existing navigation systems lack task context and cannot meet the requirements for switching parameters at the workstation level; Third, existing solutions lack operational-level positioning confirmation and robotic arm safety interlocking mechanisms.

[0006] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0007] This invention provides a method, system, and equipment for deploying a SLAM system for humanoid robots in lithium battery factories. The main objective of this invention is to solve the technical problems mentioned in the background section of the prior art.

[0008] The first aspect of this invention provides a method for deploying a SLAM system for humanoid robots in lithium battery factories, comprising: Obtain the scene tags of the humanoid robot's current location and collect data from multiple sensors; Based on the multi-sensor data, quality perception is performed, state vectors are calculated, and composite degradation modes of multi-source fusion SLAM are identified to generate degradation vectors. Combining the scene label, the degradation vector, and the current task stage of the humanoid robot, the corresponding parameter strategy is matched from the preset workstation-level parameter template library to generate candidate parameter groups; The candidate parameter set is sent to the humanoid robot, which then performs edge-priority security arbitration based on the real-time security status and executes parameter adaptive deployment. Before the humanoid robot enters the workstation operation stage and the robotic arm performs autonomous operation, workstation-level positioning confirmation is performed. If the location confidence level is greater than or equal to the safe operation threshold, the robotic arm is allowed to perform the autonomous operation; if the location confidence level is less than the safe operation threshold, the robotic arm is prohibited from performing the autonomous operation, and the system switches to a degraded mode of UWB-dominated + visual identification-assisted operation, limiting the maximum movement speed to v. _max ≤0.3m / s.

[0009] In an optional embodiment of the first aspect of the present invention, the state vector includes at least two of LiDAR quality, visual quality, IMU quality, foot / wheel speed odometer quality, UWB quality, visual identifier quality, multi-source consistency quality, and mission and safety status. The degradation vector includes at least two of the following: LiDAR degradation, visual degradation, IMU degradation, odometer degradation, UWB degradation, visual identifier degradation, multi-source consistency conflict, and dynamic interference.

[0010] In an optional embodiment of the first aspect of the present invention, the workstation-level parameter template library includes at least two of the following: a moving cruise template, a workstation proximity template, a pre-operation confirmation template, a dynamic interference template, a reflective / low-light template, and a conservative retreat template. The candidate parameter group includes at least two of the following: LiDAR voxel resolution, ICP corresponding point distance, local map radius, keyframe threshold, visual feature threshold, reprojection error threshold, visual identifier threshold, IMU covariance scaling, odometry fusion weight, UWB fusion weight, UWB NLOS filtering threshold, fusion weight, sensor gating threshold, maximum moving speed, and robotic arm operation permission flag.

[0011] In an optional embodiment of the first aspect of the present invention, the safety arbitration with end-side priority by the humanoid robot in conjunction with real-time safety status includes: After the candidate parameter set is generated at the edge, the humanoid robot receives the candidate parameter set and evaluates the current safety zone status of the personnel / AGV and the status of the robotic arm. If the safety requirements are met, the candidate parameter group is applied; if the current location is unreliable, the parameters are abnormal, or personnel enter the safe zone, the policy update is frozen, the conservative rollback template is switched to, the conservative fusion weight and safe speed are rolled back, and the manual confirmation or remote operation mode is entered.

[0012] In an optional embodiment of the first aspect of the present invention, the execution of station-level positioning confirmation includes: Determine the consistency between the SLAM estimated pose and the multi-source pose of UWB position, visual identifier, or workstation coordinates; Assess the quality of local map matching and the stability of the robotic arm's base posture; The robotic arm is only allowed to perform the autonomous operation when the positioning confidence is greater than the safe operation threshold, the error of the multi-source pose consistency is less than the preset threshold, and the surrounding personnel / AGVs are not in the danger zone.

[0013] In an optional embodiment of the first aspect of the present invention, when the scene label is a workstation for applying adhesive, closing a cover, electrical testing, or tab cutting, and a degradation mode caused by reflection of reflective film, metal box, or screen reflection is detected, resulting in visual overexposure, deep holes, or abnormal LiDAR reflection, the system reduces the fusion weight of the sensor that is experiencing degradation, increases the auxiliary positioning weight of UWB or visual labels, and mandates that the consistency between the visual labels and the workstation coordinates be verified before operation.

[0014] In an optional embodiment of the first aspect of the present invention, the deployment method further includes a cross-line deployment and data backflow closed-loop step: The scene recognition, degradation scoring, parameter switching, manual intervention, positional reliability judgment, and robotic arm interlock triggering events during the operation of the humanoid robot are recorded in a structured manner. The recorded data is fed back to the platform to form a degradation sample library, which optimizes and iterates the workstation-level parameter templates and generates initial parameter templates for deployment in new factories or production lines.

[0015] A second aspect of the present invention provides a SLAM system for humanoid robots in lithium battery factories, the SLAM system comprising: The data acquisition module is used to acquire the scene labels of the humanoid robot and collect data from multiple sensors; The degradation perception module is used to perform quality perception based on the multi-sensor data, calculate the state vector and identify the composite degradation mode of multi-source fusion SLAM, and generate the degradation vector. The strategy engine module is used to combine the scene label, the degradation vector and the current task stage of the humanoid robot to match the corresponding parameter strategy from the preset workstation-level parameter template library and generate candidate parameter groups. The safety arbitration module is used to send the candidate parameter group to the humanoid robot, which then performs edge-priority safety arbitration based on the real-time safety status and executes parameter adaptive deployment. The pre-execution confirmation module is used to perform station-level positioning confirmation when the humanoid robot enters the workstation operation stage and before the robotic arm performs autonomous operation. If the positioning confidence level is greater than or equal to the safe operation threshold, the robotic arm is allowed to perform the autonomous operation; if the positioning confidence level is less than the safe operation threshold, the robotic arm is prohibited from performing the autonomous operation, and the robot switches to a degraded mode of UWB-dominated + visual identification assistance, limiting the maximum movement speed to v. _max ≤0.3m / s.

[0016] A third aspect of the present invention provides a computer device, the computer device comprising: a memory and at least one processor, the memory storing instructions, and the memory and the at least one processor being interconnected via a circuit; The at least one processor invokes the instructions in the memory to cause the computer device to execute a deployment method for a SLAM system for humanoid robots in a lithium battery factory as described in any one of the first aspects of the invention.

[0017] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a deployment method for a SLAM system for a humanoid robot in a lithium battery factory as described in any one of the first aspects of the present invention.

[0018] Beneficial Effects: This invention provides a SLAM system, deployment method, and equipment for humanoid robots in lithium battery factories. The deployment method includes: first, acquiring scene labels for the humanoid robot and collecting multi-sensor data; then, generating a SLAM degradation vector after calculating the state vector through quality perception; next, combining scene labels, degradation vectors, and task stages to match candidate parameter groups from a pre-set workstation-level parameter template library; then, the humanoid robot performs edge-priority safety arbitration and parameter adaptive deployment based on its real-time state; finally, before the robotic arm performs autonomous operation, workstation-level positioning confirmation is performed. If the positioning confidence is greater than or equal to the safe operation threshold, the robotic arm is allowed to perform autonomous operation; if the positioning confidence is less than the safe operation threshold, the robotic arm is prohibited from performing autonomous operation and enters a degraded mode. This invention overcomes the problem of humanoid robot positioning degradation in complex production line environments and ensures the safety of humanoid robot operations. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of an embodiment of the deployment method of a SLAM system for humanoid robots in a lithium battery factory, which is an exemplary embodiment of the present invention. Figure 2 This is a schematic diagram of an embodiment of the SLAM system framework for humanoid robots in lithium battery factories, which is an exemplary invention. Figure 3 This is a schematic diagram illustrating an embodiment of the relationship between multi-sensor quality scoring and workstation-level parameter templates according to the present invention; Figure 4 This is a schematic diagram of an embodiment of the present invention, illustrating the humanoid robot's approach to the workstation and pre-operation positioning confirmation. Figure 5 This is a schematic diagram of an embodiment of the complete process of scene labeling, degradation recognition, parameter template, and secure execution closed loop of the present invention; Figure 6 This is a schematic diagram of an embodiment of a multi-sensor fusion SLAM degradation identification and parameter adaptive deployment system architecture, which is an example of the present invention. Figure 7 This is a schematic diagram of an embodiment of a computer device that is an example of the present invention. Detailed Implementation

[0020] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] When humanoid robots enter the PACK production line to perform tasks such as applying adhesive, closing caps, sealing / filling, feeding, electrical testing, inspection verification, and material handling, the humanoid robots not only need to complete production line channel navigation, but also need to obtain stable, reliable, and auditable pose estimation before approaching the workstation, aligning with the fixture, and entering the robotic arm operating posture. Traditional general SLAM or AGV positioning solutions are usually point-to-point navigation, which is difficult to meet the comprehensive requirements of humanoid robot workstation-level positioning + pre-operation safety confirmation + multi-source degradation adaptation.

[0022] The PACK factory environment is characterized by strong indoor attributes, unavailability or inability to directly use GNSS / RTK, numerous metal and reflective parts, many repetitive workstations and shelves, frequent dynamic interference from personnel / AGVs / robotic arms, severe local occlusion, and the need for precise alignment of workstation coordinates and production line coordinates with humanoid robot coordinates. Fixed multi-sensor fusion SLAM parameters, fixed fusion weights, and fixed sensor gating thresholds make it difficult to maintain stable performance across different workstations, different operation stages, and different degradation modes.

[0023] The main drawbacks of traditional humanoid robot positioning methods include the following: (1) The lack of a PACK factory scenario-based labeling system makes it impossible to distinguish different positioning needs such as applying adhesive, closing the cover, sealing, feeding, electrical testing, production line channel, material rack area and AGV intersection area.

[0024] (2) The lack of a unified quality score for multi-sensor degradation makes it difficult to identify complex degradation such as LiDAR reflection, low visual illumination, IMU vibration, foot slippage, UWB NLOS, visual sign occlusion and multi-source pose conflict.

[0025] (3) There is a lack of parameter template library that is bound to the workstation task stage, and the general SLAM parameters cannot be adaptively switched between the movement, approach, pre-operation confirmation, job execution and withdrawal stages.

[0026] (4) The lack of humanoid robot operation safety linkage may allow the robotic arm to enter the work station operation area even when the positioning is unreliable.

[0027] (5) There is a lack of deployment data feedback mechanism across workstations, production lines and factories, and it is difficult to precipitate on-site faults and manual takeover data into reusable degraded samples.

[0028] This invention proposes a deployment method for a SLAM system for humanoid robots in lithium battery factories, the main purpose of which is: 1. Establish a scene labeling system and workstation-level positioning requirement model for humanoid robots in PACK factories.

[0029] 2. Establish a multi-sensor fusion SLAM quality perception and degradation recognition method, covering multiple sources of information such as LiDAR, vision, IMU, foot / wheel speed odometer, UWB, visual identification and workstation map.

[0030] 3. Establish a mapping relationship between scene labels, degradation types, task stages, and parameter templates to achieve adaptive deployment of fusion weights, gating thresholds, covariance scaling, keyframe strategies, and security parameters.

[0031] 4. Establish a workstation-level positioning reliability and humanoid robot arm operation interlock mechanism to prohibit or downgrade autonomous operation when positioning is unreliable.

[0032] 5. Establish a closed loop for data feedback between the humanoid robot end, edge end, and platform end to achieve continuous iteration of parameter strategies, degradation samples, and cross-line body deployment templates.

[0033] See Figure 1 The deployment method of the SLAM system for humanoid robots in lithium battery factories includes: S100: Obtain the scene label of the humanoid robot's current location and collect multi-sensor data. Before elaborating on the method of this invention, the system architecture on which this method relies will first be described, see [link to documentation]. Figure 2 The system of this invention mainly comprises four components: robot end, edge computing end, platform end, and production line system end. The robot end is equipped with a SLAM system and a humanoid robot motion / operation control system; the edge end performs multi-sensor quality scoring, degradation scene recognition, parameter strategy generation, safety gating, and local caching; the platform end maintains the PACK scene label library, degradation sample library, parameter template library, and strategy version; the production line system end provides workstation number, equipment coordinates, MES / MOM tasks, WMS / AGV status, safety PLC, electronic fence, and operation permission information.

[0034] More specifically, the robot end is the system's data acquisition source and physical motion execution carrier. To cope with complex industrial environments, humanoid robots are equipped with a rich heterogeneous sensor network, mainly including: LiDAR (Light Detection and Ranging), RGB-D cameras, wrist cameras, IMU (Inertial Measurement Unit), joint encoders, foot or wheel speed odometers, and are equipped with UWB tags and visual identifiers (such as QR code recognition modules). This end integrates SLAM algorithms on the local running end to output real-time pose, and includes motion execution and emergency stop control modules. The robot end is not only responsible for reporting raw sensor data and SLAM status in real time, but also for receiving and applying parameter configurations sent down from the upper layer, and for performing hardware-level safety stop actions in emergency situations.

[0035] As the intermediary hub connecting the robot and the cloud platform, the edge computing terminal undertakes high-real-time computing and parameter arbitration tasks. This terminal mainly includes the following core modules: Multi-source quality scoring and degradation scene recognition: responsible for fine-grained quality assessment of multi-source data uploaded from the robot, real-time calculation of degradation vectors, and identification of whether LiDAR or visual composite degradation has occurred; Parameter strategy engine: based on the current degradation mode and task context, matching and generating appropriate SLAM fusion parameter sets; Workstation-level confidence calculation: before the robotic arm prepares for operation, comprehensively evaluating the local map matching quality and multi-source pose consistency, and outputting the operation safety assessment result; Safety gating and rollback: performing edge-priority safety arbitration on the parameter sets to be sent to the robot, triggering parameter rollback or freeze update if an anomaly is detected; Local data caching: ensuring that local control logic can be maintained even during network fluctuations, and caching data samples to be returned.

[0036] The platform serves as the brain and data asset center of the entire system, focusing on global strategy management and data iteration. It maintains multiple core databases: PACK scenario tag library, parameter template library, and failure sample library. At the same time, the platform is responsible for strategy version management and provides visualization and auditing functions for operation and maintenance personnel to monitor. More importantly, based on the continuously accumulated degradation and failure samples, the platform supports the replication of optimized parameter templates across production lines / factories, thereby significantly reducing the debugging costs of deploying new production lines.

[0037] The production line system represents the original information and automation infrastructure of the PACK factory. It interacts with the platform / edge end through API interfaces and mainly includes: MES / MOM (Manufacturing Execution System), WMS / AGV scheduling system, safety PLC / access control, workstation coordinates / fixture data, work tickets / permissions, and electronic fences. This end provides the essential task context (e.g., which workstation the robot should go to to perform what task, and the precise coordinates of the target workstation) and global safety constraints (e.g., whether the electronic fence of the current workstation is open, and whether any AGVs are traversing the conflict area).

[0038] Specifically, an exemplary scene labeling system table for a PACK factory according to the present invention can be shown in Table 1 below: Table 1. A scene labeling system of the present invention

[0039] Taking a specific scenario as an example, in step S100, the process can be as follows: The humanoid robot communicates with the production line system (such as the MES system or WMS system) or obtains the current PACK scene label (e.g., A03 adhesive application station, A15 capping station, long straight aisle of the production line, etc.) by visually recognizing the workstation QR code. At the same time, the system collects multi-sensor data from the humanoid robot in real time, including but not limited to 3D LiDAR point cloud, visual camera images, IMU data, foot / wheel speed odometer data, UWB base station signal, visual tag (AprilTag / QR code) data, and preset workstation map data.

[0040] S200: Based on the multi-sensor data, perform quality perception, calculate state vectors, identify composite degradation modes of multi-source fusion SLAM, and generate degradation vectors. The state vectors include at least two of the following: LiDAR quality, visual quality, IMU quality, foot / wheel speed odometer quality, UWB quality, visual identifier quality, multi-source consistency quality, and task and safety status. The degradation vectors include at least two of the following: LiDAR degradation, visual degradation, IMU degradation, odometer degradation, UWB degradation, visual identifier degradation, multi-source consistency conflict, and dynamic interference.

[0041] Specifically, an exemplary state vector table of the present invention can be shown in Table 2 below. The system calculates the degradation vector D based on the multi-source state vector S. The degradation vector may include: D _L Indicates LiDAR degradation, D _V Indicates visual degeneration, D _I Indicates IMU degradation, D _O Indicates degradation of the foot / wheel speed odometer, D _U Indicates UWB degradation, D _T Indicates visual identity degradation, D _C Indicates a multi-source consistency conflict, D _dyn Indicating dynamic interference, in this invention, the degradation score can be calculated using rule thresholds, weighted scoring, machine learning classifiers, or reinforcement learning strategy networks. For each sensor, the system not only determines whether it is faulty, but also determines whether it is in a normal, mildly degraded, severely degraded, or recovering state, and uses a sliding window and hysteresis mechanism to avoid instantaneous misjudgment.

[0042] Table 2. A state vector table of the present invention

[0043] This invention abandons the traditional single failure judgment and instead performs fine-grained quality perception on multi-source data. The processing can be exemplified as follows: the system extracts the quality indicators of each sensor to form a state vector S. For example, LiDAR quality indicators include Hessian condition number, reflectivity anomaly ratio, and ICP residual; visual quality indicators include feature point count, reprojection error, and depth hole ratio; UWB quality indicators include NLOS (non-line-of-sight) flag, etc. Based on the state vector S, the system calculates and generates a degradation vector D. In a preferred embodiment, a sliding window and hysteresis mechanism can be used to avoid degradation misjudgment caused by transient noise. In this invention, the system not only determines whether a sensor has failed, but also finely classifies its state as normal, slightly degraded, severely degraded, or recovering.

[0044] S300: Combining the scene label, the degradation vector, and the current task stage of the humanoid robot, a corresponding parameter strategy is matched from a preset workstation-level parameter template library to generate a candidate parameter group. The workstation-level parameter template library includes at least two of the following: mobile cruise template, approaching workstation template, pre-operation confirmation template, dynamic interference template, reflective / low-light template, and conservative retreat template; the candidate parameter group includes at least two of the following: LiDAR voxel resolution, ICP corresponding point distance, local map radius, keyframe threshold, visual feature threshold, reprojection error threshold, visual identifier threshold, IMU covariance scaling, odometry fusion weight, UWB fusion weight, UWB NLOS filtering threshold, fusion weight, sensor gating threshold, maximum moving speed, and robotic arm operation permission flag.

[0045] Specifically, this invention establishes a station-level parameter template library for the PACK factory. This parameter template library is not a single SLAM configuration file, but rather a multi-layered parameter set composed of scene tag templates, degradation mode templates, task phase templates, and conservative rollback templates. An exemplary station-level parameter template library of this invention can be shown in Table 3 below: Table 3. A workstation-level parameter template library table of the present invention

[0046] In this invention, the policy engine can generate a parameter set P based on the scene label L, degradation vector D, task stage T, and safety state A. The parameter set may include: LiDAR voxel resolution, distance to ICP corresponding point, local map radius, keyframe threshold, visual feature threshold, reprojection error threshold, visual identifier threshold, IMU covariance scaling, odometry fusion weight, UWB fusion weight, UWB NLOS filtering threshold, fusion weight, sensor gating threshold, maximum moving speed, and robotic arm operation permission flag, etc.

[0047] In one exemplary application scenario of the present invention, the processing procedure of this step can be as follows: Based on the multi-dimensional station-level parameter template library established by the system, the edge strategy engine maps and generates candidate parameter groups P according to the currently acquired scene label L, degradation vector D, and current task stage T (such as mobile cruise, approaching the station, and pre-operation preparation). For example: approaching the station template: when the robot enters the A10 sealing station from the channel, the template automatically increases the local map density, enhances the visual identifier / UWB weight, and reduces the chassis movement speed; dynamic interference template: when AGV passage is detected at the A01 loading station (dynamic point ratio increases), the template increases the robustness core strength and temporarily suspends the writing of high dynamic frames into the key frame map.

[0048] For a more complete version, see Figure 3 The specific process by which this invention maps the real-time observation status of the sensor, through quantitative evaluation and fusion transformation, to specific parameter configurations adapted to the current task context can be described as follows: Multi-sensor quality feature extraction: The system first monitors and extracts the underlying quality indicators of each sensor in real time to finely characterize its observation status. Specifically, this includes: LiDAR quality: extracting point cloud density, the condition number of point cloud registration (characterizing the strength of geometric feature constraints and used to judge geometric degradation), and the proportion of highly reflective point clouds; Visual quality: extracting the number of effective feature points, image exposure status (judging whether it is overexposed or underexposed), and reprojection error of feature tracking; IMU quality: monitoring the degree of high-frequency vibration, zero-bias stability, and whether there are abnormal attitude changes; Odometry quality: for humanoid robots, monitoring the stability of foot contact, whether there is chassis / foot slippage, and the speed consistency between multiple wheels / feet; UWB quality: monitoring whether the signal is in NLOS (non-line-of-sight) occlusion state, the jump frequency of localization calculation and its covariance; Visual label quality: evaluating the occlusion ratio of labels such as QR codes, the observation angle (tilt), and the confidence level of the recognition algorithm output.

[0049] Construction and Correction of Fusion Degradation Vector: The quality scores of each individual sensor are not used in isolation. The system aggregates these scores to construct a comprehensive fusion degradation vector D, corresponding to the degradation state of radar, vision, IMU, odometry, UWB, and other dynamic / consistency dimensions, respectively. After generating this vector, the system does not directly output parameters, but introduces three key correction mechanisms: Multi-source consistency scoring: cross-validating whether there are serious discrepancies in the pose outputs of different sensors; Scene label correction: assigning different prior confidence levels to different sensors based on the current location (such as adhesive application station, production line aisle); Task stage correction: determining whether the robot is currently in the movement or operation stage.

[0050] Workstation-level parameter template mapping: Based on the corrected fusion degradation vector D and the task context, the system maps to one of five typical preset parameter templates through the strategy engine: Moving cruise template: suitable for long straight production line channels, the strategy focuses on robustness priority, activates low-speed obstacle avoidance logic, and strengthens the filtering of dynamic obstacles (such as pedestrians and AGVs); Approaching workstation template: suitable for the process of the robot stopping at the work fixture. The strategy switches to high-precision positioning, triggering local map encryption (improving voxel resolution) and significantly enhancing the weight of visual identifiers in the fusion algorithm; Pre-operation confirmation template: applicable before the robotic arm is about to begin operation, the strategy executes extremely strict multi-source consistency threshold verification. If the consistency does not meet the standard, the robotic arm hardware-level interlock is triggered, and autonomous operation is strictly prohibited; Dynamic interference template: triggered when a large amount of dynamic interference is detected (such as forklifts passing by and obstructing the line of sight), the strategy actively reduces the robot's moving speed, suspends keyframe extraction and map updates at the SLAM front end (to prevent map pollution), and requests remote takeover in extremely severe cases; Conservative rollback template: triggered when severe composite degradation or system failure occurs, the system switches to the most conservative safety parameter group, freezes all state updates to prevent pose rollover, and stores the environment and state data at the time of the failure as failure samples in the database for subsequent data backflow and learning on the platform.

[0051] Low-level parameter set output: After determining the template, the system finally parses out the parameter set output that can be directly called by the robot's low-level SLAM nodes and motion controller. These specific parameters include, but are not limited to: the fusion weights and covariance scaling factors of each sensor, the gating threshold for outlier data removal, the voxel size of point cloud filtering, the map keyframe insertion conditions, the chassis speed limit, and the most critical robotic arm operation permission flag. Through the above refined mapping, this invention realizes the smooth adaptive switching of humanoid robot perception-decision-execution in complex factory environments.

[0052] S400. The candidate parameter set is sent to the humanoid robot, which then performs edge-side priority safety arbitration based on the real-time safety status and executes parameter adaptive deployment. The parameter sending follows the principle of "edge-side safety arbitration priority." In this invention, the platform can publish templates and policy versions, and the edge can generate candidate parameter sets, but whether to execute them ultimately is determined by the robot's safety module based on the real-time status, task stage, and robotic arm interlocking conditions.

[0053] To ensure the safety of the actual production line, the distribution of parameters adopts the principle of prioritizing end-side safety arbitration.

[0054] After the candidate parameter group P generated at the edge is sent to the robot, it is not immediately forced to take effect. The robot's safety module will arbitrate based on the current real-time status (such as whether personnel have entered the safety zone and the robot's real-time motion posture). If the environment is determined to be safe, the parameter group is applied to perform SLAM parameter hot update. If unreliable positioning, abnormal parameters, or emergency avoidance instructions issued on the production line are detected, the strategy update is frozen, and the robot is switched to a conservative rollback template, forcing the robot to roll back to conservative fusion weights and safe low speed, or even directly enter the manual remote operation mode (i.e., after the candidate parameter group is generated at the edge, the humanoid robot receives the candidate parameter group and evaluates the current safety zone status of personnel / AGV and the status of the robotic arm; if the safety requirements are met, the candidate parameter group is applied; if the current positioning is unreliable, the parameters are abnormal, or personnel have entered the safety zone, the strategy update is frozen, the robot is switched to a conservative rollback template, rolls back to conservative fusion weights and safe speed, and enters the manual confirmation or remote operation mode).

[0055] S500: Before the humanoid robot enters the workstation operation stage and the robotic arm performs autonomous operations, a workstation-level positioning confirmation is performed. This invention differs from the key steps of traditional AGV navigation in that the humanoid robot must not only be able to move but also to perform its actions accurately. Before the robot enters the workstation and the robotic arm is ready to perform autonomous operations (such as grasping, applying adhesive, or pressing), a strict multi-source positioning confirmation must be performed.

[0056] In this invention, when the humanoid robot transitions from the movement phase to the workstation operation phase, the system performs workstation-level positioning confirmation. This confirmation process includes at least: consistency judgment between SLAM pose and UWB position, consistency judgment between SLAM pose and visual identifier / workstation coordinates, local map matching quality judgment, robot arm base pose stability judgment, and personnel / AGV safety zone status judgment. Only when the workstation-level positioning confidence level C... _pose Greater than the operation threshold C _op The system only allows the robotic arm to enter the autonomous operation process when the multi-source consistency error is less than the threshold and the personnel / AGV are not in the danger zone. If the conditions are not met, the system prohibits the robotic arm from performing high-risk actions and enters the remote operation, manual confirmation, repositioning, or conservative retreat mode.

[0057] More specifically, in the complex PACK factory environment, even a slight deviation in chassis positioning can lead to serious collisions before the robotic arm of a humanoid robot performs physical interaction tasks such as grasping and assembly. Therefore, this invention designs a strict safety interlock and state transition mechanism based on positional confidence. See [link to relevant documentation]. Figure 4 The specific steps are as follows: Production line navigation: First, when the humanoid robot moves a long distance in a relatively open production line aisle or a production line aisle with many dynamic obstacles, the system is in the production line navigation stage. At this time, the system calls the aforementioned robust positioning template, which focuses on maintaining the continuity of positioning and anti-interference ability, and avoids navigation interruption due to the loss of sporadic features.

[0058] Approaching the target workstation: Subsequently, when the robot navigates to a preset distance from the target workstation, the process enters the approaching the target workstation stage. At this time, the system triggers a seamless switch of parameter strategies based on the location label, calls a high-precision positioning template (or a workstation approach template), and the system automatically increases the resolution of the local map and increases the weight of sensor observation data to prepare for pose convergence for the upcoming fine-tuning operation.

[0059] Multi-source pose confirmation: Before the robot chassis reaches the predetermined operation point and is ready to extend the robotic arm, the system forcibly enters the multi-source pose confirmation stage. Specifically, the system will extract and compare the output results of the multi-source positioning system at the current moment, which mainly include: the relative continuous pose output by the multi-sensor fusion SLAM algorithm; the global absolute positioning reference provided by the factory UWB system; the relative pose calculated by recognizing the visual markers of the workstation (such as ArUco codes and QR codes) through the robot's end effector or head camera; and the prior true value of the target workstation coordinates issued by the production line MES system.

[0060] Confidence determination and safety gating: Next, the system performs confidence determination based on the above multi-source data. The system calculates the cross error (error threshold) between multi-source poses, evaluates the size of the covariance matrix of the current SLAM algorithm posterior optimization, and checks the confidence of the label output by the vision algorithm. Based on the above indicators, the system determines whether the current comprehensive positioning accuracy meets the stringent requirements of the robotic arm operation.

[0061] Action arbitration based on the judgment result (safety interlock): Based on the confidence judgment result, the system executes strict branch arbitration logic: If the judgment is passed: that is, the multi-source pose height is consistent, the error is lower than the safety threshold, and the covariance converges well, the system releases the safety interlock, the state switches to allow operation, and at this time the humanoid robot's robotic arm can safely enter the work area to perform the predetermined physical interaction task.

[0062] If the system fails the test, it indicates that there has been a subtle positioning drift, degradation, or multi-source data conflict. The system immediately triggers hardware and software interlocks and switches to a prohibited / degraded state. In this state, the system adheres to the bottom line of not allowing autonomous operation when the positioning is unreliable and forcibly prohibits the robotic arm from operating to prevent equipment collisions. The system can request remote "manual confirmation / remote operation" to take over, or instruct the chassis to perform a conservative retreat action to return to the previous safe and well-positioned node to try again.

[0063] Failure Recording and Feedback Optimization: For cases that fail the judgment, the system will automatically record the original sensor data, environmental characteristics, and reasons for failure when the interlock is triggered, while prohibiting operation. This data, as high-value negative samples, will be fed back to the platform for strategy optimization. The platform will then optimize the parameter template or adjust the threshold for that specific workstation, enabling the robot to adaptively adopt a better strategy the next time it approaches the target workstation, forming a continuous closed-loop evolution of system safety and robustness.

[0064] S600. If the positioning confidence level is greater than or equal to the safe operation threshold, the robotic arm is allowed to perform the autonomous operation. If the positioning confidence level is less than the safe operation threshold, the robotic arm is prohibited from performing the autonomous operation, and the system switches to a degraded mode of UWB-dominated + visual identification assistance, limiting the maximum movement speed to v. _max ≤0.3m / s.

[0065] In this invention, taking a specific scenario as an example, the system compares the errors between the SLAM-estimated pose and the UWB position and visual identifier coordinates, performs multi-source pose consistency judgment, evaluates the stability of the robotic arm base pose and the matching quality of the local map, and calculates the comprehensive station-level positioning reliability C. _pose Only when C _pose Greater than the set operation threshold C _op Only when the production line safety PLC confirms that personnel / AGVs are not in a danger zone will the system release the interlock and allow the robotic arm to perform high-risk actions (i.e., judging the consistency of multi-source poses between SLAM estimated pose and UWB position, visual identifier or workstation coordinates; judging the local map matching quality and the stability of the robotic arm base pose; the robotic arm is only allowed to perform the autonomous operation when the positioning confidence is greater than the safe operation threshold, the error of the multi-source pose consistency is less than the preset threshold, and the surrounding personnel / AGVs are not in a danger zone). Otherwise, the robotic arm is prohibited from extending, the system reports an anomaly and enters degraded mode.

[0066] In an optional embodiment of the first aspect of the present invention, the deployment method further includes a cross-line deployment and data feedback closed-loop step: performing structured recording of scene recognition, degradation scoring, parameter switching, manual takeover, positional reliability judgment, and robotic arm interlock triggering events during the operation of the humanoid robot; feeding the recorded data back to the platform to form a degradation sample library, optimizing and iterating the workstation-level parameter template, and generating an initial parameter template for deployment in a new factory or new production line.

[0067] In this invention, the system records in a structured manner each scene recognition, degradation score, parameter switching, location reliability judgment, pre-operation confirmation, manual takeover, conservative rollback, and task result. The recorded content includes scene label, workstation number, sensor quality status, parameter group version, fusion weight, positioning residual, UWB / visual identification consistency, personnel / AGV status, whether the robotic arm is allowed to operate, failure reason, and recovery strategy. After the above data is fed back to the platform, it forms a PACK factory degradation sample library, parameter template optimization records, and cross-line replication templates. When deploying in a new factory or new production line, the initial parameter template can be selected according to the scene similarity and quickly adapted using a small amount of on-site data.

[0068] Taking a specific scenario as an example, in this invention, every instance of robot degradation identification, parameter switching, positioning confidence alarm, robotic arm interlock triggering, and manual intervention on the production line is fed back to the platform as structured data (including the current state vector and parameter version). Based on this real production line data, the platform continuously optimizes the workstation-level parameter templates. When faced with deployment tasks in new factories or new PACK production lines, the system can directly call mature template copies from the platform based on similar scenario tags. Only a small amount of on-site data is needed for fine-tuning to complete the adaptive adaptation of humanoid robot SLAM parameters, greatly shortening the cycle of manual parameter tuning on-site.

[0069] To better illustrate the deployment method of the SLAM system for humanoid robots in lithium battery factories according to the present invention, the present invention provides the following embodiments based on the technical verification indicators in Table 1 below: Table 1. Technical Effect Verification Index Table of the Invention

[0070] Example 1: Reflective film at the A03 adhesive application station causes visual degradation: For example, before a humanoid robot performs the A03 adhesive application task, the robot needs to move from the production line aisle to the vicinity of the adhesive application fixture and identify the tape, the application area, and the edge of the fixture. Due to the obvious reflective properties of the tape and protective film, the vision camera experiences localized overexposure, the depth camera produces holes, the number of visual feature points decreases, and the reprojection error increases. Based on the A03 adhesive application station label and visual degradation score, the system reduces the weight of visual measurement and increases the weight of UWB and visual identifier-assisted positioning; simultaneously, it enables proximity to the station template, reduces the movement speed, increases the local map density, and requires consistency verification between the visual identifier and the station coordinates before operation. If the verification passes, the robotic arm is allowed to enter the adhesive application preparation posture; if the verification fails, it enters the remote operation or manual confirmation mode (that is, when the scene label is adhesive application (A03), cover closing (A15), electrical testing (A05) or tab cutting (A10) station, and the system identifies a degradation mode caused by reflective film, metal box or screen reflection resulting in visual overexposure, deep holes or abnormal LiDAR reflection, the system reduces the fusion weight of the degraded sensor, increases the auxiliary positioning weight of UWB or visual sign, and forces the visual sign and station coordinates to be verified before operation).

[0071] Example 2: Metal Box Reflection and Personnel Obstruction at the A15 Closing Station: For example, at the A15 closing station, there are metal boxes, lid edges, and human personnel obstructing the view. The LiDAR generates abnormal reflection points on the metal surface, and the visual camera is affected by personnel obstruction, resulting in a decrease in local map matching scores. After the system detects an increase in the proportion of abnormal LiDAR reflections, an increase in the proportion of visual obstruction, and a decrease in the multi-source consistency score, it switches to a conservative approach to the template at the closing station: enhancing dynamic target filtering, limiting keyframe map updates, reducing movement speed, and increasing the pre-operation positioning confirmation threshold. Only after personnel leave the safe zone and multi-source pose consistency is restored is the system allowed the robot to perform pre-closing observation or auxiliary operations.

[0072] Example 3: Dynamic Interference Caused by AGV Movement at Loading Station A01: For example, near the loading station A01, AGVs frequently move around and share the passage with humanoid robots. The system obtains AGV proximity information from the AGV scheduling system and simultaneously detects an increase in the proportion of dynamic points and fluctuations in ICP residuals from the LiDAR point cloud. The parameter strategy engine switches to the dynamic interference template, improves the robustness core strength, temporarily suspends the writing of high-dynamic frames to the keyframe map, reduces the robot's maximum speed, and enables avoidance or waiting strategies. If the AGV enters the safe zone, the robot's safety module refuses to perform new parameter exploration and robotic arm movements, maintaining only conservative positioning and standby mode.

[0073] Example 4: Repetitive geometry in long, straight production line corridors leads to LiDAR degradation: For example, when a humanoid robot moves along a long, straight corridor in a PACK production line, the repeated arrangement of workstations, columns, and guardrails on both sides weakens the LiDAR geometric constraints in the forward direction, increases the Hessian condition number, and decreases the minimum eigenvalue. The system combines labels for the long, straight production line corridor, lowers the keyframe insertion threshold, increases the local map radius based on UWB and visual marker positioning, appropriately increases the local map radius, and limits the movement speed. If a UWB or visual marker is occluded, the system switches to a conservative corridor template, requiring the robot to reconfirm its pose at the next visible marker.

[0074] Example 5: Multi-source pose inconsistency triggers robot arm interlock: For example, after the robot arrives at the A12 electrical test station, the SLAM estimated pose differs from the UWB localization result by more than a threshold, while the visual tag recognition confidence level is below the threshold. The system calculates the station-level positioning confidence level C. _pose Below the operating threshold C _op Therefore, the robotic arm is prohibited from entering the test probe operation area and reports an event to the platform that "positioning is unreliable and autonomous operation is not allowed". The robot can perform repositioning, remote operation or manual confirmation process. The event is recorded as a failure sample, including sensor status, parameter group, workstation label, pose difference and final processing result.

[0075] In summary, see Figure 5 The entire process of deploying a SLAM system for humanoid robots in lithium battery factories, as described in this invention, can be summarized as follows: Workstation / Area Recognition (Task Context Acquisition): The starting point of this invention's process is to acquire the semantic environment in which the robot is currently located. The system determines the current workstation or area label through interaction with the production line system or local visual recognition. Typical scenarios include the "A03 Adhesive Application" workstation, the "A15 Cover Closing" workstation, ordinary "production line passages," and AGV intersection areas with complex dynamic environments. This recognition result provides a basic task context for subsequent parameter matching.

[0076] Multi-sensor quality acquisition: While acquiring scene labels, the robot's underlying system collects raw data from various sensors in real time and extracts quality features. The collection objects include, but are not limited to, LiDAR (Light Detection and Ranging), vision (camera), IMU (Inertial Measurement Unit), odometry (foot or wheel speed), UWB (Ultra-Wideband), and visual labels. The system not only focuses on the data values ​​of the sensors, but also focuses on extracting underlying indicators that reflect the current observation quality (such as the number of feature points, point cloud matching residuals, etc.).

[0077] Degradation Score Calculation: Based on the collected multi-source quality data, the system performs refined degradation score calculations to identify specific degradation patterns. The system can identify a variety of typical industrial degradation scenarios: Geometric degradation: For example, degradation of LiDAR due to lack of longitudinal structural features in long straight production line aisles; Low-light reflection: For example, visual feature failure caused by metal enclosures or reflective films in the A15 cover-up station; Slippage vibration: Odometry and IMU noise caused by robot chassis slippage or high-frequency movements of robotic arms; NLOS (Non-line-of-sight): UWB signals blocked by large equipment; Multi-source conflict: Severe discrepancies in pose estimation outputs from different sensors.

[0078] Parameter Template / Strategy Generation: After identifying a specific degradation pattern and combining it with the current workstation label, the system extracts or generates a response strategy from a pre-set template library. The SLAM and control parameters dynamically adjusted by the system mainly include: Fusion Weights and Covariance: reducing the weight of degraded sensors (increasing their covariance) and increasing the weight of healthy sensors; Gating Thresholds and Keyframes: dynamically adjusting the gating threshold for removing abnormal data and the insertion logic of map keyframes (such as temporarily suspending keyframe insertion in highly dynamic AGV intersection areas); Speed ​​Limitation: actively generating chassis deceleration commands in severe degradation scenarios.

[0079] Safety gating (end-side arbitration): This is a crucial step in ensuring production line safety. Before the generated parameter strategy can take effect, it must undergo rigorous safety gating verification. Gating conditions include comprehensive location reliability assessment, task stage determination (whether it is in a stage where parameter mutations are allowed), the intrusion status of personnel / AGVs in the current area, and the operation permissions issued on the production line. Only when all safety gating conditions are met is the strategy allowed to be issued and executed; otherwise, the system will trigger parameter freezing or safety rollback.

[0080] Robot execution: After passing the safety gate, the robot continues to perform physical actions using the updated SLAM parameters. The execution phase includes routine navigation, approaching the workstation, pre-operation confirmation after reaching the target point (i.e., multi-source pose consistency interlock verification before the robotic arm moves), and finally autonomous operation. If the pre-operation confirmation fails, the robot exits the autonomous mode and requests manual remote operation takeover.

[0081] Data Feedback and Data Loop: The method of this invention is not unidirectional but forms a continuously iterative data loop. When the robot completes its execution or an anomaly occurs, the system feeds back the execution results. The fed-back data includes: the success / failure status of the task, human intervention records, SLAM posterior residuals, the system's parameter switching history, and recovery results after the anomaly. This high-value, real-world production line structured data flows to the platform through the bottom data loop channel for continuous updating and optimization of the degradation sample library, scenario template library, parameter template library, and strategy version. Through this closed-loop mechanism, the system can exhibit increasingly stronger adaptability and robustness when deployed across production lines or factories.

[0082] In summary, the deployment method of the SLAM system for humanoid robots in lithium battery factories of the present invention has at least the following technical advantages: (1) Compared with general SLAM systems, this invention incorporates PACK factory scene semantics, workstation labels and work safety requirements into the deployment of fusion positioning parameters.

[0083] (2) Compared with single sensor positioning, the present invention integrates the quality status of LiDAR, vision, IMU, UWB, odometer and visual marker, and can identify composite degradation and adjust the fusion strategy.

[0084] (3) Compared with the AGV navigation scheme, the present invention addresses the dual needs of humanoid robot movement and robotic arm operation by adding pre-operation positioning confirmation and robotic arm interlocking mechanism.

[0085] (4) Compared with manual parameter tuning, the present invention enables rapid deployment and continuous optimization across workstations, production lines and factories through parameter template library and data feedback mechanism.

[0086] (5) Compared with pure algorithm adaptation, this invention adds end-side security arbitration, conservative rollback, remote operation degradation and task security linkage, which is suitable for deployment on real production lines.

[0087] It should also be noted that, for the technical solution of this invention, there are the following alternative implementations, all of which should fall within the protection scope of this invention: Sensor combinations can be replaced: depending on the robot's form factor, combinations such as LiDAR+IMU+UWB, LiDAR+vision+IMU, and visual identification+UWB+odometer can be selected.

[0088] Alternative strategy generation methods are available: parameter strategies can be generated from rule templates, reinforcement learning policy networks, supervised learning classifiers, Bayesian optimization, or human expert templates.

[0089] The source of scene labels can be replaced: they can be generated automatically by manual configuration, MES / MOM workstation data, robot map semantic annotation, or visual recognition.

[0090] Alternative positioning confirmation methods include: UWB / SLAM consistency, visual identification / workstation coordinate consistency, end-point visual alignment results, or multi-source confidence fusion.

[0091] Deployment architecture can be alternative: parameter policies can be generated on the robot, edge workstation, or platform, but final execution should be arbitrated by the robot's security module.

[0092] See Figure 6 The second aspect of the present invention provides a SLAM system for humanoid robots in lithium battery factories, the SLAM system comprising: The data acquisition module 10 is used to acquire the scene label where the humanoid robot is currently located and to collect data from multiple sensors; The degradation perception module 20 is used to perform quality perception based on the multi-sensor data, calculate the state vector and identify the composite degradation mode of multi-source fusion SLAM, and generate the degradation vector. The strategy engine module 30 is used to combine the scene label, the degradation vector and the current task stage of the humanoid robot to match the corresponding parameter strategy from the preset workstation-level parameter template library and generate a candidate parameter group. The safety arbitration module 40 is used to send the candidate parameter group to the humanoid robot, which then performs edge-priority safety arbitration based on the real-time safety status and executes parameter adaptive deployment. The pre-execution confirmation module 50 is used to perform station-level positioning confirmation when the humanoid robot enters the workstation operation stage and before the robotic arm performs autonomous operation; if the positioning confidence is greater than or equal to the safe operation threshold, the robotic arm is allowed to perform the autonomous operation; if the positioning confidence is less than the safe operation threshold, the robotic arm is prohibited from performing the autonomous operation, and the robot switches to a degraded mode of UWB-dominated + visual identification assistance, and the maximum movement speed is limited to v. _max ≤0.3m / s.

[0093] Figure 7This is a schematic diagram of a computer device according to an embodiment of the present invention. The computer device can vary significantly due to differences in configuration or performance, and may include one or more processors 60 (central processing units, CPUs) (e.g., one or more processors) and memory 70, and one or more storage media 80 (e.g., one or more mass storage devices) for storing recording programs or data. The memory and storage media can be temporary or persistent storage. The program stored in the storage media may include one or more modules (not shown in the figure), each module including a series of instruction operations on the computer device. Furthermore, the processor may be configured to communicate with the storage media and execute the series of instruction operations stored in the storage media on the computer device.

[0094] The computer device in this embodiment of the invention may further include one or more power supplies 90, one or more wired or wireless network interfaces 100, one or more input / output interfaces 110, and / or one or more operating systems, such as Windows Server, Mac OS, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 7 The computer device structure shown does not constitute a limitation on the computer device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0095] This invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the deployment method for the SLAM system of humanoid robots for lithium battery factories.

[0096] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system, module, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0097] If the integrated unit is implemented as a software functional unit or a combination of software and hardware and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0098] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for deploying a SLAM system for humanoid robots in lithium battery factories, characterized in that, include: Acquire scene tags of the humanoid robot's current location and collect data from multiple sensors; Based on the multi-sensor data, quality perception is performed, state vectors are calculated, and composite degradation modes of multi-source fusion SLAM are identified to generate degradation vectors. Combining the scene label, the degradation vector, and the current task stage of the humanoid robot, the corresponding parameter strategy is matched from the preset workstation-level parameter template library to generate candidate parameter groups; The candidate parameter set is sent to the humanoid robot, which then performs edge-priority security arbitration based on the real-time security status and executes parameter adaptive deployment. Before the humanoid robot enters the workstation operation stage and the robotic arm performs autonomous operation, workstation-level positioning confirmation is performed. If the location confidence level is greater than or equal to the safe operation threshold, the robotic arm is allowed to perform the autonomous operation; if the location confidence level is less than the safe operation threshold, the robotic arm is prohibited from performing the autonomous operation, and the system switches to a degraded mode of UWB-dominated + visual identification-assisted operation, limiting the maximum movement speed to v. _max ≤0.3m / s.

2. The deployment method of the SLAM system for humanoid robots in lithium battery factories according to claim 1, characterized in that, The state vector includes at least two of the following: LiDAR quality, visual quality, IMU quality, foot / wheel speed odometer quality, UWB quality, visual identifier quality, multi-source consistency quality, and mission and safety status. The degradation vector includes at least two of the following: LiDAR degradation, visual degradation, IMU degradation, odometer degradation, UWB degradation, visual identifier degradation, multi-source consistency conflict, and dynamic interference.

3. The deployment method of the SLAM system for humanoid robots in lithium battery factories according to claim 1, characterized in that, The workstation-level parameter template library includes at least two of the following: mobile cruise template, workstation proximity template, pre-operation confirmation template, dynamic interference template, reflective / low-light template, and conservative retreat template. The candidate parameter group includes at least two of the following: LiDAR voxel resolution, ICP corresponding point distance, local map radius, keyframe threshold, visual feature threshold, reprojection error threshold, visual identifier threshold, IMU covariance scaling, odometry fusion weight, UWB fusion weight, UWB NLOS filtering threshold, fusion weight, sensor gating threshold, maximum moving speed, and robotic arm operation permission flag.

4. The deployment method of the SLAM system for humanoid robots in lithium battery factories according to claim 1, characterized in that, The safety arbitration method, which prioritizes the edge side based on the humanoid robot's real-time safety status, includes: After the candidate parameter set is generated at the edge, the humanoid robot receives the candidate parameter set and evaluates the current safety zone status of the personnel / AGV and the status of the robotic arm. If the safety requirements are met, the candidate parameter group is applied; if the current location is unreliable, the parameters are abnormal, or personnel enter the safe zone, the policy update is frozen, the conservative rollback template is switched to, the conservative fusion weight and safe speed are rolled back, and the manual confirmation or remote operation mode is entered.

5. The deployment method of the SLAM system for humanoid robots in lithium battery factories according to claim 1, characterized in that, The execution station-level positioning confirmation includes: Determine the consistency between the SLAM estimated pose and the multi-source pose of UWB position, visual identifier, or workstation coordinates; Assess the quality of local map matching and the stability of the robotic arm's base posture; The robotic arm is only allowed to perform the autonomous operation when the positioning confidence is greater than the safe operation threshold, the error of the multi-source pose consistency is less than the preset threshold, and the surrounding personnel / AGVs are not in the danger zone.

6. The deployment method of the SLAM system for humanoid robots in lithium battery factories according to claim 1, characterized in that, When the scene label is an adhesive application, cover closing, electrical testing, or tab cutting station, and a degradation mode such as visual overexposure, deep voids, or abnormal LiDAR reflection caused by reflective film, metal box, or screen reflection is detected, the system reduces the fusion weight of the sensor that caused the degradation, increases the auxiliary positioning weight of UWB or visual labels, and forces a consistency check between the visual labels and the station coordinates before operation.

7. The deployment method of the SLAM system for humanoid robots in lithium battery factories according to claim 4, characterized in that, It also includes the closed-loop steps of cross-line deployment and data backflow: The scene recognition, degradation scoring, parameter switching, manual intervention, positional reliability judgment, and robotic arm interlock triggering events during the operation of the humanoid robot are recorded in a structured manner. The recorded data is fed back to the platform to form a degradation sample library, which optimizes and iterates the workstation-level parameter templates and generates initial parameter templates for deployment in new factories or production lines.

8. A SLAM system for humanoid robots in lithium battery factories, characterized in that, The SLAM system includes: The data acquisition module is used to acquire the scene labels of the humanoid robot and collect data from multiple sensors; The degradation perception module is used to perform quality perception based on the multi-sensor data, calculate the state vector and identify the composite degradation mode of multi-source fusion SLAM, and generate the degradation vector. The strategy engine module is used to combine the scene label, the degradation vector and the current task stage of the humanoid robot to match the corresponding parameter strategy from the preset workstation-level parameter template library and generate candidate parameter groups. The safety arbitration module is used to send the candidate parameter group to the humanoid robot, which then performs edge-priority safety arbitration based on the real-time safety status and executes parameter adaptive deployment. The pre-execution confirmation module is used to perform station-level positioning confirmation when the humanoid robot enters the workstation operation stage and before the robotic arm performs autonomous operation. If the positioning confidence level is greater than or equal to the safe operation threshold, the robotic arm is allowed to perform the autonomous operation; if the positioning confidence level is less than the safe operation threshold, the robotic arm is prohibited from performing the autonomous operation, and the robot switches to a degraded mode of UWB-dominated + visual identification assistance, limiting the maximum movement speed to v. _max ≤0.3 m / s.

9. A computer device, characterized in that, The computer device includes: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a line; The at least one processor invokes the instructions in the memory to cause the computer device to execute the deployment method of the SLAM system for humanoid robots in lithium battery factories as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the deployment method of the SLAM system for humanoid robots in lithium battery factories as described in any one of claims 1-7.

Images