Garbage recycling method

Abstract

The invention discloses a method applied to a garbage collection robot. The robot receives the garbage bag through the delivery port and obtains user category priori, and an automatic mouth tying device behind the delivery port instantly seals the recognized garbage bag; physical characteristics in the bag are obtained through non-intrusive recognition, and harm is judged; target classification and confidence are determined by fusing prior and features, a sorting mechanism is driven to throw garbage bags into corresponding collection grids, deodorization and sterilization devices are arranged in the collection grids, and smoke detectors are arranged in dangerous goods isolation boxes; the system monitors the full load degree of each collection grid in real time, and when the danger is identified or any collection grid is predicted to be full, a relay dumping task is dynamically re-planned and inserted based on the travel, the road condition and the recovery point queuing information, so that the collection-dumping collaborative scheduling is realized. According to the method, accurate classification is completed on the entrance side, a harmful recognition, isolation and priority clearance closed loop is established, stagnation caused by single-class full load is avoided, and accuracy, efficiency, safety and operation continuity are improved.

Description

Technical Field

[0001] This application relates to the field of intelligent robots, and more particularly to a waste recycling method. Background Technology

[0002] With the advancement of community waste sorting and intelligent recycling, existing solutions primarily rely on platform-triggered door-to-door collection. After completing tasks at several collection points, the vehicles or robots typically return to the recycling station (or collection point) along a fixed route for unified disposal. Entry-side identification often depends on resident declarations, bag color / labels, or single visual algorithms, making it difficult to obtain the physical characteristics inside the bag when it is intact. Vehicle-mounted multi-sorting bins operate on a static schedule, lacking linkage with real-time / predictive loading status. Faced with dynamic changes in the availability of recycling point dispensing slots and queuing times, route planning lacks rapid response and a coordinated mechanism for "transfer and disposal en route." For suspected or identified hazardous materials, the common practice is primarily alarm alerts, lacking closed-loop capabilities for isolation, priority enhancement, and linkage with disposal routes, easily leading to mis-disposal, operational interruptions, and safety risks.

[0003] Based on the above situation, the following technical problems urgently need to be solved: First, the accuracy of inlet sorting is limited by resident declarations or single identification methods, lacking the ability to obtain the physical characteristics inside the bag under the condition that the bag is intact, integrate it with user category information, and output target classification and classification confidence. Second, there is a lack of diversion and control mechanisms for different judgment results, making it difficult to implement differentiated storage and verification isolation for sorted items, suspected hazardous items, and hazardous items. Third, the loading status of each container on the vehicle lacks real-time monitoring and predictive triggering, making it impossible to dynamically update the route based on factors such as travel distance, road traffic, and queuing information at recycling points, and to organize "garbage collection / transfer and dumping" into an orderly task sequence when "hazardous items are detected" or "a container has reached / is about to reach a threshold." Fourth, the disposal of hazardous materials lacks priority enhancement and linkage with route replanning, making it difficult to form a closed-loop treatment in a timely and safe manner. The above problems collectively restrict the sorting accuracy, operational continuity, and system safety of community waste automatic recycling. Summary of the Invention

[0004] This application aims to overcome the problems in existing technologies, such as inaccurate entry point identification, insufficient capacity coordination, lack of linkage between path scheduling and recycling points, and non-closed-loop disposal of hazardous materials.

[0005] According to one aspect of this application, a waste recycling method is provided, applied to a waste recycling robot, the waste recycling robot including a delivery port, an identification module, a control unit, a sorting mechanism, multiple internal collection compartments, a capacity monitoring module, and a navigation planning module; the method includes the following steps: S10. Receive the garbage bag delivered by the user through the delivery port and obtain the category information sent by the user. An automatic sealing device is provided behind the delivery port to seal the garbage bag after identification. S20. Using the identification module, the physical characteristics of the items inside the garbage bag are obtained without damaging the garbage bag, and it is determined whether the garbage bag contains hazardous waste. S30. Using the control unit, the category information and physical characteristics are integrated to determine the target category and classification confidence level of the garbage bag; S40. The sorting mechanism is driven by the control unit to sort the garbage bags into the corresponding target collection compartments according to the target classification and classification confidence level; wherein, the target collection compartment includes a regular collection compartment, a waiting-for-inspection compartment, and a hazardous materials isolation box; the multiple internal collection compartments are equipped with deodorizing and sterilizing devices for odor removal and disinfection of the environment within the collection compartments; the hazardous materials isolation box is equipped with a smoke detector for monitoring fire and triggering a safety warning; S50. The full load of each collection cell is monitored in real time through the capacity monitoring module. S60. The navigation planning module dynamically updates the robot's work route based on at least one of the following triggering conditions: S61 and S20 determine that the garbage bag contains hazardous waste; S62. The real-time or predicted full load of any of the collection cells reaches a preset threshold. The updated work routes are used to guide robots to continue performing collection tasks or to go to recycling points for transfer and dumping.

[0006] Preferably, S20 includes: S21. The appearance image information of the garbage bag is acquired through the image acquisition device in the recognition module; S22. In the presence of an electronic tag, the information of the electronic tag attached to the garbage bag is read by the radio frequency identification reader in the identification module; S23. The control unit determines the preliminary category and preliminary confidence level of the garbage bag based on the appearance image information and electronic tag information. S24. The control unit determines whether a hazard warning visual feature or a label field matches the hazard blacklist. If so, the garbage bag is determined to contain hazardous waste. S25. If not, obtain the weight of the garbage bag through the weight sensor in the identification module, and obtain its volume through visual ranging based on the image information; S26. The control unit calculates the apparent density of the garbage bag and determines whether the apparent density reaches or exceeds the preset hazardous density criterion. If so, it is determined that the bag contains hazardous waste. S27. If not, near-infrared spectral data are collected and material composition analysis is performed using the near-infrared spectrometer in the identification module to obtain the probability of hazardous materials. S28. The control unit determines whether the probability of the hazardous material reaches or exceeds a preset hazardous material threshold. If so, it is determined that the material contains hazardous waste.

[0007] Preferably, the fusion in S30 specifically includes: The control unit uses the category information sent by the user as a decision bias factor. When the identified physical characteristics match the user category information, the control unit performs an operation to increase the classification confidence and relaxes the action threshold for sorting the garbage bag into the corresponding regular collection grid based on the increased confidence.

[0008] Preferably, S40 specifically includes: If the garbage bag is identified as a hazardous material, it is sorted into a hazardous material isolation box by the sorting mechanism. If the target is classified as general waste and its classification confidence is higher than the first threshold, it is sorted into the corresponding regular collection grid. If the classification confidence level is lower than the first threshold but higher than the second threshold, then the item is sorted into the inspection cell.

[0009] Preferably, after S28, the method further includes: When the probability of the hazardous material does not reach the hazardous material threshold but is within the preset suspicious range, it is sorted into the inspection cell by the sorting mechanism in S40. A timed verification mechanism is activated, and a second verification is performed through the identification module within the preset maximum retention time. The second verification includes performing near-infrared spectral acquisition and material composition analysis again. Subsequently, based on the secondary identification results, the target classification and confidence level are redefined, and S40 and subsequent steps are executed again.

[0010] Preferably, the dynamic updating of the job route in S60 includes: When the trigger condition is S61, the hazardous materials priority handling procedure is executed: S61a. The control unit generates a highest priority transfer task to the hazardous waste disposal port. S61b: The current task sequence is immediately interrupted by the navigation planning module to calculate the fastest path and perform replanning and travel. S61c, Send alarm information to the community management platform and record the entire handling process; When the trigger condition is S62, the normal relay dumping procedure is executed: S62a. A transfer task to clear the corresponding collection cell is generated through the control unit; S62b: Insert the task into the current sequence through the navigation planning module, and replan and proceed with the goal of optimizing the overall operation time.

[0011] Preferably, the path replanning employs a periodic rolling optimization mechanism: The navigation planning module receives environmental status information again at specific time intervals and calculates the optimal path. After each calculation, the robot is driven to travel along the updated path.

[0012] Preferably, the preset condition in S62 is based on a prediction model that the first collection cell is about to reach full capacity; The generation of the prediction model includes: The robot's current task sequence is obtained through the control unit or cloud server; Predict the type and quantity of waste at each task location based on historical data; Predict future loading status based on predicted waste disposal patterns and the current fullness of collection grids; If the predicted state exceeds the threshold, it is determined that the preset conditions are met.

[0013] Preferably, the generation of the prediction model further includes a learning optimization step: The robot's data recording unit or cloud server records historical data including user identification, delivery time, waste type and weight. Based on the historical data, parameters for user or region waste disposal habits are generated. The habitual parameters are fed back into the prediction model to optimize the accuracy of predicting waste disposal patterns.

[0014] Preferably, the waste recycling robot is also equipped with a wireless transmission module for communicating with community waste stations; The navigation planning module receives the availability status and estimated queuing time of different types of drop-off points in real time and uses them as key parameters for path planning.

[0015] This application offers the following advantages: More accurate and interpretable entry point identification. Physical characteristics of the bag's interior are acquired under intact conditions and fused with user-submitted category priors, outputting a "target classification + classification confidence level." This significantly reduces reliance on resident declarations or single visual rules, improving the reliability and interpretability of entry point identification. Tiered sorting and a safe closed-loop system are implemented, with sorting based on confidence level: high-confidence samples are placed in regular collection compartments, suspicious samples in suspected hazardous material testing compartments, and those determined to be hazardous are immediately transferred to a hazardous materials isolation box. The isolation box incorporates a smoke detector for early fire warning, preventing hazardous materials from entering the normal process and forming a rapid "identification-isolation-verification" closed loop. Environmental hygiene and secondary risk control are enhanced. An automatic sealing device behind the disposal port immediately seals the identified garbage bags, and each collection compartment integrates deodorizing and sterilizing devices, effectively suppressing odor and bacterial transmission and reducing the risk of secondary pollution during operation and transportation. Capacity coordination and uninterrupted operation: The capacity monitoring module tracks the full capacity of each collection grid in real time and uses prediction results as trigger conditions to detect the risk of "almost full" in advance, avoiding the entire line from being shut down due to a single category filling up first, and maintaining operational continuity. Route scheduling and recycling point linkage improve efficiency: When hazardous materials are identified or any collection grid reaches or is about to reach its threshold, the navigation planning module dynamically updates the operation route based on the route, road conditions, and recycling point queuing information, inserting transfer and dumping tasks as needed. This achieves coordinated scheduling of collection and dumping, reducing waiting and detour time, and improving the single task completion rate and shift throughput.

[0016] In summary, this application achieves accurate classification at the entry point, hierarchical isolation and environmental sterilization at the mid-stage, and ensures continuous operation and safety priority through capacity-path linkage at the execution side. Overall, it achieves the technical effects of accurate entry point identification, sufficient capacity coordination, efficient linkage between path scheduling and recycling points, and a closed loop of safe disposal. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a module relationship diagram of the garbage recycling robot described in one embodiment of this application; Figure 2 This is a logic block diagram of the waste recycling method described in one embodiment of this application.

[0019] Figure descriptions: 100, Waste recycling robot; 200, User terminal; 10, Drop-off port; 11, Automatic sealing device; 20, Identification module; 30, Control unit; 40, Sorting mechanism; 50, Collection compartment; 51, Regular collection compartment; 52, Inspection compartment; 53, Hazardous materials isolation box; 54, Smoke detector; 60, Capacity monitoring module; 70, Navigation planning module. Detailed Implementation

[0020] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings. Preferred embodiments of this application are shown in the drawings. However, this application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the disclosure of this application.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] Please refer to Figure 1-2 This application provides a waste recycling method applied to a waste recycling robot 100. The waste recycling robot 100 includes a delivery port 10, an identification module 20, a control unit 30, a sorting mechanism 40, multiple internal collection compartments 50, a capacity monitoring module 60, and a navigation planning module 70. The method includes the following steps: S10: Receive the garbage bag delivered by the user through the delivery port 10 and obtain the category information sent by the user through the user terminal 200. An automatic sealing device 11 is installed behind the delivery port 10 to seal the garbage bag after identification. In this step, it should be noted that: the delivery port 10 is equipped with bag entry detection (light curtain / torque / infrared). After the detection is stable, the anti-rebound gate opens to guide the garbage bag into the identification channel. The category information can be sent to the control unit 30 via the device's touchscreen, resident code (QR code / barcode), IC card / NFC, or mobile app as a priori category. The automatic sealing device 11 is located at the end of the identification channel and is triggered by the control unit 30: when S30 / S40 confirms that the garbage bag can enter the "regular collection compartment 51" or "inspection compartment 52", it performs sealing (cable tie or heat seal) to prevent leakage and odor transfer; if it is determined to be a hazardous material, it is sealed in a separate isolation channel or directly introduced into an isolation box and then its lid is closed. The sealing device should have the functions of bag breakage detection and bag jamming reset, and should send the sealing result (success / failure / number of retries) back to the log.

[0023] S20. Using the identification module 20, the physical characteristics of the contents of the garbage bag are obtained without damaging it, and it is determined whether the garbage bag contains hazardous waste. In this step, it should be noted that the identification module 20 consists of image acquisition, RFID / NFC (optional), weighing sensing, volume / depth acquisition (visual ranging / structured light / binocular three-choice or combination), near-infrared spectroscopy (NIR, 900-2500nm, optional), and metal detection (optional). It completes the following sequentially by channel: ① Appearance image acquisition including supplementary lighting and OCR / icon detection, searching for a blacklist of hazardous colors / icons / text; ② Reading the electronic tag (if any) to obtain the category / risk field; ③ Acquiring weight and volume, calculating apparent density; ④ Performing NIR acquisition if direct judgment is not triggered, outputting the probability of material category and the probability of hazardous materials; ⑤ Determining "whether it contains hazardous waste" based on comprehensive evidence (rules + model). The entire process is bag-free. Channels that fail to acquire data are automatically downgraded or retried, and those that do not meet quality standards are triggered for re-acquisition.

[0024] S30. Through the control unit 30, category information and physical features are fused to determine the target classification and classification confidence of the garbage bag. In this step, it should be noted that: the control unit 30 performs feature-level or decision-level fusion of the "user's prior category" and the multimodal features output by S20, outputting the target classification (e.g., food waste / recyclable / hazardous / other) and classification confidence (0-1). When the prior and perception are consistent, the confidence of the corresponding category is increased according to rules (bias factor); when the prior and perception conflict and there is direct evidence of danger, a safety-first strategy is adopted to cover it as hazardous; all intermediate evidence (image fragments, label fields, NIR key bands, density values) and fusion logs are written locally / to the cloud for traceability and retraining.

[0025] S40. The sorting mechanism 40, driven by the control unit 30, sorts the garbage bags to the corresponding target collection grids 50 according to the target classification and classification confidence level. The target collection grids 50 include regular collection grids 51, inspection grids 52, and hazardous materials isolation boxes 53. Multiple internal collection grids 50 integrate deodorization and sterilization devices to deodorize and disinfect the environment within each grid. The hazardous materials isolation box 53 is equipped with a smoke detector 54 to monitor for fire and trigger a safety warning. It should be noted that the sorting mechanism 40 can be a flap / fork / diverter valve / flow belt with position feedback and jam detection. Control logic example: Hazardous assessment → direct entry into the isolation box; Non-hazardous and confidence level ≥ first threshold → entry into regular collection grid 51; Confidence level between the first and second thresholds → entry into inspection grid 52. The standard / inspection collection compartment 50 has a built-in deodorization and sterilization unit (activated carbon + ozone / UV-C or plasma), which runs periodically and is linked to the opening of the lid; the isolation box has a sealing lid, smoke / temperature and humidity / VOC monitoring and electrical interlock. When the smoke alarm is triggered, it reports to the platform and triggers the body's safety strategy (power off, evacuation or dumping nearby).

[0026] S50. The capacity monitoring module 60 monitors the full load of each collection cell 50 in real time. In this step, it should be noted that the capacity monitoring module 60 can use one or more of the following methods: weighing (strain gauge), volume / level (ultrasound / ToF), visual volume estimation, etc., to output a standardized full load (0-100%) and growth rate, and report it to the control unit 30 at a fixed period (e.g., 1-5s); median / exponential smoothing is used for abnormal transition values; and a capacity vector C = {cell 1...cell n, to be inspected, isolated} and a predicted value (which can be from the cloud) are formed to provide a basis for triggering S60.

[0027] S60. The navigation planning module 70 dynamically updates the robot's work route based on at least one of the following trigger conditions: S61 and S20 determine that the garbage bag contains hazardous waste.

[0028] S62, The real-time or predicted full load of any collection grid 50 reaches a preset threshold.

[0029] The updated work routes are used to guide robots to continue performing collection tasks or to go to recycling points for transfer and dumping.

[0030] In this step, it is necessary to explain that: Trigger S61 (Hazardous Goods Priority): Control unit 30 generates the highest priority transfer task "to the hazardous dumping port", navigation planning module 70 saves the current task breakpoint, calculates the fastest path based on cost function J = travel time + estimated queuing time + safety penalty (low battery / restricted area / nighttime restrictions, etc.) (A / Dijkstra dynamic weights are both acceptable), immediately replans and proceeds in parallel; at the same time, it reports to the platform and leaves a full record.

[0031] Trigger S62 (Capacity Relay): When the full load of a collection cell is ≥50 or the threshold is predicted to be reached, a relay task to "empty the cell" is generated, inserted into the current task sequence, and rearranged according to "optimal overall operation time" (heuristic insertion +2-opt / 3-opt, etc.), and the optimal collection point is selected by combining the availability of multiple collection points and queuing time. After dumping is completed, the original sequence is restored according to the breakpoint.

[0032] The navigation module recalculates periodically using a rolling mechanism (e.g., every 5-10 seconds or triggered by an event) and only issues the next waypoint, adapting to rapid changes in road and docking station status.

[0033] The technical solution implemented in this embodiment provides more accurate and interpretable entry point identification. It acquires the physical characteristics of the bag's interior while the bag remains intact and fuses these characteristics with user-submitted category priors, outputting a "target classification + classification confidence level." This significantly reduces reliance on resident declarations or single visual rules, improving the reliability and interpretability of entry point identification. A tiered sorting and safety closed-loop system is implemented, with sorting based on confidence level: high-confidence samples are placed in the regular collection compartment 51, suspicious samples in the suspected hazardous material inspection compartment 52, and those determined to be hazardous are immediately transferred to the hazardous materials isolation box 53. The isolation box contains a smoke detector 54 for early fire warning, preventing hazardous materials from entering the normal process and forming a rapid "identification-isolation-verification" closed loop. For environmental hygiene and secondary risk control, an automatic sealing device 11 behind the delivery port 10 immediately seals the identified garbage bags. Each collection compartment 50 integrates a deodorizing and sterilizing device, effectively inhibiting the spread of odors and bacteria, reducing the risk of secondary pollution during operation and transportation. Capacity coordination and uninterrupted operation: The capacity monitoring module 60 tracks the full load of each collection grid 50 in real time and uses the prediction results as trigger conditions to detect the risk of "about to be full" in advance, avoiding the entire line from being shut down due to a single category being full first, and maintaining the continuity of operations. Route scheduling and recycling point linkage improve efficiency: When hazardous materials are identified or any collection grid 50 reaches or is about to reach its threshold, the navigation planning module 70 dynamically updates the operation route based on the route, road conditions, and recycling point queuing information, inserting transfer and dumping tasks as needed, realizing coordinated scheduling of collection and dumping, reducing waiting and detour time, and improving the single task completion rate and shift throughput.

[0034] In summary, this application achieves accurate classification at the entry point, hierarchical isolation and environmental sterilization at the mid-stage, and ensures continuous operation and safety priority through capacity-path linkage at the execution side. Overall, it achieves the technical effects of accurate entry point identification, sufficient capacity coordination, efficient linkage between path scheduling and recycling points, and a closed loop of safe disposal.

[0035] Furthermore, S20 includes: S21. The image acquisition device in the recognition module 20 acquires the appearance image information of the garbage bag. In this step, it should be noted that the image acquisition device preferably uses a color camera with supplementary lighting. Before acquisition, camera calibration, white balance, and distortion correction are performed. After entering the field of view, background modeling and bag segmentation are performed to obtain the bag's bounding box and outline. Color histogram, primary color category, geometric shape, printed text (OCR), and icon features are extracted, and quality scores are given for overexposure, motion blur, etc. If the quality is insufficient, automatic re-acquisition is performed. The acquired features are transmitted to the control unit 30 in structured vector form, and the original image and timestamp are archived.

[0036] S22. In the presence of an electronic tag, the RFID reader in the identification module 20 reads the information of the electronic tag attached to the garbage bag. In this step, it should be noted that: the RFID reader polls according to the read / write protocol to obtain the tag's unique identifier, category indication code, risk field, batch and expiration information; it performs CRC verification and multiple rereads for jitter reduction; a failed read is recorded as "no tag". Upon successful reading, the fields are mapped using a dictionary and their format is verified, and then entered into the fusion process along with the image features; the original reading, success / failure status, and antenna power information are written to the log.

[0037] S23. Using control unit 30, based on appearance image information and electronic tag information, determine the preliminary category and preliminary confidence level of the garbage bag. In this step, it should be noted that control unit 30 employs either a feature-level or decision-level fusion strategy. Feature-level fusion embeds image features and tags into a lightweight classification model, outputting a category probability vector; decision-level fusion introduces tag priority rules and confidence level weighting on the model output. To improve usability, a probability calibration method is used to calibrate the output, and the preliminary category and preliminary confidence level are returned, while supporting evidence and model version are recorded.

[0038] S24. The control unit 30 determines whether a hazard warning visual feature or a label field matches the hazard blacklist. If so, the garbage bag is determined to contain hazardous waste. In this step, it should be noted that: the hazard warning visual features include three types of templates: pre-set warning color codes, hazard icons, and hazard text, which are matched using a combination of object detection and OCR; the label blacklist includes risk material codes, recall batch codes, and expired high-risk labels. A match in any channel triggers a "direct hazard assessment," outputting the conclusion that hazardous waste is present, while simultaneously solidifying the evidence (hit area, field, confidence level) and initiating subsequent isolation and priority disposal procedures.

[0039] S25. If not, the weight of the garbage bag is obtained through the weight sensor in the identification module 20, and its volume is obtained through visual ranging based on the image information. In this step, it should be noted that: weight measurement uses a high-resolution weighing unit, performs tare, vibration-resistant filtering, and stability window determination to obtain a steady-state weight value. Volume estimation is based on monocular visual ranging: a scale calibration reference is placed in the delivery channel, and the outer three-dimensional envelope is fitted using the bag segmentation contour involving both inside and outside the camera. The volume is calculated using voxel integration or geometric approximation (such as a solid of revolution), and an estimated confidence level is given. If the confidence level is insufficient due to viewpoint occlusion, a slight displacement resampling is triggered.

[0040] S26. The apparent density of the garbage bag is calculated by the control unit 30, and it is determined whether the apparent density reaches or exceeds the preset hazardous density criterion. If so, it is determined that the bag contains hazardous waste. In this step, it should be noted that apparent density equals steady-state weight divided by estimated volume. The hazardous density criterion is obtained through offline calibration and on-site calibration, and includes two sets of standards: absolute threshold and category-related threshold. When the density reaches or exceeds either criterion, the bag is judged to contain hazardous waste. To suppress random errors, an uncertainty range and safety margin are introduced, and critical samples proceed to the next step of component confirmation.

[0041] S27. If not, near-infrared spectral data is collected using the near-infrared spectrometer in identification module 20, and material composition analysis is performed to obtain the probability of hazardous materials. In this step, it should be noted that: near-infrared spectral acquisition uses diffuse reflectance; dark field and whiteboard calibration are performed before acquisition; the raw spectrum is denoised, baseline corrected, standardized, and its derivatives are processed before being input into a chemometrics or machine learning model for material discrimination, identifying hazardous materials such as battery electrolytes, organic compounds containing solvents / detergents, chlorinated and fluorinated polymers, and coatings; the probabilities of each risk group are fused to form the probability of hazardous materials; simultaneously, quality control indicators and key absorption bands are output as evidence.

[0042] S28. The control unit 30 determines whether the probability of hazardous materials reaches or exceeds a preset hazardous materials threshold. If so, it determines that hazardous waste is contained. It should be noted in this step that the hazardous materials threshold is set based on the validation set and the target false negative rate, and a more conservative threshold can be set for specific risk groups. After calibrating the probability, the control unit 30 compares it with the threshold. If the probability reaches or exceeds the threshold, it outputs "Contains hazardous waste" and triggers isolation and priority disposal. If it falls within the suspicious range, it is marked as suspicious and imported into the inspection process, while spectral, image, and density evidence are retained for review and retraining.

[0043] The technical solution implemented in this embodiment enables rapid, low-interference, and interpretable identification of hazardous waste, provided the bag remains intact, following a step-by-step process of "appearance and label prior assessment - weight, volume, and density constraints - near-infrared component confirmation." Hazardously identified samples are immediately isolated and prioritized for transport; samples with abnormal density are intercepted at the physical quantity level; and boundary samples are confirmed at the spectral composition level and undergo closed-loop diversion for suspected hazardous waste. This significantly reduces the risk of mis-disposal, mixed disposal, and missed detection, shortens hazardous waste disposal response time, and provides reliable input for subsequent sorting and route planning, thereby improving the overall accuracy, safety, and operational efficiency of inlet identification.

[0044] In one specific embodiment, the fusion in S30 specifically includes: The category information sent by the user is used as a decision bias factor through the control unit 30.

[0045] When the identified physical characteristics match the user category information, the control unit 30 performs an operation to increase the classification confidence and relaxes the action threshold for sorting the garbage bag into the corresponding regular collection grid 51 based on the increased confidence.

[0046] In this embodiment, it should be noted that the control unit 30 uniformly accesses and integrates the "category information sent by the user" and the "physical feature evidence output by the identification module 20". First, the control unit 30 registers the user category information as a decision bias factor and labels its source and credibility level (such as resident selection, property preset, historical stability, etc.). At the same time, it performs quality verification and consistency checks on the perceived evidence such as images / tags / density / near infrared by channel. Subsequently, the control unit 30 performs consistency comparison: when the preliminary category pointed to by the perceived evidence is consistent with the user category information, it is identified as a "consistent scenario"; if a conflict or dangerous direct judgment signal occurs, it is classified as a "conflict / risk scenario". In the "consistent scenario", the control unit 30 increases the classification confidence of the category (using a tiered increase and upper limit protection to avoid excessive amplification) and relaxes the sorting threshold at the action level: for example, it allows the triggering condition of "entering the corresponding regular collection grid 51" to be met in advance while maintaining the safety rules unchanged, thereby shortening the waiting time for secondary evidence or review. To prevent risk spillover from relaxed thresholds, this embodiment sets protective constraints: once any channel triggers a danger warning (visual blacklist, excessive density, or spectral hit on a dangerous group), the bias is immediately disabled and switched to a "safety-first" strategy; if consistency is insufficient or evidence quality is substandard, threshold relaxation is not implemented, and the system either enters a pending inspection phase or continues acquisition. The entire fusion and threshold adjustment process is logged (including the reason for the bias, the level of adjustment, and whether protective constraints were triggered) for auditing and subsequent parameter adaptive optimization.

[0047] The technical solution implemented in this embodiment can fully utilize user-side prior knowledge without compromising security, and combine it with robot multimodal perception to form a fusion path of "prior support, mutual evidence verification, and adaptive thresholds": when consistent, it can complete the warehouse entry decision faster and more stably, reducing unnecessary diversion and stagnation; when inconsistent or risky, it can immediately remove biases and enter isolation or review processes to prevent harmful substances from mixing into the ordinary link. This improves the interpretability and pass rate of entry identification, while maintaining strong constraints and rapid closure of anomalies and dangers, resulting in higher sorting accuracy, shorter decision latency, and more consistent work rhythm in actual operation.

[0048] In one specific embodiment, S40 specifically includes: If the garbage bag is identified as a hazardous material, it is sorted into the hazardous material isolation box 53 by the sorting mechanism 40.

[0049] If the target is classified as general waste and its classification confidence is higher than the first threshold, it is sorted into the corresponding regular collection grid 51.

[0050] If the classification confidence is lower than the first threshold but higher than the second threshold, then sort to cell 52.

[0051] In this embodiment, it should be noted that when the system has determined the hazardous materials in S20 / S30, the control unit 30 immediately issues an "isolation and delivery" action sequence to the sorting mechanism 40. The sorting mechanism 40 (such as a flip-up plate / fork / channel valve) first performs a self-check, opens the independent chute leading to the hazardous materials isolation box 53, and locks the isolation box cover after delivery. The isolation box has a built-in smoke detector and (optional) VOC / temperature and humidity monitoring. If an abnormality occurs, it is immediately reported and the main safety strategy is triggered (audible and visual warning, shutdown, or priority dumping at the nearest location). To ensure a closed-loop delivery, the chute is equipped with inlet and outlet material ejection / pressure confirmation. When a bag is jammed, a three-level unblocking process of "reverse shaking - deceleration and re-delivery - manual intervention" is performed, and the event is recorded. After delivery is completed, the control unit 30 updates the "isolation bin capacity vector" and triggers path replanning for priority handling of hazardous materials (linked with S60).

[0052] When the target waste is classified as general waste and its classification confidence level is higher than the first threshold, the control unit 30 selects the disposal trajectory of the corresponding conventional collection grid 51 and controls the action according to the principle of "gate opening time = bag length / conveying speed + safety margin". Before and after disposal, the bag passage is double-confirmed by inlet / outlet sensors. To reduce the risk of mis-disposal, the first threshold is equipped with a hysteresis loop (anti-shake), which can adaptively fine-tune for similar samples from the same household / time period without exceeding the safety limit. After disposal, the capacity monitoring module 60 (weighing / volume / visual) instantly updates the full load of the collection grid 50. If it approaches or reaches the threshold, it will remind S60 to insert a transfer and dumping task. The collection grid 50 has a built-in deodorization / sterilization unit that is linked to the lid opening to ensure environmental hygiene and subsequent maintenance safety.

[0053] When the classification confidence level is between the first and second thresholds, the system routes the sample to inspection grid 52. Simultaneously, an inspection work order is generated (including sample ID, prior category, key image / density / spectral evidence, suspicious causes, and acquisition quality indicators), and a scheduled review is initiated: secondary identification is prioritized within the longest dwell time (which may include multi-angle image re-acquisition, weight / volume re-measurement, NIR re-acquisition, and quality verification), and a manual safety review process is triggered when necessary. Inspection grid 52 has independent masking and visual queue indicators to avoid interference with the regular route; when secondary identification is completed, the control unit 30 automatically performs re-sorting or transfers to isolation based on the new conclusions, and synchronously updates the adaptive statistics of the model and thresholds (without affecting the safety threshold baseline).

[0054] The technical solution implemented in this embodiment can clearly separate three types of results—hazardous direct judgment, high-confidence routine, and suspicious pending inspection—in terms of device operation: hazardous samples are immediately isolated and linked to safety and path priority to reduce accident risks; high-confidence samples are directly put into the warehouse with stable cycle time, reducing unnecessary verification and congestion; suspicious samples are verified periodically and evidence is recorded, improving overall accuracy and interpretability without sacrificing safety. Combined with real-time capacity updates and card unblocking fault tolerance mechanisms, it can effectively avoid mis-dispensing, mixed dispensing, and stagnation caused by single-category fullness, forming an executable, traceable, and optimizable closed loop from entry judgment to warehouse management and subsequent disposal, significantly improving operational continuity, sorting accuracy, and system security.

[0055] Furthermore, S28 also includes: When the probability of a dangerous item does not reach the dangerous item threshold but is within the preset suspicious range, it is sorted to the inspection cell 52 by the sorting mechanism 40 in S40.

[0056] The timed verification mechanism is activated, and a second verification is performed through the identification module 20 within the preset maximum retention time. The second verification includes a second near-infrared spectral acquisition and material composition analysis.

[0057] Subsequently, based on the secondary identification results, the target classification and confidence level are redefined, and S40 and subsequent steps are executed again.

[0058] In this embodiment, it should be noted that when the probability of a hazardous material does not reach the hazardous material threshold but is within a preset suspicious range, it is sorted to the inspection cell 52 by the sorting mechanism 40 in S40. The suspicious range is pre-defined by the system and is dedicated to the safe handling of "boundary samples". After receiving the result of S28, the control unit 30 immediately generates an inspection work order with sample identification, suspicious reasons and first-round evidence snapshots (appearance elements, weight, volume and apparent density, near-infrared key absorption segment, etc.) and issues a sorting action. The inspection cell 52 adopts an independent shielding and anti-odor structure, and bag confirmation sensors are set at the entrance and inside the cell to prevent accidental entry or missed disposal. After successful disposal, the system synchronously updates the capacity vector and inspection queue information, and inserts a "review task" placeholder in the task sequence to avoid forgetting or exceeding the deadline.

[0059] A timed verification mechanism is initiated, and secondary verification is performed by the identification module 20 within the preset maximum dwell time. Secondary verification includes re-acquiring near-infrared spectra and analyzing material composition. The timed verification is scheduled by the control unit 30 and can be executed according to arrival order or risk weight. Before secondary verification, data quality self-checks and light-blocking calibration are performed. If necessary, the illumination angle and acquisition position are changed to reduce posture and background interference. Appearance and text / icon information are re-acquired from multiple angles, and weight and volume are re-measured after settling to correct deformation errors. Near-infrared spectra re-acquisition and material re-judgment are the key tasks, preserving acquisition quality indicators and evidence. If the identification module 20 detects an anomaly during verification (such as smoke detection, VOCs, temperature rise, etc.), the system can interrupt the verification and temporarily transfer the sample to an isolation channel, while simultaneously reporting to the platform.

[0060] Subsequently, based on the secondary identification results, the target classification and confidence level are redefined, and S40 and subsequent steps are executed again. If the review conclusion is hazardous, the control unit 30 immediately instructs the sorting mechanism 40 to transfer the sample to the hazardous materials isolation box 53, and triggers a "hazardous materials priority handling" path replanning in conjunction with S60; if the review conclusion is non-hazardous and the confidence level reaches the sorting threshold, the sample is imported into the corresponding regular collection grid 51 and regular operations resume; if the review is still suspicious or the quality does not meet the standards, the pending inspection status can be maintained according to the safety strategy, and manual safety review or escalation of handling can be triggered. The judgment criteria, action records, and timing status of the entire process are all written into the log for auditing and model optimization.

[0061] The technical solution implemented in this embodiment can provide a time-controlled, evidence-sufficient, and traceable re-judgment channel for "boundary samples" at the entrance: on the one hand, by diverting samples to be inspected and periodically reviewing them, potential hazardous substances are avoided from being directly incorporated into the ordinary process, thus strengthening safety redundancy; on the other hand, by multi-angle re-sampling, re-testing, and near-infrared component-level confirmation, the misjudgment and missed judgment caused by a single detection are significantly reduced. Finally, the review conclusion is seamlessly fed back to the sorting of S40 and the path linkage of S60, realizing a closed loop of identification-diversion-review-disposal, thereby improving the overall classification accuracy, operational continuity, and system safety.

[0062] In one specific embodiment, dynamically updating the job route in S60 includes: When the trigger condition is S61, the hazardous materials priority handling procedure is executed: S61a: The control unit 30 generates a highest-priority transfer task to the hazardous waste disposal chute. In this step, it should be noted that: upon receiving the trigger in S61, the control unit 30 immediately creates a "hazardous materials disposal" type transfer task. Task elements include at least: target recycling point and disposal chute number, segregation box identifier and current loading capacity, trigger source (identification evidence and timestamp), expected arrival time limit, task validity period, execution prerequisites (segregation box lid lock closed, smoke detector normal, battery level ≥ safety threshold), and safety policy (preemptible, non-downgradeable). The task enters the highest priority queue globally, allowing preemption of current non-critical tasks; and resource occupancy verification (disposal chute availability, road access permit) is performed during task creation. If the task fails, it automatically switches to an alternative disposal chute and records the reason.

[0063] S61b: The navigation planning module 70 immediately interrupts the current task sequence, calculates the fastest path, and performs replanning and rerouting. In this step, it should be noted that: the navigation planning module 70 first saves the breakpoint (current location, remaining path, current task ID, execution progress, and control mechanism status), and releases temporary resource locks related to the preempted task; then, based on the goal of "shortest arrival time," it performs path search and evaluation by comprehensively considering travel distance, road restrictions / congestion, construction closures, temporary events, and estimated queuing time at the dock. Planning employs rolling recalculation: the nearest waypoint is issued and continuously updated at fixed intervals or event triggers; if the target dock status changes abruptly (closed / queue surge), the module automatically switches to the next preferred dock and seamlessly reroutes. During the journey, safety strategies (speed limit, obstacle avoidance, low battery fallback return strategy) are activated, and the ETA is continuously transmitted back to the control unit 30 and the platform.

[0064] S61c: Send alarm information to the community management platform and record the entire handling process. In this step, it should be noted that: While issuing the transfer task, the control unit 30 pushes an alarm to the platform, including the robot's ID, location, hazard category, triggering basis (image / spectral / density summary), current isolation box status, estimated arrival time, and handling plan; it can also be linked to SMS / APP notifications for on-duty personnel according to configuration. The system records key nodes: triggering, task issuance, breakpoint saving, start-up, arrival, dumping completion, and reset, all with timestamps and evidence snapshots (cropped images, key spectral points, logs) for auditing and traceability.

[0065] When the trigger condition is S62, the normal relay dumping procedure is executed: S62a: The control unit 30 generates a transfer task to empty the corresponding collection grid 50. In this step, it should be noted that when the trigger condition is S62 (a collection grid 50 reaches its real-time or predicted full load threshold), the control unit 30 selects candidate recycling point outlets matching that category, eliminating closed or unacceptable outlets to form a candidate set. For each candidate outlet, accessibility (shortest path travel time), estimated queuing time, outlet operating hours, remaining battery power, and the cost of connecting with subsequent tasks upon returning to the station are evaluated. Multiple task drafts are generated, and the final task (including the target outlet, suggested execution window, and suggested task insertion location) is selected based on the best result. The task priority is lower than hazardous materials disposal but higher than ordinary collection tasks.

[0066] S62b: The navigation planning module 70 inserts the task into the current sequence and replans and proceeds with the goal of optimizing the overall operation time. In this step, it should be noted that the navigation planning module 70 performs sequence optimization with insertion based on the current set of incomplete tasks. It employs heuristics such as nearest insertion / minimum incremental insertion, combined with local improvements using 2-opt / 3-opt, aiming for optimal overall operation time (balancing timeout penalties and sequence stability, avoiding frequent and significant reordering that could negatively impact user experience). After determining the insertion position, a new route is generated, the nearest waypoint is issued, and rolling execution begins. If the environmental conditions change during execution (congestion, increased queue at the dumping point, decreased battery power), the module can make minor adjustments again according to the same principles. After dumping is complete, the system restores the original collection sequence at the breakpoint and continues execution.

[0067] The technical solution implemented in this embodiment enables rapid closed-loop handling with the highest priority and shortest arrival time when hazardous materials are identified, significantly shortening the risk exposure time. In conventional capacity-triggered scenarios, through plug-in relay dumping and rolling replanning, collection and dumping are coordinated and scheduled, avoiding line-wide stagnation caused by single-type materials reaching full capacity first, reducing waiting and detour times, and improving throughput and operational continuity per unit time. Simultaneously, end-to-end alarm linkage and auditing ensure traceability and replayability of the handling process, enhancing the overall system's security, efficiency, and adaptability to dynamic environments.

[0068] In one specific embodiment, path replanning employs a periodic rolling optimization mechanism: The navigation planning module 70 receives environmental status information again at specific time intervals and calculates the optimal path.

[0069] After each calculation, the robot is driven to travel along the updated path.

[0070] In this embodiment, it should be noted that the navigation planning module 70 receives environmental state information again at specific time intervals and calculates the optimal path. The system sets a periodic timer (e.g., 3-10 seconds, configurable according to the scenario). When the timer expires or an event is triggered (new task, hazardous material trigger, collection grid 50 is about to be full, sudden road congestion, surge in collection point queues, low battery, etc.), it enters a rolling optimization cycle. This cycle first gathers an environmental snapshot: the robot's own position and posture, remaining battery power and estimated range, current task queue and time window constraints, road traffic status (restricted / congested / closed), availability of collection point drop-off points and expected queuing time, and fullness and predicted trend of each collection grid 50. The module checks the freshness and completeness of the snapshot (if it times out or is missing, it rolls back to the most recent reliable data), then generates a set of feasible candidate paths (satisfying restrictions such as restricted areas / time windows / battery power), and evaluates indicators such as arrival time, detour cost, queue risk and sequence stability, producing the current optimal path and a short-term execution "horizon" (e.g., several minutes in the future). To reduce the jitter caused by frequent route changes, the system sets change thresholds and hysteresis: updates are only submitted when the new path has a clear advantage over the current path; at the same time, the previous version is retained for quick rollback.

[0071] After each calculation, the robot is driven to travel along the updated path. The planning results are issued as a set of instructions: "nearest waypoint + speed / behavioral constraints." The lower-level motion controller executes these instructions and continuously reports trajectory deviations and obstacle events. If a deviation, sudden obstacle, or abrupt change in the feed state occurs, the navigation module can immediately trigger a micro-recalculation, fine-tuning the local path without altering the global sequence. To ensure safety and continuity, the execution chain includes: pre-action checks (braking / steering / sensor online), in-journey safety protection (speed limit, obstacle avoidance, emergency braking strategies), a degraded mode for communication anomalies (conservatively driving along the last valid path or waiting safely in place), and a task-level breakpoint-recovery mechanism (automatically returning to the breakpoint to continue data collection after a transfer or tipping). All version switches, reasons, and key metrics are logged for auditing and subsequent parameter optimization.

[0072] The technical solution implemented in this embodiment enables continuous optimal driving in a dynamic community environment while maintaining stability and speed: periodic rolling optimization allows the system to respond promptly to changes in road and recycling point queues, significantly reducing unnecessary detours and waiting; threshold and hysteresis control avoids the jitter caused by frequent detours, maintaining a stable work rhythm; and breakpoint recovery and degradation strategies ensure continuous and safe operation even in abnormal scenarios. This improves on-time performance and throughput per unit time, balancing efficiency, stability, and safety.

[0073] Furthermore, the preset condition in S62 is based on the prediction model that the first collection cell is about to reach full capacity.

[0074] The generation of the prediction model includes: The robot's current task sequence can be obtained through the control unit 30 or the cloud server.

[0075] Predict the type and quantity of waste at each task site based on historical data.

[0076] Predict future loading status based on predicted waste disposal patterns and the current fullness of collection grid 50.

[0077] If the predicted state exceeds the threshold, it is determined that the preset conditions are met.

[0078] In this embodiment, it should be noted that the preset condition of S62 is based on the prediction model's judgment that the first collection grid is about to reach full capacity. The prediction judgment can be performed in the robot's local control unit 30, or the conclusion can be issued by the cloud server after calculation. This embodiment adopts a rolling triggering method: at each fixed cycle or state change (new task added, sudden increase in full capacity), the "whether the threshold will be reached within the foreseeable time window" is re-evaluated as the basis for whether to insert a relay dumping task.

[0079] The robot's current task sequence is obtained through the control unit 30 or a cloud server. The task sequence includes at least: a list of incomplete collection points, the order of visits and estimated arrival times, the geographical location and service window of each point, currently inserted transfer and dumping nodes, and robot position and power constraints. To ensure reproducibility and auditability, each task item has a unique identifier and timestamp; a prediction refresh is immediately triggered when the task sequence changes (insertion, deletion, rearrangement).

[0080] The system predicts the type and quantity of waste at each task location based on historical data. It generates an "estimated disposal volume" for each collection point, providing the quantity and confidence level by category. Historical data is derived from robot operation records and station reconciliation, and its features may include: average disposal volume at that point during similar time periods, differences between weekdays / weekends and holidays, seasonal variations, external factors such as weather and community activities, and user or area disposal habit parameters (learned from long-term data). When single-point samples are sparse, the system backtracks based on group characteristics of similar points in the same area, with a significant uncertainty label to facilitate conservative judgments later.

[0081] The system predicts future loading status based on the predicted waste disposal pattern and the current full load of collection grid 50. It performs capacity simulation along the task sequence timeline: when encountering a collection point, the predicted waste disposal volume for the corresponding category is added to the first collection grid; when encountering a planned transfer and dumping node, collection grid 50 is reset to empty or with a safety margin; simultaneously, travel time, service windows, and arrival offsets caused by queuing are considered to form a "future loading trajectory." To reduce the risk of underestimation, the trajectory calculation adopts a conservative compromise for points with high uncertainty (such as increasing the predicted value and shortening the reserved space), and outputs key milestones (expected when the alarm threshold / full load threshold will be reached).

[0082] If the predicted state exceeds a threshold, the preset conditions are deemed met. Thresholds can be of two types: one is a capacity percentage threshold (e.g., reaching a certain percentage triggers an alarm or requires dumping), and the other is a time window threshold (e.g., reaching a threshold within a set number of minutes is also considered a trigger). If the future loading trajectory shows within the prediction window that it will exceed any threshold, a "triggered" conclusion is returned; if it does not exceed the threshold, a "not triggered" conclusion is returned. The trigger result is accompanied by an evidence package (trigger time, trigger point, key prediction segments, and the conservative adjustments used) for the control unit 30 to record and make decisions.

[0083] The technical solution implemented in this embodiment can identify the risk of "the first collection slot being full" in advance without interrupting regular collection, and insert intermediate dumping nodes through a rolling prediction-evidence triggering-path linkage approach. On the one hand, it avoids the entire line from being shut down and reworked due to a single category being full first, maintaining the continuity of the operation cycle; on the other hand, due to the use of conservative processing and evidence recording, it reduces the overflow risk caused by underestimation, making the dumping timing more controllable and the path adjustment more timely. Combined with the aforementioned hazardous materials priority strategy, the system as a whole achieves proactive scheduling and closed-loop execution of capacity and safety events, significantly improving throughput per unit time, on-time completion rate, and operational safety.

[0084] In an optional embodiment, the generation of the prediction model further includes a learning optimization step: The robot's data recording unit or cloud server records historical data including user identification, delivery time, waste type, and weight.

[0085] Parameters for garbage disposal habits of users or regions are generated based on historical data.

[0086] Habitual parameters are fed back into the prediction model to optimize the accuracy of predicting waste disposal patterns.

[0087] In this embodiment, it should be noted that historical data, including user identification, delivery time, waste category, and weight, is recorded through the robot's data recording unit or a cloud server. The data is stored in a structured format of "task order - delivery details - sensor summary": a unique identifier is generated for each delivery, and the system records the anonymized user or area identifier, delivery time (unified to a fixed time zone), geographical unit / site, category label, weight and number of bags, collection grid number 50, and quality indicators of the identification evidence for that delivery (image quality, density estimation reliability, spectral acquisition quality, etc.). The system performs deduplication, outlier handling (such as automatic marking and verification of abnormal weights), time alignment, and missing data completion on the data; it also retains the version number and source label to ensure subsequent traceability and playback. To reduce on-site workload, the robot first caches the data locally and then uploads it in batches, employing a breakpoint resumption and retry strategy when the network is weak.

[0088] Based on historical data, parameters for user or area waste disposal habits are generated. Habit profiles are constructed in the cloud (or locally) at different granularities: user / building / community, extracting typical features: hourly and weekday / weekend disposal time distribution, proportion and trends of each category, mean and dispersion of single bag weight, peak and off-peak periods, seasonal and holiday effects, and stability of short-term changes. To address the sample sparsity problem, hierarchical aggregation is introduced: for users with insufficient data, group parameters from the same area or similar profile groups are referenced; simultaneously, each set of parameters is labeled with a confidence level and an indicator of "whether significant drift has occurred" (e.g., a significant difference between recent and long-term distributions is marked as drift), allowing for a more conservative strategy during the prediction phase. Profiles are updated regularly (e.g., daily / weekly) or immediately refreshed when a sudden change in distribution is detected.

[0089] The system feeds back habitual parameters to the prediction model to optimize the accuracy of waste disposal pattern predictions. Specifically, when generating the "estimated categories and quantities of task points to be executed," it first provides reasonable initial values ​​and time-period weights based on habitual parameters, then overlays real-time signals (queueing at disposal points, weather, temporary events) for correction. For objects with low reliability or those prone to drift, it increases the safety margin (e.g., using higher empirical upper limits or reserving more storage capacity) to reduce the risk of underestimation. For stable objects, it moderately converges to their historical average to reduce excessive fluctuations. The prediction output also generates uncertainty alerts to guide the setting of trigger thresholds and lead times for transfer and dumping. The system records the deviation between each prediction and the actual situation, performs periodic calibration and parameter adaptation, forming a closed loop of "prediction-execution-evaluation-update."

[0090] The technical solution implemented in this embodiment can continuously accumulate interpretable user / region delivery habits without increasing the complexity of front-end operations, enabling more accurate and robust predictions of the "category and quantity" of each collection point. It provides safety boundaries and fallback paths for scenarios with sparse samples and behavioral drift, reducing false positives and false negatives of "near full" events. It also makes the triggering of transfer and dumping more timely and path adjustments more reasonable, thereby reducing stagnation and rework caused by single-category filling first, and improving overall throughput and on-time completion rate. Simultaneously, because the entire process has quality control and traceability, it facilitates subsequent auditing and model iteration, enhancing the system's robustness and continuous optimization capabilities in the face of uncertain environments.

[0091] In an optional embodiment, the waste recycling robot 100 is also equipped with a wireless transmission module for communicating with community waste stations.

[0092] The navigation planning module 70 receives the availability status and estimated queuing time of different types of drop-off points in real time and uses them as key parameters for path planning.

[0093] In this embodiment, it should be noted that the waste recycling robot 100 is equipped with a wireless transmission module for communication with the community waste station. This module supports one or more of 4G / 5G, Wi-Fi (enterprise network), and low-power IoT (such as LoRa / Ethernet fallback), and exchanges data bidirectionally with the station service via an encrypted link (such as HTTPS / MQTT over TLS). The navigation planning module 70 receives key operating parameters for each type of waste disposal point in real time using a subscription / push method. These parameters include, but are not limited to: availability / disabling status, maintenance / cleaning time window, remaining acceptance capacity or full load alarm, current queuing time and queuing uncertainty, estimated recovery time, entry restrictions (vehicle / time / security level), and station event alarms (temporary shutdown, abnormal congestion). To avoid using outdated data, the module adds a timestamp and time-to-live (TTL) check to each message; information exceeding the TTL is automatically downgraded to "for reference only" and triggers a conservative strategy (prioritizing backup waste disposal points). The link layer provides heartbeat and reconnection capabilities. In the event of a weak network or link failure, the system enables local fallback: temporary parameters are estimated based on recent historical queuing samples and station reliability scores, and the frequency of rerouting is limited to suppress jitter. Hazardous materials disposal messages are set as high-priority channels, allowing for preemptive transmission even under limited bandwidth. For security, two-way authentication and token rotation are employed, and communication and log recording meet auditing requirements. The navigation planning module 70 integrates the aforementioned station information with vehicle capacity vectors, road congestion, task time windows, and remaining battery power into a unified model, using this as a dynamic term in the path cost for rolling replanning. Smoothing and hysteresis thresholds are applied to queuing time, triggering rerouting only when the benefit is significant, balancing efficiency and route stability.

[0094] The technical solution implemented in this embodiment enables the robot to perceive the availability of disposal outlets and queue changes in real time. When handling hazardous materials, it prioritizes the "fastest accessible and acceptable" outlet, significantly reducing waiting time and risk exposure. In conventional relay dumping scenarios, it selects the optimal station based on the combined cost of "distance + estimated queue" and makes rolling adjustments, reducing unnecessary round trips and blind queuing, and improving the single-task completion rate and shift throughput. Even in cases of network instability or temporary station closure, the system can maintain predictable operational rhythm and continuity thanks to TTL constraints, backoff estimation, and rerouting hysteresis. Overall, it achieves high-frequency linkage and robust decision-making between path scheduling and collection points, improving operational efficiency, safety, and adaptability to dynamic environments.

[0095] The embodiments described above are merely illustrative of several implementations of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the appended claims.

Claims

1. A waste recycling method applied to a waste recycling robot, the waste recycling robot comprising a delivery port, an identification module, a control unit, a sorting mechanism, multiple internal collection compartments, a capacity monitoring module, and a navigation planning module; characterized in that, The method includes the following steps: S10. Receive the garbage bag delivered by the user through the delivery port and obtain the category information sent by the user. An automatic sealing device is provided behind the delivery port to seal the garbage bag after identification. S20. Using the identification module, the physical characteristics of the items inside the garbage bag are obtained without damaging the garbage bag, and it is determined whether the garbage bag contains hazardous waste. S30. Using the control unit, the category information and physical characteristics are integrated to determine the target category and classification confidence level of the garbage bag; S40. The sorting mechanism is driven by the control unit to sort the garbage bags into the corresponding target collection compartments according to the target classification and classification confidence level; wherein, the target collection compartment includes a regular collection compartment, a waiting-for-inspection compartment, and a hazardous materials isolation box; the multiple internal collection compartments are equipped with deodorizing and sterilizing devices for odor removal and disinfection of the environment within the collection compartments; the hazardous materials isolation box is equipped with a smoke detector for monitoring fire and triggering a safety warning; S50. The full load of each collection cell is monitored in real time through the capacity monitoring module. S60. The navigation planning module dynamically updates the robot's work route based on at least one of the following triggering conditions: S61 and S20 determine that the garbage bag contains hazardous waste; S62. The real-time or predicted full load of any of the collection cells reaches a preset threshold. The updated work routes are used to guide robots to continue performing collection tasks or to go to recycling points for transfer and dumping.

2. The waste recycling method according to claim 1, characterized in that, S20 includes: S21. The appearance image information of the garbage bag is acquired through the image acquisition device in the recognition module; S22. In the presence of an electronic tag, the information of the electronic tag attached to the garbage bag is read by the radio frequency identification reader in the identification module; S23. The control unit determines the preliminary category and preliminary confidence level of the garbage bag based on the appearance image information and electronic tag information. S24. The control unit determines whether a hazard warning visual feature or a label field matches the hazard blacklist. If so, the garbage bag is determined to contain hazardous waste. S25. If not, obtain the weight of the garbage bag through the weight sensor in the identification module, and obtain its volume through visual ranging based on the image information; S26. The control unit calculates the apparent density of the garbage bag and determines whether the apparent density reaches or exceeds the preset hazardous density criterion. If so, it is determined that the bag contains hazardous waste. S27. If not, near-infrared spectral data are collected and material composition analysis is performed using the near-infrared spectrometer in the identification module to obtain the probability of hazardous materials. S28. The control unit determines whether the probability of the hazardous material reaches or exceeds a preset hazardous material threshold. If so, it is determined that the material contains hazardous waste.

3. The waste recycling method according to claim 2, characterized in that, The fusion in S30 specifically includes: The control unit uses the category information sent by the user as a decision bias factor. When the identified physical characteristics match the user category information, the control unit performs an operation to increase the classification confidence and relaxes the action threshold for sorting the garbage bag into the corresponding regular collection grid based on the increased confidence.

4. The waste recycling method according to claim 3, characterized in that, S40 specifically includes: If the garbage bag is identified as a hazardous material, it is sorted into a hazardous material isolation box by the sorting mechanism. If the target is classified as general waste and its classification confidence is higher than the first threshold, it is sorted into the corresponding regular collection grid. If the classification confidence level is lower than the first threshold but higher than the second threshold, then the item is sorted into the inspection cell.

5. The waste recycling method according to claim 4, characterized in that, Following S28, the following is also included: When the probability of the hazardous material does not reach the hazardous material threshold but is within the preset suspicious range, it is sorted into the inspection cell by the sorting mechanism in S40. A timed verification mechanism is activated, and a second verification is performed through the identification module within the preset maximum retention time. The second verification includes performing near-infrared spectral acquisition and material composition analysis again. Subsequently, based on the secondary identification results, the target classification and confidence level are redefined, and S40 and subsequent steps are executed again.

6. The waste recycling method according to claim 1, characterized in that, The dynamically updated job route in S60 includes: When the trigger condition is S61, the hazardous materials priority handling procedure is executed: S61a. The control unit generates a highest priority transfer task to the hazardous waste disposal port. S61b: The current task sequence is immediately interrupted by the navigation planning module to calculate the fastest path and perform replanning and travel. S61c, Send alarm information to the community management platform and record the entire handling process; When the trigger condition is S62, the normal relay dumping procedure is executed: S62a. A transfer task to clear the corresponding collection cell is generated through the control unit; S62b: Insert the task into the current sequence through the navigation planning module, and replan and proceed with the goal of optimizing the overall operation time.

7. The waste recycling method according to claim 1, characterized in that, The path replanning employs a periodic rolling optimization mechanism: The navigation planning module receives environmental status information again at specific time intervals and calculates the optimal path. After each calculation, the robot is driven to travel along the updated path.

8. The waste recycling method according to claim 7, characterized in that, The preset condition in S62 is based on the prediction model that the first collection cell is about to reach full load. The generation of the prediction model includes: The robot's current task sequence is obtained through the control unit or cloud server; Predict the type and quantity of waste at each task location based on historical data; Predict future loading status based on predicted waste disposal patterns and the current fullness of collection grids; If the predicted state exceeds the threshold, it is determined that the preset conditions are met.

9. The waste recycling method according to claim 1, characterized in that, The generation of the prediction model also includes a learning optimization step: The robot's data recording unit or cloud server records historical data including user identification, delivery time, waste type and weight. Based on the historical data, parameters for user or region waste disposal habits are generated. The habitual parameters are fed back into the prediction model to optimize the accuracy of predicting waste disposal patterns.

10. The waste recycling method according to claim 1, characterized in that, The garbage recycling robot is also equipped with a wireless transmission module for communicating with community garbage stations; The navigation planning module receives the availability status and estimated queuing time of different types of drop-off points in real time and uses them as key parameters for path planning.

Images