Automatic recovery system and method for empty trays of airport luggage turntable

Abstract

The invention discloses an airport luggage turntable empty tray automatic recovery system and method. The system comprises a sensor network, a control module and an execution module. The sensor network detects the state and position of the tray, the control module processes data and controls the execution module, and the execution module completes grabbing, carrying, storage and redistribution of the empty tray. According to the invention, a closed-loop empty tray intelligent recovery system is constructed, and full-automatic identification, grabbing, storage and redistribution of empty trays of a luggage turntable are realized for the first time through three major technologies of multi-sensor fusion sensing, dynamic path planning and real-time motion compensation. Compared with a traditional manual recycling mode, the single-time recycling period can be shortened, and meanwhile manual intervention is reduced. Under the special scene of an airport with high dynamic and high safety requirements, the problem of cooperative control of the mechanical arm and the rotating disc is solved through modular design, and a complete technical chain from physical perception to decision execution is established.

Description

Technical Field

[0001] This invention relates to the technical field of airport baggage handling, and in particular to an automatic empty tray recycling system and method for airport baggage carousels. Background Technology

[0002] In the air transport chain, airport baggage handling systems are a crucial link in ensuring smooth baggage flow and enhancing passenger experience. Baggage carousels are a key component of this system, and the efficiency of empty pallet collection directly impacts passenger experience and overall operational costs. Currently, airports primarily rely on manual methods to collect empty pallets. This traditional method has several drawbacks, such as low efficiency in manual identification and handling, potential congestion in baggage carousel areas during peak flight periods, safety hazards due to staff frequently navigating high-traffic areas, and increased security risks due to misplaced or piled-up empty pallets, which can prolong subsequent baggage loading wait times and affect overall operational efficiency.

[0003] To address these issues, some airports have attempted to introduce semi-automated recycling solutions, such as setting up designated recycling stations to guide the return of empty pallets. However, such semi-automated solutions still struggle to dynamically track and respond in real-time to the rotation needs of empty pallets, failing to achieve true automated management. Meanwhile, current automated material handling systems are widely used in industrial logistics; for example, the integration of robotic arms and sensors can significantly improve handling accuracy and efficiency. However, these technologies have not yet been systematically applied to the specific scenario of airport baggage carousels. The dynamic rotation of the carousel increases the difficulty of compensating for the relative motion between the end-point and the target, making it impossible to accurately identify and handle the random distribution and stacking of baggage baskets in real time. Furthermore, the complexity and high demands of the airport environment place extremely high standards on the reliability and safety of the system, all of which present significant technical challenges that need to be overcome. Summary of the Invention

[0004] The purpose of this invention is to provide an automatic empty tray recycling system and method for airport baggage carousels, so as to solve the problems of low efficiency, high safety hazards, and inability to dynamically track and respond in the existing technology of manual recycling.

[0005] To achieve the above objectives, the solution adopted by the present invention is: An automatic empty tray recycling system for airport baggage carousels includes: a sensor network, a control module, and an execution module; the sensor network, control module, and execution module are interconnected via industrial Ethernet, and the data flow follows the OPCUA standard protocol; The sensor network consists of a distributed sensor array used to detect the occupancy status and position of the tray on the turntable. It includes a photoelectric sensor group, a 3D-ToF camera, and a pressure sensing pad. The photoelectric sensor group is arranged around the circumference of the turntable to detect the presence of the tray; the 3D-ToF camera is installed above the turntable to obtain the three-dimensional coordinates and depth information of the tray; and the pressure sensor is embedded in the surface of the turntable to help determine the load status of the tray. The execution module includes a multi-axis collaborative robotic arm, a storage subsystem, and a transportation mechanism. The robotic arm is used to grasp and transport empty pallets. The storage subsystem uses a high-density automated warehouse for temporary storage of recycled empty pallets. The automated warehouse uses vertical lifting racks that support bidirectional access. The transportation mechanism uses multiple AGV transport vehicles. The AGV transport vehicles are based on laser SLAM navigation and equipped with RFID pallet identification modules to redistribute empty pallets to the storage subsystem. The control module includes a real-time PLC controller, a motion planning module, and a safety monitoring module. The real-time PLC controller is used to process the data stream from the sensor network; the motion planning module is used to run an improved RRT algorithm for path optimization; and the safety monitoring module implements three levels of protection through area gratings and emergency stop circuits.

[0006] Furthermore, the robotic arm is a 6-axis robotic arm, and the end of the robotic arm is equipped with an adaptive electromagnetic-vacuum composite gripper.

[0007] Furthermore, the real-time PLC controller uses a dynamic target detection model based on a lightweight YOLOv5s network to classify the status of the luggage tray. The data stream processed by the real-time PLC controller comes from the photoelectric sensor group, the 3D-ToF camera, and the pressure sensing pad. The photoelectric sensor group outputs switch signals through the digital input module with a sampling period of ≤10ms to detect the presence of the tray. The 3D-ToF camera transmits the 3D point cloud data of the tray via industrial Ethernet using the UDP protocol. The pressure sensing pad is embedded in the surface of the turntable, and the analog signal is converted into a digital signal by the analog input module to determine the load status of the tray.

[0008] Furthermore, data from the photoelectric sensor group, the 3D-ToF camera, and the pressure sensing pad are synchronized with timestamps via the OPCUA standard protocol, with an error of ≤1ms. After data preprocessing, the data is uniformly converted into a format that includes timestamps, sensor IDs, physical quantity values, and confidence levels. Then, the presence of the tray is sensed by multiple sensors through a mechanism of "photoelectric sensor triggering + ToF point cloud verification + pressure assistance". After the photoelectric sensor group is triggered, the ToF point cloud of the 3D-ToF camera confirms the position of the tray, and the pressure data of the pressure sensing pad excludes the load state. When the three are consistent, a retrieval command is triggered, and finally, the processed tray information is output.

[0009] Furthermore, the method for classifying pallet status using a lightweight YOLOv5s network in a real-time PLC controller is as follows: The real-time PLC controller presets the confidence threshold and the load of the empty pallet to be recycled. The input image resolution is adjusted to 640×480 pixels. The confidence Pi for the i-th detection frame is calculated as follows: Pi=σ(w1×IoU+w2×DepthConsist) Where σ is the Sigmoid function; w1=0.6, w2=0.4, w1 and w2 are both weight coefficients; IoU represents the overlap rate between the detection box and the ground truth annotation; DepthConsist is the depth consistency score, calculated through ToF point cloud matching; When Pi exceeds the threshold and the load detected by the pressure sensor is lower than the load of the empty pallet, it is determined to be an "empty pallet to be recycled", triggering the subsequent robotic arm grasping process. When Pi is below the threshold, or the pressure sensor pad detects a load greater than the empty pallet load, it is determined to be a non-empty pallet and the recycling action is not triggered.

[0010] Furthermore, the motion planning module of the control module uses an improved RRT algorithm to optimize the motion path of the robotic arm in real time. Its goal is to avoid dynamic obstacles, minimize the robotic arm's motion time, and ensure a smooth grasping trajectory. Specifically, it includes the following steps: SA, Initialize Path Planning Parameters and Environment Model SA1. Construct a kinematic model of the robotic arm, inputting the joint angle range, maximum joint velocity, and acceleration constraints as hard constraints for path search. SA2. Based on the initial data from the 3D-ToF camera, determine the initial three-dimensional coordinates of the target tray and the rotation parameters of the turntable; SA3. Initialize the improved RRT, with the current end pose of the robotic arm as the root node, set the sampling space boundary as the robotic arm's workspace, and load the coordinates of static obstacles detected by the sensor network. SA4. Set the weights for path evaluation indicators; SB, Real-time Iterative Path Optimization SB1. Periodically acquire sensor network data: The 3D-ToF camera updates the tray's real-time coordinates and the turntable's angular velocity; Sensor networks update the dynamic obstacle status; SB2, Dynamic Obstacle Trajectory Prediction: Based on the turntable angular velocity, calculate the positional offset of the tray and other objects on the turntable within the future Δt. For moving obstacles, a linear prediction model is used to calculate their future position within Δt. SB3, Biased Random Sampling: Sampling points q_rand are randomly generated within the workspace of the robotic arm, with a 60% probability of sampling within a 0.5m radius around the target tray and a 40% probability of uniform sampling across the entire area. Check whether the sampling point q_rand satisfies the kinematic constraints of the robotic arm: the joint angle and speed are within the range set by SA1. If not, resample. SB4, Nearest Node Search and Path Expansion: Search the RRT* tree for the node q_near that is closest to the sample point q_rand; Starting from q_near, generate a new node q_new along the direction from q_near to q_rand, and dynamically adjust the expansion step size d according to the current angular velocity; SB5, Collision Detection: For the path segment from q_near to q_new, check if it collides with a static obstacle; Based on the prediction results of SB2, check whether the path segment collides with the dynamic obstacle in the future Δt. If it collides, discard the path segment and return to SB3 for resampling. SB6, Path Optimization and Reconnection: If q_new passes the collision detection, search for tree node q_near' within its radius r and calculate the path cost from q_near' to q_new; If the cost is lower than the original path, update the parent node to q_near' to achieve local path optimization; SB7, Target Area Determination: If the distance between q_new and the real-time coordinates of the tray is less than the allowable error for grabbing, then backtrack from q_new to the root node to generate the initial path P; SB8, Trajectory Smoothing and Time Optimization: B-spline interpolation is performed on path P to make the rate of change of joint angles continuously differentiable and eliminate abrupt change points; Based on the speed / acceleration constraints of the robotic arm, a time-optimal control algorithm is used to allocate the motion time of each joint and minimize the total path time. SB9, Output and Iteration: The optimized trajectory is decomposed into a sequence of joint angles and sent to the robotic arm controller for execution. Return to SB1 and repeat the iteration based on the next round of sensor data until the grasping action is completed.

[0011] Furthermore, the motion planning module of the control module performs motion compensation on the robotic arm and establishes a transformation matrix from the turntable coordinate system to the robotic arm's base coordinate system. :

[0012] in, Let be the rotation matrix about the Z-axis; ω is the angular velocity of the turntable, which is obtained through an encoder; t is the time it takes for the turntable to rotate. This represents the offset between the origins of the two coordinate systems; The end effector, i.e., the target pose of the robotic arm, is dynamically updated as follows:

[0013] The pose of the turntable at its original position; It refers to the target pose after a rotation over time t.

[0014] Furthermore, the storage subsystem maximizes the storage efficiency of the automated warehouse racking through a dynamic programming algorithm, with racking storage efficiency indicators... :

[0015] in, Let be the accessibility coefficient of the k-th shelf layer; Slot Used k The status of the shelf slots in use; k represents the shelf layer number, from 1 to N; N represents the total number of shelves; This represents the maximum value of the accessibility coefficient across all shelf levels.

[0016] Furthermore, the safety monitoring module of the control module implements three-level protection through area gratings and emergency stop circuits as follows: Level 1 Protection: Area Light Grating Early Warning An infrared grating is installed on the outer edge of the working radius of the robotic arm to form the first protective boundary. When the grating is blocked, the photoelectric sensor group sends a signal to the real-time PLC controller in real time, triggering the first-level response: the movement speed of the robotic arm decreases and the audible and visual alarm device is activated. Level 2 Protection: Monitoring and Suspension of Core Areas A second layer of optical grating or millimeter-wave radar is installed in the direct working area of ​​the robotic arm to form a core protection zone. When personnel or objects are detected entering this zone: The real-time PLC controller immediately cuts off the robotic arm drive signal, causing it to pause at its current position. The relay trips, cutting off some unnecessary power circuits; The audible and visual alarm device switched to a high-frequency warning sound and a red light, and simultaneously sent a warning message to the airport's BHS system; Level 3 protection: Emergency stop circuit triggering and complete shutdown In case of an emergency: The safety contactor in the emergency stop circuit immediately disconnects, cutting off all power sources to the robotic arm. The robotic arm brake is locked to prevent displacement due to gravity or inertia; The system sends an emergency signal to the airport security center, triggers on-site surveillance cameras to focus on the incident area, and simultaneously activates the physical feedback of the on-site emergency stop button.

[0017] The present invention also provides a method for automatically collecting empty trays from airport baggage carousels, based on the aforementioned automatic empty tray collection system, comprising the following steps: S1. Empty baggage carousel tray generation: Baggage is transported to the rotating baggage carousel via the baggage conveyor belt. After passengers take their baggage, empty trays are left on the carousel. S2. Empty Pallet Detection and Judgment: The sensor network detects the pallets on the turntable. When the pressure sensor pad detects that the pallet load is less than the preset empty pallet load, and the lightweight YOLOv5s network identifies the pallet status as an empty pallet, the pallet is determined to be an empty pallet to be recycled. S3, Empty Pallet Recycling: The real-time PLC controller uses the 3D coordinates of the empty pallet to be recycled and the dynamic position of the turntable obtained by the 3D-ToF camera to plan the motion path of the robotic arm through the improved RRT* algorithm. The robotic arm grabs the empty pallet and transports it to a high-density automated warehouse for storage. S4. Empty Pallet Redistribution: The AGV transport fleet uses laser SLAM navigation to the automated warehouse, retrieves empty pallets through the RFID pallet identification module, transports them to the baggage conveyor belt starting point, and puts the empty pallets back into the baggage conveyor belt starting point for reuse.

[0018] By adopting the above solution, the automated empty tray recycling system and method for airport baggage carousels provided by this invention constructs a closed-loop intelligent recycling system. Through multi-sensor fusion sensing, dynamic path planning, and real-time motion compensation, it achieves, for the first time, fully automated identification, grasping, storage, and redistribution of empty baggage carousels. Compared to traditional manual recycling methods, this invention can shorten the single recycling cycle and reduce human intervention. In the special scenario of airports, characterized by high dynamics and high security requirements, this invention solves the problem of coordinated control between the robotic arm and the rotating carousel through modular design and establishes a complete technology chain from physical perception to decision execution. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the overall architecture of the present invention and the airport baggage handling system.

[0020] Figure 2 This is a schematic diagram of the empty tray recycling system for airport baggage carousels according to the present invention. Detailed Implementation

[0021] To further explain the technical solution of the present invention, the present invention will be described in detail below through specific embodiments. For example... Figure 1 and Figure 2 As shown, this invention discloses an automatic empty tray recycling system for airport baggage carousels. Its integration with the Airport Baggage Handling System (BHS) is achieved through a RESTful API, enabling real-time acquisition of operational data such as flight numbers and carousel rotation speeds. The baggage handling system includes a conveyor belt, a baggage carousel, and a baggage storage area. The baggage conveyor belt transports baggage from the check-in area to the baggage carousel, which serves as a rotating platform for passengers to retrieve their baggage. The baggage storage area is used to temporarily store baggage that has not been collected in a timely manner.

[0022] The automatic empty tray recycling system for airport baggage carousels of the present invention includes a sensor network, a control module, and an execution module. The sensor network, control module, and execution module are interconnected via industrial Ethernet, and the data flow follows the OPCUA standard protocol.

[0023] The sensor network consists of a distributed sensor array used to detect the occupancy status and position of the trays on the turntable. It includes a photoelectric sensor group, a 3D-ToF camera, and a pressure sensor pad. The photoelectric sensor group is arranged around the circumference of the turntable to detect the presence of the trays; the 3D-ToF camera is installed above the turntable to obtain the three-dimensional coordinates of the trays; and the pressure sensor is embedded in the surface of the turntable to help determine the load status of the trays.

[0024] The execution module includes a multi-axis collaborative robotic arm, a storage subsystem, and a transportation mechanism.

[0025] In this embodiment, the robotic arm is a 6-axis robotic arm, and the end of the robotic arm is equipped with an adaptive electromagnetic-vacuum composite gripper for performing the gripping and handling of empty pallets.

[0026] The control module includes a real-time PLC controller, a motion planning module, and a safety monitoring module. The real-time PLC controller is used to process the data stream of the sensor network with a sampling period of ≤10ms. The motion planning module is used to run an improved RRT algorithm for path optimization. The safety monitoring module implements three levels of protection through area gratings and emergency stop circuits.

[0027] The real-time PLC controller employs a dynamic target detection model based on a lightweight YOLOv5s network to classify the status of luggage trays. The data stream processed by the real-time PLC controller primarily originates from photoelectric sensor arrays, a 3D-ToF camera, and a pressure sensing pad. The photoelectric sensor array is arranged circumferentially around the turntable, outputting switch signals through a digital input module with a sampling period of ≤10ms to detect the presence of the tray. The 3D-ToF camera is mounted above the turntable, transmitting the tray's 3D point cloud data (X, Y, Z coordinates and depth information) via industrial Ethernet using the UDP protocol. The pressure sensing pad is embedded in the turntable surface, converting analog signals to digital signals via an analog input module to determine the tray's load status. These three types of sensor data are time-stamped and synchronized (error ≤1ms) using the OPCUA standard protocol. After data preprocessing, the data is uniformly converted into a format containing timestamps, sensor IDs, physical quantity values, and confidence levels. A mechanism of "photoelectric sensor triggering + ToF point cloud verification + pressure assistance" is then used to fuse the multiple sensor data to perceive the presence of the tray. After the photoelectric sensor group is triggered, the ToF point cloud confirms the pallet position, the pressure data excludes the load state, and when the three are consistent, the retrieval command is triggered. Finally, the processed information such as the pallet's three-dimensional coordinates, movement speed, and grasping priority is output.

[0028] The method for classifying pallet status using a lightweight YOLOv5s network in a real-time PLC controller is as follows: The real-time PLC controller has a preset confidence threshold (e.g., 0.5) for empty pallets to be recycled with a load capacity of ≤5kg. The input image resolution is adjusted to 640×480 pixels. The confidence Pi for the i-th detection box (i.e., the tray on the turntable) is calculated as: Pi=σ(w1×IoU+w2×DepthConsist) Where σ is the Sigmoid function; w1=0.6, w2=0.4, w1 and w2 are both weight coefficients; IoU represents the overlap rate between the detection box and the ground truth annotation; DepthConsist is the depth consistency score, calculated by ToF point cloud matching.

[0029] The motion planning module of the control module uses an improved RRT algorithm to optimize the robotic arm's motion path in real time. Its goal is to avoid dynamic obstacles, minimize the robotic arm's motion time, and ensure a smooth grasping trajectory. Specifically, it includes the following steps: S1. Initialize path planning parameters and environment model S1.1 Construct the kinematic model of the robotic arm, inputting the joint angle range, maximum joint velocity, and acceleration constraints as hard constraints for path search; S1.2 Based on the initial data from the 3D-ToF camera, determine the initial three-dimensional coordinates of the target tray and the rotation parameters of the turntable (initial angular velocity, rotation center coordinates). S1.3 Initialize the improved RRT, with the current end pose of the robotic arm as the root node, set the sampling space boundary as the robotic arm's workspace, and load the coordinates of static obstacles detected by the sensor network; S1.4 Set the weights of the path evaluation indicators.

[0030] S2. Real-time Iterative Path Optimization S2.1 Acquire sensor data periodically (period T=50ms): The 3D-ToF camera updates the real-time coordinates of the tray and the angular velocity of the turntable (calculated by the coordinate difference between adjacent moments). Sensor networks update dynamic obstacle states: output passenger position and movement speed vectors, and bounding box coordinates of other dynamic devices (such as temporary obstacles).

[0031] S2.2 Dynamic Obstacle Trajectory Prediction: Based on the turntable angular velocity, calculate the positional offset of the tray and other objects on the turntable within the future Δt. For moving obstacles such as passengers, a linear prediction model is used to calculate their future position within Δt.

[0032] S2.3 Biased Random Sampling: Sampling points q_rand are randomly generated within the workspace of the robotic arm, with a 60% probability of sampling within a 0.5m radius around the target tray and a 40% probability of uniform sampling across the entire area. Check whether the sampling point q_rand meets the kinematic constraints of the robotic arm: the joint angle and speed are within the range set in S1.1. If not, resample.

[0033] S2.4 Nearest Node Search and Path Expansion: Search the RRT* tree for the node q_near that is closest to the sample point q_rand; Starting from q_near, generate a new node q_new along the direction from q_near to q_rand. The expansion step size d is dynamically adjusted according to the current angular velocity (the larger the angular velocity, the smaller d).

[0034] S2.5 Collision Detection: For the path segment from q_near to q_new, check whether it collides with a static obstacle (Euclidean distance < safety threshold). Based on the prediction results of S2.2, check whether the path segment collides with dynamic obstacles (including the area around the tray) in the future Δt. If a collision occurs, discard the path segment and return to S2.3 for resampling.

[0035] S2.6 Path Optimization and Reconnection: If q_new passes the collision detection, search for tree node q_near' within its radius r, and calculate the path cost (including time and smoothness metrics) from q_near' to q_new. If the cost is lower than the original path, update the parent node to 'q_near' to achieve local path optimization.

[0036] S2.7 Target Area Determination: If the distance between q_new and the real-time coordinates of the tray is ≤0.1m (allowable error for grabbing), then backtrack from q_new to the root node to generate the initial path P.

[0037] S2.8 Trajectory Smoothing and Time Optimization: B-spline interpolation is performed on path P to make the rate of change of joint angles continuously differentiable and eliminate abrupt change points; Based on the speed / acceleration constraints of the robotic arm, a time-optimal control algorithm is used to allocate the motion time of each joint and minimize the total path time.

[0038] S2.9 Output and Iteration: The optimized trajectory is decomposed into a sequence of joint angles and sent to the robotic arm controller for execution. Return to S2.1 and repeat the iteration based on the next round of sensor data until the grasping action is completed.

[0039] This invention utilizes the 3D coordinates of the tray and the dynamic position of the turntable provided by a 3D-ToF camera. Real-time path optimization of the robotic arm is achieved by using sensor networks to detect obstacles on the turntable (such as passengers or other equipment) and kinematic constraints (joint angles, speed limits) to enable real-time path optimization.

[0040] The motion planning module of the control module performs motion compensation for the robotic arm and establishes the transformation matrix from the turntable coordinate system to the robotic arm base coordinate system. :

[0041] in, Let be the rotation matrix about the Z-axis; ω is the angular velocity of the turntable, which is obtained through an encoder; t is the time it takes for the turntable to rotate. This represents the offset between the origins of the two coordinate systems; The target pose of the end effector, i.e., the end gripper of the robotic arm, needs to be dynamically updated to:

[0042] The pose of the turntable at its original position; It refers to the target pose after a rotation over time t.

[0043] The motion planning module of this invention combines turntable encoder data with the inverse kinematics of the robotic arm. By calculating the positional changes caused by the rotation of the turntable in real time, it compensates for the relative motion of the robotic arm during grasping, achieves motion synchronization, and ensures accurate positioning.

[0044] This embodiment of the storage subsystem employs a high-density automated warehouse for temporary storage of recycled empty pallets. The automated warehouse utilizes vertical lift racks, supporting bidirectional access. The storage subsystem maximizes the storage efficiency of the automated warehouse racks through a dynamic programming algorithm. Define shelf storage efficiency metrics :

[0045] in, Let be the accessibility coefficient of the k-th shelf. In this embodiment, the top shelf = 1.0, and it decreases by 0.2 for each subsequent shelf. Slot Used k The status of the shelf slots in use; k represents the shelf layer number, from 1 to N; N represents the total number of shelves; This represents the maximum accessibility coefficient across all shelf levels, with the maximum value being 1.0 for the top shelf.

[0046] The transportation agency uses multiple AGV transport vehicles, which are based on laser SLAM navigation and equipped with RFID pallet identification modules to redistribute empty pallets to the storage subsystem.

[0047] The safety monitoring module of the control module achieves three levels of protection through area light grids and emergency stop circuits. It is a layered security mechanism that combines hardware detection and logic control, and the specific implementation method is as follows: Level 1 Protection: Area Light Grating Early Warning An infrared grating is installed along the outer edge of the robotic arm's working radius to form the first line of defense. When the grating is blocked (e.g., by a passenger approaching or accidentally entering), the sensor sends a signal to the real-time PLC controller, triggering a first-level response: the robotic arm's movement speed decreases, and an audible and visual alarm is activated (a yellow warning light flashes and a low-frequency warning sound is heard), prompting personnel to stay away.

[0048] Level 2 Protection: Monitoring and Suspension of Core Areas A second optical grating or millimeter-wave radar is installed in the direct working area of ​​the robotic arm (radius 0.5-1 meter) to form a core protection zone. When personnel or objects are detected entering this zone: The real-time PLC controller immediately cuts off the robotic arm drive signal, pausing it at its current position (maintaining the clamping state to prevent the empty basket from falling). The relay trips, cutting off some unnecessary power circuits (such as the vacuum / electromagnetic adsorption power of the end clamp). The audible and visual alarm device switched to a high-frequency warning sound and a red light, and simultaneously sent a warning message to the airport's BHS system.

[0049] Level 3 protection: Emergency stop circuit triggering and complete shutdown In the event of the following emergency situations (such as direct contact between personnel and the robotic arm, continuous obstruction of the light grid for more than 3 seconds, or overtravel of the robotic arm joints): The safety contactor in the emergency stop circuit immediately disconnects, cutting off all power sources for the robotic arm (including servo motors, gripper drives, etc.). The robotic arm brake is locked to prevent displacement due to gravity or inertia; The system sends an emergency signal to the airport security center, triggers on-site surveillance cameras to focus on the incident area, and simultaneously triggers physical feedback from the on-site emergency stop button (such as the button popping out and locking).

[0050] Level 3 protection employs a progressive response of "early warning - pause - emergency stop," combined with spatial layer detection of regional gratings and hard-wired protection of the emergency stop circuit, to achieve full-process safety control from early warning to emergency braking, ensuring operational safety in human-machine interaction scenarios.

[0051] The present invention also provides a method for automatically collecting empty trays from airport baggage carousels, based on the aforementioned automatic empty tray collection system, comprising the following steps: S1. Empty baggage carousel tray generation: Baggage is transported to the rotating baggage carousel via a baggage conveyor belt. After passengers take their baggage, empty trays are left on the carousel. S2. Empty Pallet Detection and Judgment: The sensor network detects the pallets on the turntable. When the pressure sensor pad detects that the pallet load is ≤5kg and the lightweight YOLOv5s network identifies the pallet status as empty, the pallet is determined to be an empty pallet to be recycled. S3. Empty Pallet Recycling: The real-time PLC controller uses the 3D coordinates of the empty pallet to be recycled and the dynamic position of the turntable obtained by the 3D-ToF camera to plan the motion path of the robotic arm through the improved RRT* algorithm, and controls the adaptive electromagnetic-vacuum composite gripper at the end of the 6-axis collaborative robotic arm to grab the empty pallet and transport it to a high-density three-dimensional warehouse for storage. S4. Empty Pallet Redistribution: The AGV transport fleet uses laser SLAM navigation to the automated warehouse, retrieves empty pallets through the RFID pallet identification module, transports them to the baggage conveyor belt starting point, and puts the empty pallets back into the baggage conveyor belt starting point for reuse.

[0052] This invention relates to an automated empty tray recycling system and method for airport baggage carousels. Through dynamic compensation technology—combining carousel encoder data with the inverse kinematics of the robotic arm to achieve motion synchronization—it overcomes the complex relative motion compensation problem between the robotic arm's end effector and the target caused by the dynamic rotation of the traditional carousel. Furthermore, through multimodal perception—integrating photoelectric sensors, ToF cameras, and pressure detection—it solves the problem of misjudgment by a single sensor under complex lighting conditions. Finally, through a tiered safety strategy, it utilizes a real-time PLC controller to monitor the robotic arm's working radius in real time, ensuring safe human-machine interaction and meeting the high reliability and safety requirements of the airport environment.

[0053] The above embodiments and figures are not intended to limit the product form and style of this invention. Any appropriate changes or modifications made by those skilled in the art should be considered as not departing from the scope of this patent.

Claims

1. An automatic empty tray recycling system for airport baggage carousels, characterized in that, include: Sensor network, control module, and actuation module; the sensor network, control module, and actuation module are interconnected via industrial Ethernet, and the data flow follows the OPCUA standard protocol; The sensor network consists of a distributed sensor array used to detect the occupancy status and position of the trays on the turntable. It includes a photoelectric sensor group, a 3D-ToF camera, and a pressure sensing pad. The photoelectric sensor group is arranged around the circumference of the turntable to detect the presence of the trays. A 3D-ToF camera is mounted above the turntable to acquire the tray's three-dimensional coordinates and depth information; Pressure sensors are embedded in the surface of the turntable to help determine the load status of the pallet; The execution module includes a multi-axis collaborative robotic arm, a storage subsystem, and a transportation mechanism. The robotic arm is used to grasp and transport empty pallets. The storage subsystem uses a high-density automated warehouse for temporary storage of recycled empty pallets. The automated warehouse uses vertical lifting racks and supports bidirectional access. The transportation mechanism uses multiple AGV transport vehicles. The AGV transport vehicles are based on laser SLAM navigation and equipped with RFID pallet identification modules to redistribute empty pallets to the storage subsystem. The control module includes a real-time PLC controller, a motion planning module, and a safety monitoring module. The real-time PLC controller is used to process the data stream from the sensor network. The motion planning module is used to run an improved RRT algorithm for path optimization; the safety monitoring module implements three levels of protection through area gratings and emergency stop circuits.

2. The automatic empty tray recycling system for airport baggage carousels as described in claim 1, characterized in that: The robotic arm is a 6-axis robotic arm, and its end effector is equipped with an adaptive electromagnetic-vacuum composite gripper.

3. The automatic empty tray recycling system for airport baggage carousels as described in claim 1, characterized in that: The real-time PLC controller uses a dynamic target detection model based on a lightweight YOLOv5s network to classify the status of luggage trays. The data stream processed by the real-time PLC controller comes from a photoelectric sensor group, a 3D-ToF camera, and a pressure sensing pad. The photoelectric sensor group outputs switch signals through a digital input module with a sampling period of ≤10ms to detect the presence of the tray. The 3D-ToF camera transmits the 3D point cloud data of the tray via industrial Ethernet using the UDP protocol. The pressure sensing pad is embedded in the surface of the turntable, and the analog signal is converted into a digital signal by the analog input module to determine the load status of the tray.

4. The automatic empty tray recycling system for airport baggage carousels as described in claim 3, characterized in that: Data from three types of sensors—the photoelectric sensor group, the 3D-ToF camera, and the pressure sensing pad—are synchronized via timestamps using the OPCUA standard protocol, with an error ≤1ms. After data preprocessing, the data is uniformly converted into a format containing timestamps, sensor IDs, physical quantity values, and confidence levels. Then, the presence of the tray is sensed by fusing multiple sensors through a mechanism of "photoelectric sensor triggering + ToF point cloud verification + pressure assistance." After the photoelectric sensor group is triggered, the ToF point cloud data from the 3D-ToF camera confirms the tray position, and the pressure data from the pressure sensing pad excludes the load state. When all three are consistent, a retrieval command is triggered, and finally, the processed tray information is output.

5. An automatic empty tray recycling system for airport baggage carousels as described in claim 3, characterized in that: The method for classifying pallet status using a lightweight YOLOv5s network in a real-time PLC controller is as follows: The real-time PLC controller presets the confidence threshold and the load of the empty pallet to be recycled. The input image resolution is adjusted to 640×480 pixels. The confidence Pi for the i-th detection frame is calculated as follows: Pi=σ(w1×IoU+w2×DepthConsist) Where σ is the Sigmoid function; w1=0.6, w2=0.4, w1 and w2 are both weight coefficients; IoU represents the overlap rate between the detection box and the ground truth annotation; DepthConsist is the depth consistency score, calculated through ToF point cloud matching; When Pi exceeds the threshold and the load detected by the pressure sensor is lower than the load of the empty pallet, it is determined to be an "empty pallet to be recycled", triggering the subsequent robotic arm grasping process; When Pi is below the threshold, or the pressure sensor detects a load greater than the empty pallet load, it is determined to be a "non-empty pallet" and the recycling action is not triggered.

6. The automatic empty tray recycling system for airport baggage carousels as described in claim 1, characterized in that: The motion planning module of the control module uses an improved RRT algorithm to optimize the motion path of the robotic arm in real time. Its goal is to avoid dynamic obstacles, minimize the robotic arm's motion time, and ensure a smooth grasping trajectory. Specifically, it includes the following steps: SA, Initialize Path Planning Parameters and Environment Model SA1. Construct a kinematic model of the robotic arm, inputting the joint angle range, maximum joint velocity, and acceleration constraints as hard constraints for path search. SA2. Based on the initial data from the 3D-ToF camera, determine the initial three-dimensional coordinates of the target tray and the rotation parameters of the turntable; SA3. Initialize the improved RRT, with the current end pose of the robotic arm as the root node, set the sampling space boundary as the robotic arm's workspace, and load the coordinates of static obstacles detected by the sensor network. SA4. Set the weights for path evaluation indicators; SB, Real-time Iterative Path Optimization SB1. Periodically acquire sensor network data: The 3D-ToF camera updates the tray's real-time coordinates and the turntable's angular velocity; Sensor networks update the dynamic obstacle status; SB2, Dynamic Obstacle Trajectory Prediction: Based on the turntable angular velocity, calculate the positional offset of the tray and other objects on the turntable within the future Δt. For moving obstacles, a linear prediction model is used to calculate their future position within Δt. SB3, Biased Random Sampling: Sampling points q_rand are randomly generated within the workspace of the robotic arm, with a 60% probability of sampling within a 0.5m radius around the target tray and a 40% probability of uniform sampling across the entire area. Check whether the sampling point q_rand satisfies the kinematic constraints of the robotic arm: the joint angle and speed are within the range set by SA1. If not, resample. SB4, Nearest Node Search and Path Expansion: Search the RRT* tree for the node q_near that is closest to the sample point q_rand; Starting from q_near, generate a new node q_new along the direction from q_near to q_rand, and dynamically adjust the expansion step size d according to the current angular velocity; SB5, Collision Detection: For the path segment from q_near to q_new, check if it collides with a static obstacle; Based on the prediction results of SB2, check whether the path segment collides with the dynamic obstacle in the future Δt. If it collides, discard the path segment and return to SB3 for resampling. SB6, Path Optimization and Reconnection: If q_new passes the collision detection, search for tree node q_near' within its radius r and calculate the path cost from q_near' to q_new; If the cost is lower than the original path, update the parent node to q_near' to achieve local path optimization; SB7, Target Area Determination: If the distance between q_new and the real-time coordinates of the tray is less than the allowable error for grabbing, then backtrack from q_new to the root node to generate the initial path P; SB8, Trajectory Smoothing and Time Optimization: B-spline interpolation is performed on path P to make the rate of change of joint angles continuously differentiable and eliminate abrupt change points; Based on the speed / acceleration constraints of the robotic arm, a time-optimal control algorithm is used to allocate the motion time of each joint and minimize the total path time. SB9, Output and Iteration: The optimized trajectory is decomposed into a sequence of joint angles and sent to the robotic arm controller for execution. Return to SB1 and repeat the iteration based on the next round of sensor data until the grasping action is completed.

7. An automatic empty tray recycling system for airport baggage carousels as described in claim 1, characterized in that: The motion planning module of the control module performs motion compensation for the robotic arm and establishes the transformation matrix from the turntable coordinate system to the robotic arm base coordinate system. : in, Let be the rotation matrix about the Z-axis; ω is the angular velocity of the turntable, which is obtained through an encoder; t is the time it takes for the turntable to rotate. This represents the offset between the origins of the two coordinate systems; The end effector, i.e., the target pose of the robotic arm, is dynamically updated as follows: The pose of the turntable at its original position; It refers to the target pose after a rotation over time t.

8. An automatic empty tray recycling system for airport baggage carousels as described in claim 1, characterized in that: The storage subsystem maximizes the storage efficiency of the automated warehouse racking through a dynamic programming algorithm. The racking storage efficiency index... : in, Let be the accessibility coefficient of the k-th shelf layer; Slot Used k The status of the shelf slots in use; k represents the shelf layer number, from 1 to N; N represents the total number of shelves; This represents the maximum value of the accessibility coefficient across all shelf levels.

9. An automatic empty tray recycling system for airport baggage carousels as described in claim 1, characterized in that: The safety monitoring module of the control module implements a three-level protection method through area gratings and emergency stop circuits as follows: Level 1 Protection: Area Light Grating Early Warning An infrared grating is installed on the outer edge of the working radius of the robotic arm to form the first protective boundary. When the grating is blocked, the photoelectric sensor group sends a signal to the real-time PLC controller in real time, triggering the first-level response: the movement speed of the robotic arm decreases and the audible and visual alarm device is activated. Level 2 Protection: Monitoring and Suspension of Core Areas A second layer of optical grating or millimeter-wave radar is installed in the direct working area of ​​the robotic arm to form a core protection zone. When personnel or objects are detected entering this zone: The real-time PLC controller immediately cuts off the robotic arm drive signal, causing it to pause at its current position. The relay trips, cutting off some unnecessary power circuits; The audible and visual alarm device switched to a high-frequency warning sound and a red light, and simultaneously sent a warning message to the airport's BHS system; Level 3 protection: Emergency stop circuit triggering and complete shutdown In case of an emergency: The safety contactor in the emergency stop circuit immediately disconnects, cutting off all power sources to the robotic arm. The robotic arm brake is locked to prevent displacement due to gravity or inertia; The system sends an emergency signal to the airport security center, triggers on-site surveillance cameras to focus on the incident area, and simultaneously activates the physical feedback of the on-site emergency stop button.

10. A method for automatically collecting empty trays from an airport baggage carousel, characterized in that, The automatic empty tray recycling system for airport baggage carousels according to any one of claims 1 to 9 includes the following steps: S1. Empty baggage carousel tray generation: Baggage is transported to the rotating baggage carousel via the baggage conveyor belt. After passengers take their baggage, empty trays are left on the carousel. S2. Empty Pallet Detection and Judgment: The sensor network detects the pallets on the turntable. When the pressure sensor pad detects that the pallet load is less than the preset empty pallet load, and the lightweight YOLOv5s network identifies the pallet status as an empty pallet, the pallet is determined to be an empty pallet to be recycled. S3, Empty Pallet Recycling: The real-time PLC controller uses the 3D coordinates of the empty pallet to be recycled and the dynamic position of the turntable obtained by the 3D-ToF camera to plan the motion path of the robotic arm through the improved RRT* algorithm. The robotic arm grabs the empty pallet and transports it to a high-density automated warehouse for storage. S4. Empty Pallet Redistribution: The AGV transport fleet uses laser SLAM navigation to the automated warehouse, retrieves empty pallets through the RFID pallet identification module, transports them to the baggage conveyor belt starting point, and puts the empty pallets back into the baggage conveyor belt starting point for reuse.

Images