Cart autonomous homing and flexible stacking method, system, apparatus, and storage medium

Abstract

The application provides a trolley autonomous collection and flexible stacking method, system, device and storage medium, relates to the technical field of mobile operation robots, and the method comprises the following steps: fusing multi-modal sensing data to identify a trolley target and a pose; a mechanical arm performs grabbing and compliant docking to obtain a grabbed trolley; a trajectory is planned under the conditions of satisfying a narrow channel passing constraint and a minimum turning radius constraint, a driving speed is automatically adjusted according to crowd risk assessment, and the trolley is transported to a stacking area; multi-trolley continuous stacking is performed: a generalized coordinate comprising a mobile chassis pose, a relative nesting depth of each trolley and a yaw angle is defined, a multi-trolley coupling dynamics model is constructed, docking geometric constraints and anti-collision constraints are established, and cooperative control instructions are generated; the chassis main traction is controlled, and the mechanical arm impedance fine adjustment is controlled to complete nesting; and the stacking queue is transported to a recovery point to complete arrangement and parking. The application improves the passing safety and overall operation efficiency of the robot autonomous collection and stacking trolley.

Description

Technical Field

[0001] This invention relates to the field of mobile operation robot technology, and more specifically, to a method, system, device, and storage medium for autonomous collection and flexible stacking of handcarts. Background Technology

[0002] Currently, trolleys, as a high-frequency public service facility, face the persistent problem of scattered distribution and a shortage of collection points in areas such as baggage claim areas on arrival levels, check-in areas on departure levels, boarding gate corridors, and commercial corridors at airports. Manually pushing trolleys is not only labor-intensive and costly, but also difficult to replenish in a timely manner at night or during peak hours. Furthermore, pushing trolleys through narrow passages, serpentine barriers, and dense crowds poses a significant risk of collisions with passengers, luggage, or facilities, creating public safety hazards. Therefore, there is a need for a robotic system and control method capable of autonomously identifying, grasping, stacking, and transporting trolleys in the complex and dynamic environment of airports. This would reduce labor costs, improve collection efficiency, and meet the safety and social distancing requirements of public spaces.

[0003] While some automated or robotic solutions have been developed for trolley retrieval, existing solutions remain ineffective in the complex realities of airport environments. Specifically, due to challenges such as dynamic crowds, narrow passageways, interference from similar objects, and complex physical interactions during multi-trolley stacking, existing solutions fall short in areas like target recognition, path maneuverability, stable multi-trolley stacking, and safe human-machine collaboration, failing to achieve stable and reliable autonomous operation. Therefore, trolley retrieval remains highly reliant on manual labor, making it difficult to meet the actual needs of intelligent airport operations. Summary of the Invention

[0004] The problem solved by this invention is how to achieve autonomous collection and flexible stacking of handcarts in the complex and dynamic environment of an airport.

[0005] To address the aforementioned problems, this invention provides a method, system, device, and storage medium for the autonomous collection and flexible stacking of handcarts.

[0006] In a first aspect, the present invention provides a method for autonomous collection and flexible stacking of handcarts, applied to a robot system, the robot system comprising a mobile chassis and a robotic arm, the method comprising: Acquire multimodal sensing data and fuse and identify the multimodal sensing data to obtain the handcart target and the pose information of the handcart target; Based on the pose information, the robotic arm performs grasping and compliant docking on the handcart target to obtain the grasped handcart and a clamping status confirmation signal. Based on the current environmental perception information and the status of the grabbed handcarts, a driving trajectory is planned under the conditions of satisfying the narrow passage passage constraint and the minimum turning radius constraint. The mobile chassis is controlled to track the driving trajectory. At the same time, the driving speed of the mobile chassis is automatically adjusted according to the risk assessment results of the dynamic crowd, and the grabbed handcarts are transported to the stacking area to obtain the current stacking queue status. Based on the already grabbed trolleys and the current stacking queue state, perform continuous stacking of multiple trolleys: Define a generalized coordinate system, which includes the pose of the mobile chassis, the relative nesting depth between each trolley, and the relative yaw angle between each trolley. A multi-vehicle coupled dynamics model is constructed based on the generalized coordinates, and docking geometric constraints and anti-collision constraints are established. Based on the multi-vehicle coupled dynamics model, the geometric constraints, and the collision avoidance constraints, collaborative control commands for the mobile chassis and the robotic arm are generated. According to the cooperative control command, the mobile chassis is controlled to perform the main traction motion, and the robotic arm is controlled to perform anisotropic impedance control to perform attitude fine-tuning, so that the grabbed handcart is nested into the tail of the stacking queue to obtain the updated stacking queue. The updated stacked queue is moved to the target recycling point to complete the queue arrangement and parking of the handcarts.

[0007] Optionally, the step of acquiring multimodal perception data and fusing and recognizing the multimodal perception data to obtain the handcart target and its pose information includes: Acquire multimodal sensing data from at least two types of sensors, including a surround-view camera, a 3D LiDAR, or a depth camera; The multimodal sensing data is time-synchronized and extrinsic parameters are calibrated to obtain aligned sensing data with a unified time reference and a unified spatial coordinate system. The aligned perception data is input into the target detection model to obtain the detection box and key points of the handcart target; Based on the detection box and the key points, and combined with the 3D point cloud data in the aligned perception data, the 3D position and spatial pose of the handcart target are obtained, and the 3D position and spatial pose are used as the pose information of the handcart target.

[0008] Optionally, the step of performing grasping and compliant docking of the handcart target by the robotic arm based on the pose information, and obtaining a grasped handcart and a clamping status confirmation signal, includes: Based on the pose information, an approach trajectory for the robotic arm is generated, and the robotic arm is controlled to move to the initial docking position of the handcart target based on the approach trajectory. After the robotic arm reaches the initial docking position, the robot system enters the visual servoing stage, using the key points of the handcart target as feedback to correct the pose of the robotic arm end in real time, so that the guide structure of the robotic arm end is aligned with the docking entrance of the handcart target. After the end of the robotic arm contacts the target trolley, the robot system switches to impedance control mode to correct the pose of the end of the robotic arm to compensate for the docking pose deviation. Based on the online estimated coefficient of friction of the contact surface and the load mass of the handcart target, the required clamping force is obtained, and the gripper of the robotic arm is controlled to clamp the handcart target with the required clamping force according to the required clamping force. The displacement of the gripper, tactile feedback, and reprojection error of the key points are detected. When the displacement, tactile feedback, and reprojection error are all within their respective preset ranges, the gripping is confirmed to be successful, and a signal confirming the gripped trolley and the gripping status is obtained.

[0009] Optionally, the step of planning a driving trajectory based on the current environmental perception information and the state of the already grasped handcart, while satisfying the narrow passage constraint and the minimum turning radius constraint, and controlling the mobile chassis to track the driving trajectory, includes: Acquire current environmental perception information and the status of the grabbed handcart, wherein the current environmental perception information includes the channel width and the location of obstacles; Update the equivalent envelope width of the mobile chassis based on the status of the grabbed handcart; Based on the equivalent envelope width and the channel width, the required safety gap is obtained, and based on the safety gap, it is determined whether the narrow channel passage constraint is met. Based on the preset minimum turning radius, the maximum allowable curvature of the trajectory is set, and the minimum turning radius constraint is determined based on the maximum allowable curvature of the trajectory. Under the premise of satisfying the narrow passage constraint and the minimum turning radius constraint, and in combination with the obstacle position, a driving trajectory from the current position of the mobile chassis to the target position of the stacking area is planned, and the mobile chassis is controlled to follow the driving trajectory.

[0010] Optionally, the step of automatically adjusting the travel speed of the mobile chassis based on the risk assessment results of the dynamic crowd, and transporting the grabbed trolleys to the stacking area to obtain the current stacking queue status, includes: Acquire dynamic crowd perception data, including the location of surrounding pedestrians; Based on the position of the mobile chassis and the position of each of the surrounding pedestrians, the distance between the mobile chassis and each of the surrounding pedestrians is determined, resulting in multiple distances. The minimum value among the multiple distances is selected as the minimum distance between the mobile chassis and the surrounding pedestrians. A speed adjustment coefficient is determined based on the minimum distance, wherein the speed adjustment coefficient is positively correlated with the minimum distance; Based on the desired speed of the mobile chassis and the speed adjustment coefficient, a target speed is obtained, and the mobile chassis is controlled to travel towards the stacking area at the target speed. After the mobile chassis arrives at the stacking area, the relative pose information of each trolley in the existing stacking queue is obtained, and the relative pose information is used as the current stacking queue state.

[0011] Optionally, the step of constructing a multi-vehicle coupled dynamics model based on the generalized coordinates and establishing docking geometric constraints and collision avoidance constraints includes: Obtain the generalized coordinates; Based on the generalized coordinates, determine the inertial matrix, Coriolis force matrix, and gravity vector of the coupled system consisting of the mobile chassis, the grabbed handcart, and the existing stacked queue. Introducing a constraint Jacobian matrix to determine the constraint force terms of the coupled system; The multi-vehicle coupled dynamics model is constructed based on the inertia matrix, the Coriolis force matrix, the gravity vector, and the constraint force terms. Based on the generalized coordinates, establish the docking geometric constraints between each of the handcarts during the docking process; Based on the generalized coordinates, anti-collision constraints are established between each of the handcarts and between the handcarts and the external environment.

[0012] Optionally, generating coordinated control commands for the mobile chassis and the robotic arm based on the multi-vehicle coupled dynamics model, the geometric constraints, and the collision avoidance constraints includes: By combining the multi-vehicle coupled dynamics model, the docking geometric constraints, and the anti-collision constraints, a dynamic equation containing the constraints is constructed. Solving the dynamic equations yields coordinated control commands that satisfy the docking geometric constraints and the anti-collision constraints.

[0013] Secondly, the present invention provides an autonomous collection and flexible stacking system for handcarts, applied to a robot system, the robot system comprising a mobile chassis and a robotic arm, the autonomous collection and flexible stacking system for handcarts comprising: A multimodal perception fusion unit is used to acquire multimodal perception data and perform fusion recognition on the multimodal perception data to obtain the handcart target and the pose information of the handcart target; The gripping and docking unit is used to perform gripping and compliant docking on the handcart target by the robotic arm according to the pose information, and to obtain the gripped handcart and the clamping status confirmation signal. The motion planning and control unit is used to plan a driving trajectory based on the current environmental perception information and the status of the grabbed handcart, under the conditions of satisfying the narrow passage passage constraint and the minimum turning radius constraint, control the mobile chassis to track the driving trajectory, and automatically adjust the driving speed of the mobile chassis according to the risk assessment results of the dynamic crowd, so as to transport the grabbed handcart to the stacking area to obtain the current stacking queue status. A continuous stacking unit is used to perform continuous stacking of multiple carts based on the already grabbed carts and the current stacking queue state; the continuous stacking unit includes: The state definition subunit is used to define generalized coordinates, which include the pose of the mobile chassis, the relative nesting depth between each trolley, and the relative yaw angle between each trolley. The model building subunit is used to construct a multi-vehicle coupled dynamic model based on the generalized coordinates, and to establish docking geometric constraints and anti-collision constraints; The instruction generation subunit is used to generate coordinated control instructions for the mobile chassis and the robotic arm based on the multi-vehicle coupled dynamics model, the geometric constraints, and the anti-collision constraints. The execution subunit is used to control the mobile chassis to perform the main traction motion according to the cooperative control command, and at the same time control the robotic arm to perform anisotropic impedance control for attitude fine-tuning, so as to nest the grabbed handcart into the tail of the stacking queue to obtain the updated stacking queue. The queue arrangement and parking unit is used to move the updated stacked queue to the target recycling point, thereby completing the queue arrangement and parking of the handcart.

[0014] Thirdly, an electronic device according to the present invention includes: a processor and a memory, the memory being used to store a computer program; When the computer program is loaded by the processor, it causes the processor to execute the above-described method for autonomous collection and flexible stacking of handcarts.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for autonomous collection and flexible stacking of handcarts.

[0016] The present invention relates to a method, system, device, and storage medium for autonomous collection and flexible stacking of trolleys. In the target acquisition stage, it first fuses and identifies multimodal sensing data to accurately output the pose information of the trolley target. This effectively distinguishes between interference from similar objects such as passenger luggage and cleaning equipment in an airport environment, avoiding the missed or false detection problems caused by occlusion or changes in lighting in traditional vision solutions, thus providing a reliable prerequisite for subsequent grasping and stacking. During the grasping and docking process, a robotic arm performs compliant docking and confirms the gripping state, enabling the system to adapt to minor pose deviations caused by human parking or uneven ground, achieving low-impact, high-success-rate single-trolley grasping. During the handling and passage phase, this invention, based on the current environmental perception information and the status of the grasped trolley, explicitly satisfies the narrow passage constraints and minimum turning radius constraints when planning the driving trajectory. At the same time, it automatically adjusts the driving speed of the mobile chassis by incorporating the risk assessment results of dynamic crowds. This invention, through perception first, planning then control, enables the robot to drive smoothly in areas with limited space and high traffic flow, such as check-in areas, boarding gate corridors, and serpentine barriers, actively avoiding passengers and luggage. It completely avoids the safety hazards such as collisions and jams that are prone to occur in traditional automated equipment, and meets the strict safety and social norms of airport public spaces.

[0017] This invention also constructs a generalized coordinate system encompassing the pose of the mobile chassis, the relative nesting depth between each trolley, and the relative yaw angle. Based on this, a multi-vehicle coupled dynamics model and geometric and anti-collision constraints on the contact surfaces are established. Compared to the simple force-controlled pushing or rigid alignment in existing solutions, the modeling method of this invention describes the dynamic coupling relationship between multiple vehicles during the stacking process from a mechanistic perspective, enabling the system to predict and control the evolution of nesting depth and yaw angle. On this basis, by generating coordinated control commands for the mobile chassis and the robotic arm, the mobile chassis executes the main traction motion to provide stable propulsion, while the robotic arm performs anisotropic impedance control for attitude fine-tuning. Through this control strategy, the robotic arm maintains appropriate stiffness in the propulsion direction to achieve effective nesting, while exhibiting low stiffness characteristics in the lateral and yaw directions. This adaptively tolerates pose errors between trolleys caused by manufacturing tolerances or ground slope, achieving smooth, low-impact, and high-success-rate continuous stacking of trolleys, and avoiding vehicle deformation or structural damage due to hard collisions. Simultaneously, this invention updates the stacking queue status in real time after stacking is completed and moves the entire queue to the target recycling point for parking. Through closed-loop feedback of the current stacking queue's pose and nesting depth, the system can dynamically adjust subsequent stacking strategies, ensuring that the multi-vehicle stacking process remains controllable and stable, significantly improving the continuity and reliability of batch collection operations. Furthermore, this invention, through generalized coordinate parameterization, enables the model to quickly adapt to the geometry and nesting interfaces of different airports and trolley models, possessing excellent cross-scene transfer capabilities.

[0018] In summary, this invention significantly improves the robustness of robot-driven autonomous collection and stacking of trolleys in complex dynamic environments such as airports by using multimodal perception fusion, safe trajectory planning in dynamic environments, multi-vehicle coupled dynamic modeling based on generalized coordinates, and coordinated control of main traction and anisotropic impedance. It also enhances the recognition robustness, passage safety, stacking success rate, and overall operational efficiency of trolleys in complex dynamic environments such as airports. This provides a low-manpower-dependent and highly stable automated solution for intelligent airport operations. Furthermore, it can be extended to similar trolley management scenarios such as large transportation hubs and convention centers. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the autonomous collection and flexible stacking method for handcarts according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of the handcart autonomous collection and flexible stacking system according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0020] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the accompanying drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0021] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0022] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; and the term "optionally" means "optional embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first," "second," etc., mentioned in this invention are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0023] It should be noted that the terms "a" and "a plurality of" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0024] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties. The collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0025] Combination Figure 1 As shown in the figure, an embodiment of the present invention provides a method for autonomous collection and flexible stacking of handcarts, applied to a robot system. The robot system includes a mobile chassis and a robotic arm. The method for autonomous collection and flexible stacking of handcarts includes: Multimodal sensing data is acquired and fused to obtain the handcart target and its pose information.

[0026] Specifically, after system startup, the system first acquires surround view images, 3D laser point clouds, depth maps, and inertial measurement unit (IMU) data through a 360° blind-spot-free multimodal perception module. All sensor data are then synchronized in time and calibrated to external parameters, unifying them to the robot's base coordinate system. Specifically, a bird's-eye view (BEV) fusion framework is used to project multi-camera features onto a unified plane, which are then aligned and fused with point cloud voxel features to obtain fused features containing both geometric and texture information. In the target detection stage, a target detection loss function including an instance discrimination loss term is used. Training the detection network, where, This represents the total loss in target detection. Represents the classification loss weight coefficient. Represents the classification loss term. This represents the weighting coefficient of the bounding box regression loss. This represents the bounding box regression loss term. This represents the weighting coefficient of the target confidence loss. This represents the target confidence loss term. This represents the instance-distinguishing loss weight coefficient. This indicates the instance-specific loss term. Using focus loss To address the issue of sample imbalance between carts and non-carts, among which, This represents the focus loss balance parameter. This represents the probability that the model predicts the target class. Indicates the focus loss focusing parameter. To mitigate false positives, a contrastive learning method is used to differentiate the trolley from similar structures such as luggage handles and guardrails in the feature space. The detection network simultaneously outputs the trolley's 3D bounding box, the main beam of the frame, and the key point set of the docking point. These are then combined with depth information and point cloud regression to obtain the trolley's 3D pose. And its uncertainty. In the event of short-term loss due to obstruction by passengers or reflection, the system uses multi-frame fusion and tracking compensation (such as extended Kalman filtering) to stably restore the target state.

[0027] Based on the pose information, the robotic arm performs grasping and compliant docking on the handcart target to obtain a grasped handcart and a clamping status confirmation signal.

[0028] Specifically, based on the trolley's pose information, the approach pose is planned to ensure the robotic arm is within reach and the chassis posture meets the subsequent traction direction. During the close-range phase, visual servo control is employed, using keypoint reprojection errors as feedback to calculate the desired end-effector pose, ensuring the gripper guide structure aligns with the trolley's docking inlet direction. To achieve compliant contact, an impedance control model is used at the robotic arm's end effector. ,in, The stiffness matrix represents the stiffness matrix in impedance control, and is also relevant to the stiffness matrix of the robotic arm. Anisotropic parameters are configured, and a local coordinate system is established with the docking direction as the X-axis, the lateral direction as the Y-axis, and the yaw direction as the Z-axis. The stiffness coefficient of the X-axis is set... Setting it to a small value (low stiffness), when subjected to resistance in the X direction, the robotic arm's end effector exhibits spring-like flexibility and compliantly retracts to mitigate longitudinal distance errors; simultaneously, the lateral stiffness coefficient of the Y-axis is... and the rotational stiffness coefficient of the Z-axis Setting it to a larger value (high stiffness) gives the system strong resistance when subjected to lateral pushing, forcibly keeping the axes of the two handcarts strictly aligned, effectively preventing lateral bending and skew expansion in long queues. This represents the damping matrix in impedance control. This represents the stiffness coefficient of the local coordinate system along the X-axis (dating direction). This represents the lateral stiffness coefficient along the Y-axis of the local coordinate system. This represents the rotational stiffness coefficient of the Z-axis (yaw) in the local coordinate system. This represents the desired pose vector of the robotic arm's end effector. This represents the actual pose vector of the robotic arm's end effector. This represents the desired velocity vector at the end of the robotic arm. This represents the actual velocity vector at the end of the robotic arm and limits the maximum contact force and contact moment to avoid damage to the frame. The adaptive gripping module determines the gripper opening based on visual measurements and tactile feedback, and estimates the contact friction coefficient online according to the formula... Dynamically adjust the clamping force, among which, This indicates the clamping force output by the end effector. Indicates the equivalent load mass of the handcart. Represents the gravitational acceleration constant. This represents the online estimated coefficient of friction of the contact surface. This indicates the set safety margin coefficient. After successful clamping, the clamping status is confirmed through multi-source consistency verification: the tactile force must reach the threshold, the gripper displacement must be within a reasonable range, and the key point reprojection error must be less than the set threshold. If any condition is not met, the process will stop and realign. Once all conditions are met, a clamping status confirmation signal will be output, completing the reliable gripping of a single handcart.

[0029] Based on the current environmental perception information and the status of the grabbed handcarts, a driving trajectory is planned under the conditions of satisfying the narrow passage passage constraint and the minimum turning radius constraint. The mobile chassis is controlled to track the driving trajectory, and the driving speed of the mobile chassis is automatically adjusted according to the risk assessment results of the dynamic crowd, so as to transport the grabbed handcarts to the stacking area to obtain the current stacking queue status.

[0030] Specifically, after acquiring the grasped trolley, the system plans a driving trajectory based on current environmental perception information, including dynamic crowds, static obstacles, and the status of the stacked queue. To meet the airport's 1.2-meter narrow passage requirement, the planner explicitly constructs passability conditions. These narrow passage passability conditions are used to quantify the geometric boundaries by which the robot and its envelope can safely pass through the narrow passage. ,in, Indicates the physical channel width. This represents the equivalent vehicle body envelope width of the robot or queue. Indicates the safe radius of the obstacle expansion. The minimum safe clearance requirement indicates that the vehicle body envelope width must include the equivalent width of the lateral sway of the traction or stacking queue. The obstacle expansion radius and the minimum safe clearance need to be calibrated based on actual tests. Simultaneously, path curvature constraints are applied to ensure that the minimum turning radius does not exceed 1.5 m. The path curvature constraints are used to ensure that the minimum turning radius requirement is met within narrow passages. , Represents the trajectory in path coordinates curvature at that point This represents the maximum permissible curvature of the trajectory. For dynamic crowds, the system introduces a social perception obstacle avoidance mechanism, constructing an anisotropic personal spatial risk function for each passenger. Based on the human-machine closest distance, the system automatically adjusts the driving speed and employs an automatic speed limit command function. This function automatically outputs a smooth, safe speed limit command based on the human-machine closest distance. ,in, This indicates the speed control command issued to the chassis. This represents the numerical limiting function. This represents the expected base operating speed. This represents a smooth activation function (such as Sigmoid). A coefficient representing the steepness of the speed-limiting curve. This indicates the closest distance between a human and a robot. This indicates the safe distance threshold that triggers automatic deceleration. This indicates the maximum permissible operating speed of the system, triggering a deceleration threshold. When the predicted time-to-collision time is less than the threshold or the predicted minimum distance is less than the threshold, a deceleration-stop-bypass tiered strategy is triggered. The Time-to-Collision (TTC) risk model is used to assess collision risk in dynamic environments and to trigger tiered safety strategies, prioritizing lateral detour and yielding behaviors. ,in, This indicates the time to collision parameters in the prediction. Indicates a future time-varying variable. Represents a vector of relative positions between objects. This represents the relative velocity vector between objects. This represents the critical safe distance for determining a collision. During trajectory tracking, formal safety constraints are applied through a control barrier function (CBF). The CBF safety constraints are used to impose strict collision avoidance space and speed limits in navigation control. ,in, This represents the relative safety distance defined by the control barrier function. This represents the derivative of the barrier function with respect to the rate of change over time. This represents the positive constant gain coefficient. The control variables are solved online using quadratic programming to ensure the safe transport of grabbed trolleys to the stacking area in a dynamic environment, and to update the current stacking queue state (such as existing queue length, end pose, etc.).

[0031] Based on the already grabbed trolleys and the current stacking queue state, perform continuous stacking of multiple trolleys: Define a generalized coordinate system, which includes the pose of the mobile chassis, the relative nesting depth between each trolley, and the relative yaw angle between each trolley. A multi-vehicle coupled dynamics model is constructed based on the generalized coordinates, and docking geometric constraints and anti-collision constraints are established. Based on the multi-vehicle coupled dynamics model, the geometric constraints, and the collision avoidance constraints, collaborative control commands for the mobile chassis and the robotic arm are generated. According to the cooperative control command, the mobile chassis is controlled to perform the main traction motion, and the robotic arm is controlled to perform anisotropic impedance control to fine-tune the attitude, so that the grasped handcart is nested into the tail of the stacking queue, resulting in an updated stacking queue.

[0032] Specifically, once the handcarts have been grabbed and transported to the stacking area, the system executes continuous stacking of multiple carts, which is divided into the following sub-processes: Define generalized coordinates: For the current stacked queue, including the chassis and several stacked trolleys, up to 10 trolleys are supported as a multibody system. Define a generalized coordinate vector whose components include the absolute pose of the moving chassis, the relative nesting depth between each trolley, and the relative yaw angle between each trolley.

[0033] Constructing a multi-vehicle coupled dynamics model: A dynamics model of the multi-vehicle coupled system is established based on generalized coordinates. The continuously stacked coupled system dynamics model is used to uniformly describe the nonlinear coupled motion relationship between the chassis and multiple handcarts. ,in, This represents the mass and inertia matrix of a multibody system. Represents a generalized coordinate vector. Represents the generalized velocity vector, i.e. The first derivative with respect to time, Represents the generalized acceleration vector, i.e. The second derivative with respect to time, Represents the matrix of Coriolis force and centrifugal force terms. This represents the system's viscous damping coefficient matrix. This represents the equivalent gravitational and elastic potential energy terms. This represents the control input mapping matrix. This indicates that the chassis and the control arm work together to control the input. This represents the constraint Jacobian matrix of the system. This represents the equivalent constraint term or Lagrange multiplier. express The transpose of .

[0034] Establish docking physical constraints: For nested docking of trolleys, set geometric equality constraints, such as guide structure alignment and keeping the nesting depth within the allowable range, and anti-collision inequality constraints, such as the minimum clearance between wheels and facilities and passengers. The docking physical constraints are used to describe the geometric limits and anti-pinch collision restrictions when trolleys are stacked. ,in, Represents the geometrical equality constraint equations within the system. This represents the system's collision avoidance and limit inequality constraint equations.

[0035] Generate coordinated control commands: Based on the above dynamic model and constraints, and combined with the target nesting depth, generate coordinated control commands for the chassis and robotic arm by solving the constrained optimal control problem or based on a force / position hybrid control strategy.

[0036] Main traction and anisotropic impedance control are implemented: After receiving commands, the chassis performs low-speed main traction motion, providing longitudinal nesting thrust; simultaneously, the robotic arm performs anisotropic impedance control based on feedback from the six-dimensional force / torque sensor at the end effector. Specifically, low stiffness is set in the docking forward direction to tolerate longitudinal errors and achieve compliant propulsion, while high stiffness is set in the lateral and yaw directions to prevent the queue skew from widening. Through the coordinated approach of main traction and anisotropic compliance, the grabbed handcart is smoothly nested into the tail of the stacking queue, resulting in an updated stacking queue (queue length increased by 1).

[0037] Jamming detection and anomaly handling: During stacking, the system monitors the constraint force, the rate of change of nesting depth, and the joint motor current in real time. When the constraint force exceeds the threshold and the nesting depth growth stops, jamming is detected, and the system automatically executes a retry process of withdrawal-lateral micro-movement-realignment (retrying 1 to 3 times). If the retry still fails, the system enters a safe stop and requests manual intervention. This closed-loop mechanism ensures the continuous and stable stacking of up to 10 trolleys.

[0038] The updated stacked queue is moved to the target recycling point to complete the queue arrangement and parking of the handcarts.

[0039] Specifically, after stacking a preset number of vehicles, the updated stack queue, i.e., the entire queue, is transported to the target recycling point. During the transport, the narrow passage curvature constraints and social perception obstacle avoidance rules from step three are continued, and the upper limits of angular velocity and acceleration are dynamically reduced based on the number of stacked vehicles to minimize queue swaying. Upon reaching the vicinity of the recycling point, precise positioning is achieved using ground visual anchors, such as QR codes, April Tags, or ground markings, and end-effector attitude shaping is performed: the queue axis is adjusted to align with the recycling point direction through a small "S" shaped trajectory or micro-steering. Finally, low-speed, high-precision control is used to park the queue at the recycling point, monitoring the gap between the queue end and surrounding facilities to ensure a parallel, fan-shaped, or queue-like arrangement. After parking, a release action is executed (releasing the grippers and exiting), completing the queue arrangement and parking of the handcarts. Throughout the entire process, the safety monitoring module operates independently, and the emergency stop braking response dynamic model... ,in, Represents the saturation limiting function. This indicates the maximum permissible jerk limit. This indicates the set maximum reverse braking deceleration. Indicates a future time-varying variable. This represents the response delay time of an emergency stop command, i.e., the time from when the command is triggered to when the braking system begins to generate effective braking force. Braking distance estimation is used to estimate the braking distance in emergency stop situations to ensure a safety clearance. Total braking distance... ,in, This indicates the estimated total braking distance required. This represents the robot's initial linear velocity at the moment the emergency stop command is triggered (t=0). Indicates the moment when the object has come to a complete stop. This represents the robot's real-time speed function as it brakes. The software layer triggers automatic speed limiting or emergency stopping in real time based on the time until collision and minimum distance, ensuring safety in public spaces.

[0040] The autonomous collection and flexible stacking method for trolleys in this embodiment first fuses and identifies multimodal perception data during the target acquisition stage, accurately outputting the pose information of the trolley target. This effectively distinguishes interference from similar objects such as passenger luggage and cleaning equipment in the airport environment, avoiding the missed detections and false detections caused by occlusion or changes in lighting in traditional vision solutions, thus providing a reliable prerequisite for subsequent grasping and stacking. During the grasping and docking process, a robotic arm performs compliant docking and confirms the gripping state, enabling the system to adapt to minor pose deviations of the trolleys caused by human parking or uneven ground, achieving low-impact, high-success-rate single-trolley grasping. During the handling and passage phase, this embodiment explicitly satisfies narrow passage constraints and minimum turning radius constraints when planning the driving trajectory based on the current environmental perception information and the status of the grasped trolley. At the same time, it automatically adjusts the driving speed of the mobile chassis by incorporating the risk assessment results of dynamic crowds. This embodiment enables the robot to drive smoothly in areas with limited space and high traffic flow, such as check-in areas, boarding gate corridors, and serpentine barriers, and actively avoid passengers and luggage. It completely avoids the safety hazards such as collisions and jams that are prone to occur in traditional automated equipment, and meets the strict safety and social norms of airport public spaces.

[0041] This embodiment also constructs a generalized coordinate system encompassing the pose of the mobile chassis, the relative nesting depth between each trolley, and the relative yaw angle. Based on this, a multi-vehicle coupled dynamics model and geometric and anti-collision constraints on the contact surfaces are established. Compared to the simple force-controlled pushing or rigid alignment in existing solutions, the modeling method in this embodiment describes the dynamic coupling relationship between multiple vehicles during the stacking process from a mechanistic perspective, enabling the system to predict and control the evolution of nesting depth and yaw angle. On this basis, by generating coordinated control commands for the mobile chassis and the robotic arm, the mobile chassis executes the main traction motion to provide stable propulsion, while the robotic arm performs anisotropic impedance control for attitude fine-tuning. Through this control strategy, the robotic arm maintains appropriate stiffness in the propulsion direction to achieve effective nesting, while exhibiting low stiffness characteristics in the lateral and yaw directions. This adaptively tolerates pose errors between trolleys caused by manufacturing tolerances or ground slope, achieving smooth, low-impact, and high-success-rate continuous stacking of trolleys, and avoiding vehicle deformation or structural damage due to hard collisions. Simultaneously, this invention updates the stacking queue status in real time after stacking is completed and moves the entire queue to the target recycling point for parking. Through closed-loop feedback of the current stacking queue's pose and nesting depth, the system can dynamically adjust subsequent stacking strategies, ensuring that the multi-vehicle stacking process remains controllable and stable, significantly improving the continuity and reliability of batch collection operations. Furthermore, this invention, through generalized coordinate parameterization, enables the model to quickly adapt to the geometry and nesting interfaces of different airports and trolley models, possessing excellent cross-scene transfer capabilities.

[0042] In summary, this embodiment significantly improves the robustness of robot-driven autonomous collection and stacking of trolleys in complex dynamic environments such as airports by using multimodal perception fusion, safe trajectory planning in dynamic environments, multi-vehicle coupled dynamic modeling based on generalized coordinates, and coordinated control of main traction and anisotropic impedance. It also enhances the recognition robustness, passage safety, stacking success rate, and overall operational efficiency of trolleys in complex dynamic environments such as airports. This provides a low-manpower-dependent and highly stable automated solution for intelligent airport operations. Furthermore, it can be extended to similar trolley management scenarios such as large transportation hubs and convention centers.

[0043] Optionally, the step of acquiring multimodal perception data and fusing and recognizing the multimodal perception data to obtain the handcart target and its pose information includes: Acquire multimodal sensing data from at least two types of sensors, including a surround-view camera, a 3D LiDAR, or a depth camera; The multimodal sensing data is time-synchronized and extrinsic parameters are calibrated to obtain aligned sensing data with a unified time reference and a unified spatial coordinate system. The aligned perception data is input into the target detection model to obtain the detection box and key points of the handcart target; Based on the detection box and the key points, and combined with the 3D point cloud data in the aligned perception data, the 3D position and spatial pose of the handcart target are obtained, and the 3D position and spatial pose are used as the pose information of the handcart target.

[0044] Specifically, multimodal perception data is acquired using at least two sensors (including a surround-view camera, a 3D LiDAR, or a depth camera). For example, the surround-view camera provides a 360° blind-spot-free image, the 3D LiDAR provides a dense point cloud, the depth camera provides a depth map, and the inertial measurement unit (IMU) provides attitude and acceleration information. Then, all sensor data is synchronized in time and calibrated using extrinsic parameters: hardware timestamps or software interpolation methods are used to align the sensor data to a unified time reference, and a pre-calibrated extrinsic parameter matrix is ​​used to transform the LiDAR point cloud, depth camera data, and surround-view camera image to the same spatial coordinate system (usually the robot's base coordinate system), resulting in aligned perception data. Next, the aligned perception data is input into a target detection model; this model uses a bird's-eye view (BEV) fusion framework, projecting the multi-camera surround-view image features onto a unified plane, and then aligning and fusing them with the 3D LiDAR point cloud voxel features to form a fused feature containing geometric and textural information. The loss function of the detection model is... Among them, classification loss Using focus loss To address the imbalance between cart and non-cart samples, instance discrimination loss is used. By employing contrastive learning, the model distances the trolley from similar structures in the feature space, such as luggage handles and guardrails, thereby suppressing false detections. The model outputs a 2D bounding box for the trolley target and a predefined set of key points, including stable structures such as the intersection of the main beams of the frame, the edge of the docking entrance, and guide holes. Finally, based on the bounding box and key points, combined with aligned 3D point cloud data, the 3D coordinates of the key points are calculated using depth regression or point cloud projection, thus estimating the trolley's 3D position. and orientation angle This yields the complete pose information and uncertainty of the trolley target, which is used for subsequent grasping and docking. For short-term loss due to passenger obstruction or glass reflection, multi-frame fusion and tracking compensation (such as extended Kalman filtering) are further employed to stabilize and restore the target state.

[0045] In this optional embodiment, multi-source heterogeneous data from surround-view cameras, 3D LiDAR, and depth cameras are fused and unified into the robot's base coordinate system after time synchronization and extrinsic parameter calibration. This overcomes the perception limitations of a single sensor under conditions of reflection, such as metal frames, glass curtain walls, obstructions (crowded passengers), and drastic changes in lighting. A BEV bird's-eye view fusion framework is adopted to closely align the features of multi-camera images with the voxel features of the LiDAR point cloud, enabling the model to utilize both texture and geometric information, significantly enhancing its ability to identify the metal frame of the trolley. An instance discrimination loss term is introduced into the detection loss function. Through contrastive learning, the trolley features are forced to distance themselves from similar interfering objects such as luggage handles and guardrail posts in the feature space, significantly reducing the false detection rate from the source. At the same time, the focus loss effectively alleviates the problem of imbalanced positive and negative samples, improving the detection performance for difficult examples such as partial occlusion. The model outputs predefined key points such as the intersection of the main beams of the chassis and the edge of the docking entrance. Combined with 3D point cloud data regression, it obtains the precise 3D position and orientation angle of the trolley. This not only provides the complete pose information required for grasping but also lays a reliable foundation for the subsequent visual servoing and compliant docking of the robotic arm. This ensures that the robot consistently achieves stable and accurate trolley perception results in crowded, poorly lit public areas like airports, where there are many similar objects, strongly supporting the autonomous execution of subsequent grasping, stacking, and handling tasks.

[0046] Optionally, the step of performing grasping and compliant docking of the handcart target by the robotic arm based on the pose information, and obtaining a grasped handcart and a clamping status confirmation signal, includes: Based on the pose information, an approach trajectory for the robotic arm is generated, and the robotic arm is controlled to move to the initial docking position of the handcart target based on the approach trajectory. After the robotic arm reaches the initial docking position, the robot system enters the visual servoing stage, using the key points of the handcart target as feedback to correct the pose of the robotic arm end in real time, so that the guide structure of the robotic arm end is aligned with the docking entrance of the handcart target. After the end of the robotic arm contacts the target trolley, the robot system switches to impedance control mode to correct the pose of the end of the robotic arm to compensate for the docking pose deviation. Based on the online estimated coefficient of friction of the contact surface and the load mass of the handcart target, the required clamping force is obtained, and the gripper of the robotic arm is controlled to clamp the handcart target with the required clamping force according to the required clamping force. The displacement of the gripper, tactile feedback, and reprojection error of the key points are detected. When the displacement, tactile feedback, and reprojection error are all within their respective preset ranges, the gripping is confirmed to be successful, and a signal confirming the gripped trolley and the gripping status is obtained.

[0047] Specifically, based on the 3D pose (position and orientation angle) of the target trolley output by the perception module and a predefined set of key points (including the intersection of the main beams of the frame, the edge of the docking entrance, and guide holes), an approach trajectory for the robotic arm is generated. This trajectory smoothly transitions from the current arm configuration of the robot to the initial docking position in front of the target trolley, ensuring that the gripper guide structure is basically consistent with the direction of the trolley entrance. The robotic arm is controlled to move along this trajectory to the initial docking position, during which the chassis remains stationary or is finely adjusted to ensure accessibility. After the robotic arm reaches the initial docking position, the system enters the visual servoing stage, using the reprojection error of the key points of the trolley target as a feedback signal to correct the pose (including position and orientation) of the robotic arm end in real time, so that the gripper guide structure is accurately aligned with the docking entrance of the trolley. During this process, the stiffness of the robotic arm end is set to a moderate level to quickly respond to visual errors. Once the robotic arm's end effector makes physical contact with the trolley frame, the system immediately switches to impedance control mode. This mode employs an impedance control model to reduce stiffness in the docking direction and allow for elastic yielding. Compliant posture fine-tuning compensates for docking position deviations caused by uneven ground, frame deformation, or residual visual errors, while limiting maximum contact force to prevent scratches or damage. In the contact-holding state, the adaptive gripping module estimates the contact surface friction coefficient online (based on current changes or micro-slip detection during gripper contact) and dynamically calculates the required gripping force, considering the trolley's load mass and safety factor. The system then controls the grippers to clamp the trolley frame with this force value; the gripper opening is determined jointly by visual measurement and tactile feedback. Finally, a multi-source consistency check is performed: simultaneously detecting whether the actual gripper displacement is within a preset reasonable range, whether the tactile force curve from the end effector's six-dimensional force sensor shows stable contact (without spikes or sudden drops), and whether the key point reprojection error is less than a set threshold (e.g., pixel-level error). Only when the above displacement, tactile force, and reprojection error all meet their respective preset ranges is the clamping confirmed, and a signal confirming the gripped trolley and clamping status is output; if any condition is not met, the process of retraction and realignment is stopped.

[0048] In this optional embodiment, key point feedback is used to correct the end-effector pose in real time during the visual servoing stage, ensuring that the guide structure is initially aligned with the docking entrance. After contact, the system switches to impedance control, reducing the stiffness in the docking direction and allowing elastic yielding, thus eliminating the risk of scratching or jamming that may result from rigid push-in. The adaptive clamping module estimates the coefficient of friction online and dynamically calculates the required clamping force, avoiding slippage due to insufficient clamping force or damage to the frame due to excessive clamping force. Finally, the clamping success is reliably confirmed through a three-source consistency check of gripper displacement, tactile force feedback, and key point reprojection error. If any condition is not met, the process is stopped and retried. This embodiment significantly improves the success rate of a single gripping attempt, reduces the probability of collision damage between the equipment and the trolley, and provides a stable and reliable clamping foundation for subsequent multi-cart stacking.

[0049] Optionally, the step of planning a driving trajectory based on the current environmental perception information and the state of the already grasped handcart, while satisfying the narrow passage constraint and the minimum turning radius constraint, and controlling the mobile chassis to track the driving trajectory, includes: Acquire current environmental perception information and the status of the grabbed handcart, wherein the current environmental perception information includes the channel width and the location of obstacles; Update the equivalent envelope width of the mobile chassis based on the status of the grabbed handcart; Based on the equivalent envelope width and the channel width, the required safety gap is obtained, and based on the safety gap, it is determined whether the narrow channel passage constraint is met. Based on the preset minimum turning radius, the maximum allowable curvature of the trajectory is set, and the minimum turning radius constraint is determined based on the maximum allowable curvature of the trajectory. Under the premise of satisfying the narrow passage constraint and the minimum turning radius constraint, and in combination with the obstacle position, a driving trajectory from the current position of the mobile chassis to the target position of the stacking area is planned, and the mobile chassis is controlled to follow the driving trajectory.

[0050] Specifically, based on the current environmental perception information and the status of the already grabbed trolleys, the process of planning the driving trajectory and controlling the mobile chassis to track the vehicle under the conditions of satisfying narrow passage constraints and minimum turning radius constraints is as follows: First, the system acquires the current environmental perception information, including the passage width, obstacle positions, dynamic crowd distribution, and the status of the already grabbed trolleys, such as whether multiple trolleys have been towed in a stacked queue. Then, the equivalent envelope width of the mobile chassis is updated according to the status of the grabbed trolleys: when only a single trolley is grabbed, the equivalent envelope width is the larger of the chassis width and the trolley width; when multiple trolleys have been stacked, the lateral sway envelope of the queue during turning needs to be included in the equivalent width estimation, for example, by adding a margin through empirical formulas or sway amplitude predicted based on multibody dynamics models. Next, the obstacle expansion radius is combined with... and the preset minimum safety gap (e.g., 0.05 to 0.15 m), verify narrow passage constraints: if the passage width satisfy If the path is cleared, the system considers it safe to proceed; otherwise, it will choose an alternative path or wait for the passage to clear. Regarding minimum turning radius constraints, the system uses a preset minimum turning radius... Set the maximum allowable curvature of the trajectory The curvature of all path points when planning the trajectory Implement explicit restrictions to ensure To avoid sharp turns caused by an excessively high ratio of angular velocity to linear velocity, the system, while simultaneously satisfying narrow passage constraints and minimum turning radius constraints, combines obstacle locations, including static facilities and dynamic pedestrians, and employs curvature-continuous spline trajectories or sampling planning methods based on motion primitives to generate a collision-free, curvature-constrained driving trajectory from the current position of the mobile chassis to the target position in the stacking area. If multiple candidate trajectories exist, the social risk cost is additionally assessed, and the optimal trajectory is selected based on the proximity to the crowd. Finally, the chassis controller uses a tracking control algorithm, such as pure tracking model predictive control, to track the planned trajectory in real time, continuously verifying curvature constraints and safety clearances during the tracking process, and making local adjustments when necessary. For example, when the chassis controller tracks the planned trajectory, a pure tracking model geometric path tracking method can be used: First, based on the current chassis speed and preset forward sight distance rules (e.g., the faster the speed, the farther the forward sight distance), a forward reference point is selected on the planned trajectory; then, the arc turning angle from the current chassis position to the reference point is calculated, causing the chassis to travel towards the reference point in a smooth arc; as the chassis continues to move forward, the reference point rolls forward along the trajectory, and the chassis adjusts its steering and speed in real time, continuously converging the deviation between its position and orientation and the reference trajectory, thereby achieving stable movement along the planned trajectory. Simultaneously, the controller limits the desired steering angle based on the minimum turning radius constraint to avoid exceeding the curvature upper limit, and merges it with the speed command output by the social perception obstacle avoidance module to form the final linear velocity and angular velocity control commands, achieving stable tracking of the planned trajectory.

[0051] In this optional embodiment, the equivalent envelope width is dynamically updated based on the captured trolley status (especially the stacking queue length), and the lateral sway during queue turns is included in the estimation, avoiding scraping or jamming caused by insufficient envelope estimation. Narrow passage conditions are verified by combining obstacle expansion radius and minimum safety clearance; passage is only planned when the passage width meets the requirements, otherwise, the robot actively detours or waits, eliminating the risk of forcibly entering narrow passages from the source. Simultaneously, a minimum turning radius constraint (maximum curvature limit) is explicitly applied, and a trajectory is generated using curvature-continuous spline trajectory or motion primitive sampling, preventing queue tailing or loss of control caused by sharp turns. The chassis controller uses geometric tracking methods such as pure tracking to continuously converge the deviation between the actual driving trajectory and the planned trajectory. The controller limits the desired steering angle and integrates socially perceived speed commands to ensure that the upper limit of curvature and safe speed are always met during tracking. This embodiment effectively solves the autonomous passage problem in scenarios such as narrow airport passages and small turning radii, reducing the probability of collisions between the robot and facilities or passengers, and ensuring the continuous and reliable execution of trolley handling tasks.

[0052] Optionally, the step of automatically adjusting the driving speed of the mobile chassis based on the risk assessment results of the dynamic crowd, and transporting the grabbed handcarts to the stacking area to obtain the current stacking queue status, includes: Acquire dynamic crowd perception data, including the location of surrounding pedestrians; Based on the position of the mobile chassis and the position of each of the surrounding pedestrians, the distance between the mobile chassis and each of the surrounding pedestrians is determined, resulting in multiple distances. The minimum value among the multiple distances is selected as the minimum distance between the mobile chassis and the surrounding pedestrians. A speed adjustment coefficient is determined based on the minimum distance, wherein the speed adjustment coefficient is positively correlated with the minimum distance; Based on the desired speed of the mobile chassis and the speed adjustment coefficient, a target speed is obtained, and the mobile chassis is controlled to travel towards the stacking area at the target speed. After the mobile chassis arrives at the stacking area, the relative pose information of each trolley in the existing stacking queue is obtained, and the relative pose information is used as the current stacking queue state.

[0053] Specifically, the system acquires real-time perception data of dynamic crowds through a multimodal perception module, including the three-dimensional position and speed information of each pedestrian in the surrounding area. Simultaneously, the current position of the mobile chassis itself is provided by an odometer or positioning module. Then, the Euclidean distance between the mobile chassis and each surrounding pedestrian is calculated, and the minimum value among all distances is selected as the current minimum human-machine distance. Based on this minimum distance, a smooth speed adjustment mechanism is used to determine the speed adjustment coefficient: when the minimum human-machine distance is greater than or equal to a preset safe distance threshold, such as 1.2 meters, the speed adjustment coefficient is 1, meaning no deceleration; when the minimum human-machine distance is less than this threshold, the speed adjustment coefficient gradually decreases from 1 to near 0 according to a smooth curve, making the coefficient smaller as the distance gets closer. Next, the desired speed of the mobile chassis, such as the normal cruising speed, is multiplied by this speed adjustment coefficient and then subjected to a limiting process (limited to between 0 and the maximum speed allowed by the system) to obtain the target speed. Simultaneously, if the predicted collision time is less than a set threshold or the predicted minimum distance is less than a set threshold, a graded strategy of deceleration, stopping, and detour is further triggered. When stopping, the target speed is directly set to 0. After receiving the target speed command, the chassis controller smoothly adjusts the actual speed to the target speed based on the current path tracking requirements, and actively slows down to give way in densely populated areas to avoid close contact with passengers. Once the mobile chassis safely arrives at the stacking area, the multimodal perception and tracking module detects and estimates the pose of the existing trolley stack queues within the area: by detecting key points of each trolley and combining them with point cloud data, the relative nesting depth and relative yaw angle between adjacent trolleys in the queue, as well as the spatial pose of the end of the queue, are calculated; this relative pose information is used as the current stacking queue state for docking planning and control in subsequent multi-cart continuous stacking steps.

[0054] In this optional embodiment, by calculating the minimum human-robot distance in real time and employing a smooth speed adjustment curve, the robot continuously and naturally decelerates when approaching passengers rather than abruptly stopping, avoiding disturbance and collision risks. Simultaneously, by combining a collision time-to-collision and predicted minimum distance triggering tiered strategies (deceleration, stopping, and detouring), it ensures timely stopping or proactive yielding in unexpected situations, such as when a pedestrian suddenly turns around, making its behavior more in line with human expectations. Upon reaching the stacking area, the robot further obtains the relative nesting depth and yaw angle of adjacent trolleys in the existing queue, providing a precise initial state for subsequent continuous stacking of multiple trolleys, ensuring smooth stacking connections. This embodiment achieves a balance between safety and efficiency while ensuring public space safety and avoiding the decrease in traffic efficiency caused by overly conservative deceleration.

[0055] Optionally, the step of constructing a multi-vehicle coupled dynamics model based on the generalized coordinates and establishing docking geometric constraints and collision avoidance constraints includes: Obtain the generalized coordinates; Based on the generalized coordinates, determine the inertial matrix, Coriolis force matrix, and gravity vector of the coupled system consisting of the mobile chassis, the grabbed handcart, and the existing stacked queue. Introducing a constraint Jacobian matrix to determine the constraint force terms of the coupled system; The multi-vehicle coupled dynamics model is constructed based on the inertia matrix, the Coriolis force matrix, the gravity vector, and the constraint force terms. Based on the generalized coordinates, establish the docking geometric constraints between each of the handcarts during the docking process; Based on the generalized coordinates, anti-collision constraints are established between each of the handcarts and between the handcarts and the external environment.

[0056] Specifically, by jointly modeling the robot chassis and the stacked carts in mathematical and dynamic models, when implementing the continuous stacking of multiple carts, a generalized coordinate vector q is first defined to describe the entire coupled system of the mobile chassis-grabbed carts-existing stack queue. This vector includes not only the absolute pose of the robot chassis itself, but also the relative nesting depth of each cart in the queue relative to the cart in front. and relative yaw angle Secondly, the mass and inertia matrices of the chassis and all handcarts are combined to establish a unified system mass matrix. Coriolis force matrix And introduce a constraint Jacobian matrix. This matrix describes the kinematic hard constraints arising from nested physical contact between vehicles. It is derived from the kinematic hard constraints during the docking process (such as the geometric constraints of nested guide structures). Through this joint modeling approach, the controller can predict how the chassis's motion behavior will be transmitted through physical connections and affect the entire long queue (e.g., tail swing) before issuing commands, thereby achieving overall coordination of the multibody system.

[0057] The generalized coordinates are obtained through the real-time state estimation module. The absolute position and orientation angle of the moving chassis are provided by an extended Kalman filter that integrates data from the wheel odometer and the inertial measurement unit: the wheel odometer collects the number of rotary encoder pulses of the left and right drive wheels and converts them into displacement increments, and the inertial measurement unit provides triaxial angular velocity and acceleration. After time synchronization and extrinsic parameter calibration, the two are input into the filter, for example, outputting chassis pose estimation at a frequency of 50 times per second.

[0058] The relative nesting depth between the trolleys is measured in real time by a depth camera mounted at the end of the robotic arm: During stacking, the depth camera acquires point cloud data between the front face of the trolley to be stacked and the rear face of the already stacked trolleys. After planar fitting of the point cloud, the vertical distance between the two planes is calculated to obtain the nesting depth measurement. The relative yaw angle is obtained by visually detecting predefined key points on the trolley frames on both sides, such as handlebar pivots or frame intersections, and calculating the angle between the line connecting the key points and the reference axis. The above measurements are updated at a preset frequency, such as 20 times per second, and stored in a generalized coordinate vector after being smoothed by a first-order low-pass filter.

[0059] Simultaneously, a table of mass and moment of inertia parameters for the mobile chassis and various types of handcarts is pre-stored in non-volatile memory. For example, the chassis has a mass of 120 kg and a moment of inertia of 15 kg / m² about the vertical axis; each handcart has a mass of 18 kg and a moment of inertia of 2.5 kg / m². During operation, the controller dynamically constructs the inertia matrix based on the number of handcarts currently stacked (obtained by accumulating clamping completion signals). Specifically, the inertia matrix is ​​calculated in a block format: the main diagonal elements of the submatrix corresponding to the chassis are the chassis mass and moment of inertia; the main diagonal elements of the submatrix corresponding to each handcart are the mass and moment of inertia of a single handcart; the coupling inertia term between the chassis and each handcart is calculated in real time based on the geometric offset between them (obtained from the nesting depth and yaw angle through forward kinematics). The Coriolis force and centrifugal force matrices are calculated by substituting the generalized velocity into the standard Christofel formula, specifically using a numerical method: taking the partial derivative of each element of the inertia matrix with respect to the generalized coordinates and multiplying it by the corresponding generalized velocity component before combining them. Under flat indoor airport conditions, since the mobile chassis and trolley are both in a horizontal position with no significant slope, all components of the gravity vector are set to zero.

[0060] The constraint Jacobian matrix is ​​pre-derived and stored analytically based on the physical geometry of the nested guide structure of the handcarts. Taking the nesting depth constraint as an example, the ideal nesting state is that the nesting depth equals the preset target value, and the geometric constraint equation is written as the nesting depth minus the target value equals zero. Each row of the constraint Jacobian matrix is ​​the vector of partial derivatives of the constraint equation with respect to the generalized coordinates. Specifically, only the position corresponding to the nesting depth in this row vector is 1, and the rest are 0. For multiple handcarts, the constraint row vectors corresponding to each cart are stacked vertically to form a complete constraint Jacobian matrix. In each control cycle (50 times per second), the system calculates the specific value of the constraint Jacobian matrix based on the current generalized coordinates. The Lagrange multipliers in the constraint force terms are obtained by solving the inverse solution of the multibody dynamics equations: the dynamic model established in subsequent steps is rearranged, and the known inertia matrix multiplied by the acceleration, Coriolis force, damping, gravity, and control input terms are substituted in, and the numerical solution of the Lagrange multipliers is obtained using the least squares method. Each dimension of the multiplier corresponds to a generalized force that needs to be applied in a constraint direction. The unit for translational constraints is Newton, and the unit for rotational constraints is Newton-meter.

[0061] Substituting the above results into the multi-vehicle coupled dynamics model, where the system viscous damping coefficient matrix is... The chassis was calibrated through no-load pushing tests, yielding the equivalent damping coefficient obtained by measuring the thrust required to maintain a constant speed of 0.5 meters per second and dividing it by the speed. The main diagonal elements were set to this value, while the off-diagonal elements were set to zero. The control input mapping matrix was determined by the kinematic model of the chassis drive mechanism: for a differential chassis, this matrix maps the left and right wheel drive forces to the chassis linear and angular accelerations; for the robotic arm joints, this matrix maps the torque commands of each joint to the equivalent force of the end effector in the generalized coordinate direction. The cooperative control inputs were generated in real-time by the upper-level controller, for example, using model predictive control or a quadratic programming solver. This dynamic model was numerically integrated in the embedded controller at an iteration frequency of 10,000 times per second to predict the system state and generate control commands for the next moment.

[0062] For each trolley to be stacked, a docking geometric equality constraint is established. Specifically, let the target nesting depth to be achieved after the i-th trolley is docked be set, which is pre-determined to be 0.15 meters by measuring the length of the trolley's guide groove. The geometric constraint expression is that the nesting depth minus 0.15 equals zero. This constraint is used as an equality constraint condition in the controller, and is enforced by using the Lagrange multiplier method or constraint projection method to ensure that after the stacking operation is completed, the deviation between the actual measured value of the nesting depth and the target value is less than a preset tolerance, for example, ±0.005 meters. The controller monitors this deviation in real time. Once the deviation exceeds the tolerance range for three consecutive control cycles, it is determined that the docking is not completed, and the system does not proceed to the stacking process of the next trolley.

[0063] The anti-collision inequality constraints include the following three specific forms: First, a lower limit constraint on nesting depth: the actual nesting depth minus the minimum allowable nesting depth is greater than or equal to zero. The minimum allowable nesting depth is set at 0.05 meters to prevent the stacked queue from separating during movement due to shallow docking. Second, an upper limit constraint on nesting depth: the maximum allowable nesting depth minus the actual nesting depth is greater than or equal to zero. The maximum allowable nesting depth is set at 0.18 meters to prevent excessive pushing that could cause plastic deformation of the trolley frame or damage to the guide structure. Third, a constraint on the clearance between the stacked queue and the external environment: the system deploys four ultrasonic sensors around the chassis (e.g., one each at the front, back, left, and right, with a detection range of 0.1 to 3 meters and an accuracy of ±0.01), measuring the minimum distance between the queue and obstacles such as guardrails, walls, and passengers in real time. The constraint expression is: the minimum measured distance minus the minimum safe clearance is greater than or equal to zero. The minimum safe clearance is set at 0.1 meters. The aforementioned inequality constraints are checked in each local planning iteration. When any inequality is about to be violated, for example, if it is predicted that the minimum measured distance will drop below 0.1 meters in the next 0.5 seconds, the controller immediately performs deceleration or braking and replans the path to avoid collision.

[0064] In this optional embodiment, generalized coordinates are accurately obtained by real-time fusion of wheel odometer, inertial measurement unit, depth camera, and visual keypoint detection, providing reliable input for subsequent modeling. Furthermore, based on pre-calibrated mass and moment of inertia parameters, the inertia matrix, Coriolis force matrix, and damping matrix are dynamically constructed. Constraint Jacobian matrices and Lagrange multipliers are introduced according to the geometric relationship of the nested guide structure, thereby establishing a multi-vehicle coupled dynamic model containing constraint force terms. This model can accurately describe the inertial coupling and constraint reaction forces generated by the chassis motion on each vehicle in the platoon through physical connections, enabling the controller to anticipate and suppress adverse dynamics such as tail sway during cornering and impact transmission during braking. Based on this, by setting equality constraints on nesting depth, such as a target depth of 0.15 meters and a tolerance of ±0.005 meters, as well as inequality constraints such as upper and lower limits of nesting depth (0.05–0.18 meters) and environmental safety clearance (0.1 meters), the physical limits and anti-collision requirements of the stacking process are transformed into hard constraints that can be checked online. In actual operation, the controller uses this model and constraints for real-time optimization. When any inequality is predicted to be violated, it immediately decelerates or brakes, thus proactively preventing hard impacts, excessive squeezing, queue separation, or pedestrian sweeping. Simultaneously, the constraint forces output by the dynamic model can be used for jamming detection: if the constraint force abnormally spikes and the nesting depth stagnates, the system automatically executes a withdrawal-lateral shift-realignment process, significantly improving the success rate and operational stability of continuous stacking. This embodiment, while ensuring the geometric alignment accuracy and structural compactness of the stacked queue, effectively suppresses instability factors caused by multi-body coupling, solving the engineering pain points of jamming, skewing, and scraping during trolley stacking, and achieving safe and reliable autonomous handling and queued operations in the dynamic environment of an airport.

[0065] Optionally, generating coordinated control commands for the mobile chassis and the robotic arm based on the multi-vehicle coupled dynamics model, the geometric constraints, and the collision avoidance constraints includes: By combining the multi-vehicle coupled dynamics model, the docking geometric constraints, and the anti-collision constraints, a dynamic equation containing the constraints is constructed. Solving the dynamic equations yields coordinated control commands that satisfy the docking geometric constraints and the anti-collision constraints.

[0066] Specifically, the multi-vehicle coupled dynamics model, docking geometric constraints, and anti-collision constraints are combined to construct system equations containing constraints. The method of combining these equations involves using the multi-vehicle coupled dynamics model (i.e., the equations describing the relationship between the inertial forces, Coriolis forces, damping forces, gravity, control forces, and constraint forces of the chassis and the stacked queue) as the fundamental equations of system motion, while simultaneously treating the docking geometric equality constraints and anti-collision inequality constraints as additional conditions that must be satisfied at the same time. In engineering implementation, to transform this continuous-time equation into a numerically solvable form, the controller employs a discrete-time model predictive control framework. The multi-vehicle coupled dynamics model is forward Euler discretized at a fixed time step, for example, 0.05 seconds, to obtain discrete state transition equations. Then, the docking geometric equality constraints are processed into hard constraints that allow for small errors, and embedded into the equality constraint set of the optimization problem using the Lagrange multiplier method. The collision avoidance inequality constraints are transformed into a series of linear inequalities in the prediction time domain. For example, it is required that the nesting depth always lies between a preset minimum and maximum value in multiple future control cycles, and that the minimum distance between the stacked queue and the obstacle is not less than the safety gap. Next, the discretized multi-vehicle coupled dynamics model and constraints are solved. The solution uses an online quadratic programming algorithm: with the current generalized coordinates and generalized velocity as the initial state, the control input sequence, including chassis linear velocity and angular velocity commands, and the torque or end-effector pose adjustment of each joint of the robotic arm, are used as optimization variables. The objectives are to track the desired trajectory, minimize control energy, and satisfy all constraints. A convex quadratic programming problem is constructed. The objective function of the convex quadratic programming problem includes a quadratic form of the tracking error and a quadratic form of the control increment. The constraints include the aforementioned equality constraints, inequality constraints, and actuator saturation limits. The solver, such as using the OSQP or qpOASES library, quickly solves the quadratic programming problem in each control cycle, outputting the optimal control input sequence that satisfies all constraints. The first value of the sequence is taken as the cooperative control command for the current cycle. This command is simultaneously sent to the chassis controller and the robotic arm controller, enabling coordinated movement between the chassis and the robotic arm while satisfying docking geometric constraints and collision avoidance constraints. If no feasible solution is found during the solution process, the system automatically reduces the desired speed or triggers a safety stop, and then recalculates the reference trajectory before solving again. In this way, the system can generate cooperative control commands online that satisfy multi-vehicle coupling dynamics, docking geometric constraints, and collision avoidance constraints, ensuring the safety and stability of the continuous stacking process.

[0067] In this optional embodiment, by simultaneously solving the multi-vehicle coupled dynamics model, docking geometric equality constraints, and collision avoidance inequality constraints within the model predictive control framework, online collaborative control commands for the chassis and robotic arm that satisfy multiple constraints can be generated. This embodiment treats requirements such as nesting depth target accuracy, depth upper and lower limits, and environmental safety clearances as hard constraints or penalty terms in the optimization problem, helping to reduce risks such as hard collisions, excessive squeezing, queue separation, or pedestrian sweeping caused by command deviations from constraints. Simultaneously, based on the predictive capabilities of the multi-vehicle coupled dynamics model, the controller can anticipate the swaying and impact trends generated at the tail of the queue when the chassis turns or brakes, and appropriately suppress them in the commands, thereby improving the stability and compactness of the stacked queue during movement. When no feasible solution is found, the speed can be automatically reduced or a safe stop can be triggered for replanning, enhancing robustness in complex dynamic scenarios to a certain extent.

[0068] Combination Figure 2 As shown, an embodiment of the present invention provides an autonomous collection and flexible stacking system for handcarts, applied to a robot system. The robot system includes a mobile chassis and a robotic arm. The autonomous collection and flexible stacking system for handcarts includes: A multimodal perception fusion unit is used to acquire multimodal perception data and perform fusion recognition on the multimodal perception data to obtain the handcart target and the pose information of the handcart target; The gripping and docking unit is used to perform gripping and compliant docking on the handcart target by the robotic arm according to the pose information, and to obtain the gripped handcart and the clamping status confirmation signal. The motion planning and control unit is used to plan a driving trajectory based on the current environmental perception information and the status of the grabbed handcart, under the conditions of satisfying the narrow passage passage constraint and the minimum turning radius constraint, control the mobile chassis to track the driving trajectory, and automatically adjust the driving speed of the mobile chassis according to the risk assessment results of the dynamic crowd, so as to transport the grabbed handcart to the stacking area to obtain the current stacking queue status. A continuous stacking unit is used to perform continuous stacking of multiple carts based on the already grabbed carts and the current stacking queue state; the continuous stacking unit includes: The state definition subunit is used to define generalized coordinates, which include the pose of the mobile chassis, the relative nesting depth between each trolley, and the relative yaw angle between each trolley. The model building subunit is used to construct a multi-vehicle coupled dynamic model based on the generalized coordinates, and to establish docking geometric constraints and anti-collision constraints; The instruction generation subunit is used to generate coordinated control instructions for the mobile chassis and the robotic arm based on the multi-vehicle coupled dynamics model, the geometric constraints, and the anti-collision constraints. The execution subunit is used to control the mobile chassis to perform the main traction motion according to the cooperative control command, and at the same time control the robotic arm to perform anisotropic impedance control for attitude fine-tuning, so as to nest the grabbed handcart into the tail of the stacking queue to obtain the updated stacking queue. The queue arrangement and parking unit is used to move the updated stacked queue to the target recycling point, thereby completing the queue arrangement and parking of the handcart.

[0069] The advantages of the handcart autonomous collection and flexible stacking system of the present invention compared with the prior art are the same as those of the above-mentioned handcart autonomous collection and flexible stacking method compared with the prior art, and will not be repeated here.

[0070] Combination Figure 3 As shown, an electronic device according to an embodiment of the present invention includes: a processor and a memory, wherein the memory is used to store computer programs; When the computer program is loaded by the processor, it causes the processor to execute the above-described method for autonomous collection and flexible stacking of handcarts.

[0071] The electronic device of the present invention has the same advantages over the prior art as the aforementioned autonomous collection and flexible stacking method of the handcart, and will not be repeated here.

[0072] An embodiment of the present invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for autonomous collection and flexible stacking of handcarts.

[0073] The computer-readable storage medium of the present invention has the same advantages over the prior art as the aforementioned handcart autonomous collection and flexible stacking method over the prior art, and will not be repeated here.

[0074] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A method for autonomous collection and flexible stacking of handcarts, characterized in that, Applied to a robotic system, the robotic system including a mobile chassis and a robotic arm, the method for autonomous collection and flexible stacking of handcarts includes: Acquire multimodal sensing data and fuse and identify the multimodal sensing data to obtain the handcart target and the pose information of the handcart target; Based on the pose information, the robotic arm performs grasping and compliant docking on the handcart target to obtain the grasped handcart and a clamping status confirmation signal. Based on the current environmental perception information and the status of the grabbed handcarts, a driving trajectory is planned under the conditions of satisfying the narrow passage passage constraint and the minimum turning radius constraint. The mobile chassis is controlled to track the driving trajectory. At the same time, the driving speed of the mobile chassis is automatically adjusted according to the risk assessment results of the dynamic crowd, and the grabbed handcarts are transported to the stacking area to obtain the current stacking queue status. Based on the already grabbed trolleys and the current stacking queue state, perform continuous stacking of multiple trolleys: Define a generalized coordinate system, which includes the pose of the mobile chassis, the relative nesting depth between each trolley, and the relative yaw angle between each trolley. A multi-vehicle coupled dynamics model is constructed based on the generalized coordinates, and docking geometric constraints and anti-collision constraints are established. The construction of the multi-vehicle coupled dynamics model includes: establishing a multi-vehicle coupled system dynamics model based on the generalized coordinates; and using a continuously stacked coupled system dynamics model to uniformly describe the nonlinear coupled motion relationship between the chassis and multiple handcarts. ,in, This represents the mass and inertia matrix of a multibody system. Represents a generalized coordinate vector. Represents the generalized velocity vector, i.e. The first derivative with respect to time, Represents the generalized acceleration vector, i.e. The second derivative with respect to time, Represents the matrix of Coriolis force and centrifugal force terms. This represents the system's viscous damping coefficient matrix. This represents the equivalent gravitational and elastic potential energy terms. This represents the control input mapping matrix. This indicates that the chassis and the control arm work together to control the input. This represents the constraint Jacobian matrix of the system. This represents the equivalent constraint term or Lagrange multiplier. express The transpose of the matrix; Based on the multi-vehicle coupled dynamics model, the geometric constraints, and the collision avoidance constraints, collaborative control commands for the mobile chassis and the robotic arm are generated. According to the cooperative control command, the mobile chassis is controlled to perform the main traction motion, and the robotic arm is controlled to perform anisotropic impedance control to perform attitude fine-tuning, so that the grabbed handcart is nested into the tail of the stacking queue to obtain the updated stacking queue. The updated stacked queue is moved to the target recycling point to complete the queue arrangement and parking of the handcarts.

2. The method for autonomous collection and flexible stacking of handcarts according to claim 1, characterized in that, The process of acquiring multimodal sensing data and fusing and recognizing the multimodal sensing data to obtain the handcart target and its pose information includes: Acquire multimodal sensing data from at least two types of sensors, including a surround-view camera, a 3D LiDAR, or a depth camera; The multimodal sensing data is time-synchronized and extrinsic parameters are calibrated to obtain aligned sensing data with a unified time reference and a unified spatial coordinate system. The aligned perception data is input into the target detection model to obtain the detection box and key points of the handcart target; Based on the detection box and the key points, and combined with the 3D point cloud data in the aligned perception data, the 3D position and spatial pose of the handcart target are obtained, and the 3D position and spatial pose are used as the pose information of the handcart target.

3. The method for autonomous collection and flexible stacking of handcarts according to claim 2, characterized in that, The step of grasping and compliantly docking the handcart target with the robotic arm based on the pose information, and obtaining a grasped handcart and a clamping status confirmation signal, includes: Based on the pose information, an approach trajectory for the robotic arm is generated, and the robotic arm is controlled to move to the initial docking position of the handcart target based on the approach trajectory. After the robotic arm reaches the initial docking position, the robot system enters the visual servoing stage, using the key points of the handcart target as feedback to correct the pose of the robotic arm end in real time, so that the guide structure of the robotic arm end is aligned with the docking entrance of the handcart target. After the end of the robotic arm contacts the target trolley, the robot system switches to impedance control mode to correct the pose of the end of the robotic arm to compensate for the docking pose deviation. Based on the online estimated coefficient of friction of the contact surface and the load mass of the handcart target, the required clamping force is obtained, and the gripper of the robotic arm is controlled to clamp the handcart target with the required clamping force according to the required clamping force. The displacement of the gripper, tactile feedback, and reprojection error of the key points are detected. When the displacement, tactile feedback, and reprojection error are all within their respective preset ranges, the gripping is confirmed to be successful, and a signal confirming the gripped trolley and the gripping status is obtained.

4. The method for autonomous collection and flexible stacking of handcarts according to claim 1, characterized in that, The step of planning a driving trajectory based on the current environmental perception information and the status of the already grasped handcart, while satisfying the narrow passage passage constraint and the minimum turning radius constraint, and controlling the mobile chassis to track the driving trajectory includes: Acquire current environmental perception information and the status of the grabbed handcart, wherein the current environmental perception information includes the channel width and the location of obstacles; Update the equivalent envelope width of the mobile chassis based on the status of the grabbed handcart; Based on the equivalent envelope width and the channel width, the required safety gap is obtained, and based on the safety gap, it is determined whether the narrow channel passage constraint is met. Based on the preset minimum turning radius, the maximum allowable curvature of the trajectory is set, and the minimum turning radius constraint is determined based on the maximum allowable curvature of the trajectory. Under the premise of satisfying the narrow passage constraint and the minimum turning radius constraint, and in combination with the obstacle position, a driving trajectory from the current position of the mobile chassis to the target position of the stacking area is planned, and the mobile chassis is controlled to follow the driving trajectory.

5. The method for autonomous collection and flexible stacking of handcarts according to claim 4, characterized in that, The step of automatically adjusting the travel speed of the mobile chassis based on the risk assessment results of the dynamic population, and transporting the grabbed handcarts to the stacking area to obtain the current stacking queue status, includes: Acquire dynamic crowd perception data, including the location of surrounding pedestrians; Based on the position of the mobile chassis and the position of each of the surrounding pedestrians, the distance between the mobile chassis and each of the surrounding pedestrians is determined, resulting in multiple distances. The minimum value among the multiple distances is selected as the minimum distance between the mobile chassis and the surrounding pedestrians. A speed adjustment coefficient is determined based on the minimum distance, wherein the speed adjustment coefficient is positively correlated with the minimum distance; Based on the desired speed of the mobile chassis and the speed adjustment coefficient, a target speed is obtained, and the mobile chassis is controlled to travel towards the stacking area at the target speed. After the mobile chassis arrives at the stacking area, the relative pose information of each trolley in the existing stacking queue is obtained, and the relative pose information is used as the current stacking queue state.

6. The method for autonomous collection and flexible stacking of handcarts according to claim 5, characterized in that, The construction of a multi-vehicle coupled dynamics model based on the generalized coordinates, and the establishment of docking geometric constraints and anti-collision constraints, include: Obtain the generalized coordinates; Based on the generalized coordinates, determine the inertia matrix, Coriolis force matrix, and gravity vector of the coupled system consisting of the mobile chassis, the grabbed handcart, and the existing stacked queue. Introducing a constraint Jacobian matrix to determine the constraint force terms of the coupled system; The multi-vehicle coupled dynamics model is constructed based on the inertia matrix, the Coriolis force matrix, the gravity vector, and the constraint force terms. Based on the generalized coordinates, establish the docking geometric constraints between each of the handcarts during the docking process; Based on the generalized coordinates, anti-collision constraints are established between each of the handcarts and between the handcarts and the external environment.

7. The method for autonomous collection and flexible stacking of handcarts according to claim 1, characterized in that, The step of generating coordinated control commands for the mobile chassis and the robotic arm based on the multi-vehicle coupled dynamics model, the geometric constraints, and the collision avoidance constraints includes: By combining the multi-vehicle coupled dynamics model, the docking geometric constraints, and the anti-collision constraints, a dynamic equation containing the constraints is constructed. Solving the dynamic equations yields coordinated control commands that satisfy the docking geometric constraints and the anti-collision constraints.

8. A handcart autonomous collection and flexible stacking system, characterized in that, Applied to a robotic system, the robotic system includes a mobile chassis and a robotic arm, and the autonomous collection and flexible stacking system for the handcart includes: A multimodal perception fusion unit is used to acquire multimodal perception data and perform fusion recognition on the multimodal perception data to obtain the handcart target and the pose information of the handcart target; The gripping and docking unit is used to perform gripping and compliant docking on the handcart target by the robotic arm according to the pose information, and to obtain the gripped handcart and the clamping status confirmation signal. The motion planning and control unit is used to plan a driving trajectory based on the current environmental perception information and the status of the grabbed handcart, under the conditions of satisfying the narrow passage passage constraint and the minimum turning radius constraint, control the mobile chassis to track the driving trajectory, and automatically adjust the driving speed of the mobile chassis according to the risk assessment results of the dynamic crowd, so as to transport the grabbed handcart to the stacking area to obtain the current stacking queue status. A continuous stacking unit is used to perform continuous stacking of multiple carts based on the already grabbed carts and the current stacking queue state; the continuous stacking unit includes: The state definition subunit is used to define generalized coordinates, which include the pose of the mobile chassis, the relative nesting depth between each trolley, and the relative yaw angle between each trolley. The model construction subunit is used to construct a multi-vehicle coupled dynamics model based on the generalized coordinates, and to establish docking geometric constraints and anti-collision constraints; wherein, the construction of the multi-vehicle coupled dynamics model includes: establishing a multi-vehicle coupled system dynamics model based on the generalized coordinates, and a continuously stacked coupled system dynamics model is used to uniformly describe the nonlinear coupled motion relationship between the chassis and multiple handcarts. ,in, This represents the mass and inertia matrix of a multibody system. Represents a generalized coordinate vector. Represents the generalized velocity vector, i.e. The first derivative with respect to time, Represents the generalized acceleration vector, i.e. The second derivative with respect to time, Represents the matrix of Coriolis force and centrifugal force terms. This represents the system's viscous damping coefficient matrix. This represents the equivalent gravitational and elastic potential energy terms. This represents the control input mapping matrix. This indicates that the chassis and the control arm work together to control the input. This represents the constraint Jacobian matrix of the system. This represents the equivalent constraint term or Lagrange multiplier. express The transpose of the matrix; The instruction generation subunit is used to generate coordinated control instructions for the mobile chassis and the robotic arm based on the multi-vehicle coupled dynamics model, the geometric constraints, and the anti-collision constraints. The execution subunit is used to control the mobile chassis to perform the main traction motion according to the cooperative control command, and at the same time control the robotic arm to perform anisotropic impedance control for attitude fine-tuning, so as to nest the grabbed handcart into the tail of the stacking queue to obtain the updated stacking queue. The queue arrangement and parking unit is used to move the updated stacked queue to the target recycling point, thereby completing the queue arrangement and parking of the handcart.

9. An electronic device, characterized in that, include: Processor and memory, the memory being used to store computer programs; When the computer program is loaded by the processor, it causes the processor to execute the handcart autonomous collection and flexible stacking method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the autonomous collection and flexible stacking method for handcarts as described in any one of claims 1 to 7.

Images