Supermarket shopping robot control method and system adopting intelligent path planning

Abstract

The invention discloses a supermarket shopping robot control method and system adopting intelligent path planning, relates to the technical field of path planning, and aims to collect environment and robot state data, receive user commodity positions and provide reliable data. Grids are divided based on a map, the pose of the robot A is analyzed, a people flow density map is generated, the track of the robot B is predicted, and the dynamic environment is sensed. And calculating space-time risk cost based on the people flow density and the predicted trajectory of the robot B, generating a candidate path set, and providing diversified intelligent options. And the user selects a path, and the robot A starts along the selected path, observes and records advancing data in real time, fuses user will and starts execution. Iterative optimization is carried out based on the advancing data and the path report, the data are added to a historical database, and continuous improvement of the system performance is achieved. According to the invention, by fusing multi-source perception, a dynamic environment, intelligent path generation and user interaction, the path planning efficiency and safety of the robot in a complex supermarket environment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning, and in particular to a control method and system for a supermarket shopping robot using intelligent path planning. Background Art

[0002] With the rapid development of automation and artificial intelligence technologies, robotics is gradually penetrating all aspects of social life, showing enormous potential in the service industry in particular. Among the many service robot applications, mobile robots for commercial retail environments have attracted considerable attention due to their potential to improve operational efficiency and customer experience. In large retail locations, such as modern supermarkets, customers often face challenges such as time-consuming search for items and the inconvenience of carrying heavy objects. To address this, a class of specially designed assistive mobile robots has emerged. These robots can autonomously navigate complex supermarket environments, performing tasks such as guiding customers, delivering items, or accompanying customers and carrying their purchases. These devices, commonly referred to as "supermarket shopping robots," have a core capability: their ability to safely and efficiently navigate dynamic, crowded indoor spaces. Therefore, research and development of control methods and systems for these robots, particularly technologies that implement intelligent path planning, is crucial for promoting smart retail and improving the level of automated services.

[0003] Current supermarket shopping robots still face significant challenges in path planning and control in practical applications. While existing navigation methods can handle basic static obstacle avoidance and simple dynamic target avoidance, they often struggle to cope with the complex environments unique to supermarkets. Traditional path planning algorithms typically treat pedestrians as isolated, unpredictable obstacles and passively avoid them, failing to effectively consider and utilize macroscopic information about "crowd density" in the environment. They lack the ability to perceive and predict the level of congestion in an area, which can lead robots to choose routes that are theoretically the shortest but, in reality, extremely inefficient due to crowd congestion. Furthermore, when multiple robots work together in the same area, existing control systems often lack effective coordination mechanisms. Robots usually make decisions based on their own local perception and planning, without considering the real-time position, motion status or even future action "intentions" of other collaborative robots. What's more serious is that in multi-robot scenarios, independent obstacle avoidance behaviors can easily lead to conflicts. For example, a robot suddenly changes lanes to avoid a customer, which happens to block the path of another robot, causing "traffic jams" between robots and even increasing the risk of collision. This not only affects the normal operation of the robots, but may also cause inconvenience or safety hazards to customers, ultimately weakening the original intention of introducing the robot system to improve efficiency and enhance the shopping experience. Summary of the invention

[0004] In view of the deficiencies of the prior art, the present invention provides a control method and system for a supermarket shopping robot using intelligent path planning, which solves the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A control method for a supermarket shopping robot using intelligent path planning, comprising the following steps:

[0006] S1. Obtain environmental data Ei through sensors configured in the supermarket, and through sensors and communication modules carried by robot A, collect image data around robot A in real time and receive the status data ORS_B of robot B. The user inputs specified commodities to robot A through the operation interface and locates the positions of the commodities.

[0007] S2. Divide the supermarket environment into a grid coordinate system Gcs based on the supermarket map data Map, calculate the precise pose information POSE_A of robot A, generate a crowd density map CDM in combination with the environmental data Ei, and generate a predicted motion path TRJ_B(Δt) of robot B within a future time window Δt based on the status data ORS_B of robot B.

[0008] S3. Based on the precise pose information POSE_A of robot A, the crowd density map CDM, and the predicted motion path TRJ_B(Δt) of robot B within a future time window Δt, calculate the spatio-temporal risk increment cost DC, and generate a candidate path set Path based on the spatio-temporal risk increment cost DC.

[0009] S4. Based on the generated candidate path set Path, the user selects a path and marks the selected path as Path_obt, generates a path planning report AS, and robot A departs along the path Path_obt and observes the path in real time, records the travel data, and generates a feedback data set Fh.

[0010] S5. Based on the path planning report As and the feedback data set Fh, perform iterative optimization and add the path planning report As and the actual feedback data set Fh to the historical database His.

[0011] Preferably, S1 includes S11;

[0012] S11. Robot A scans the surrounding environment through the lidar it carries, obtains the original point cloud data set PCD_A of Robot A in the current cycle, captures the images within its field of view through the camera it carries, obtains the original image frame data IMG_A of Robot A in the current cycle, records the wheel rolling information through the wheel odometer it carries, obtains the original odometer reading ODO_A of Robot A in the current cycle, listens to and decodes the status data ORS_B broadcast by Robot B in the network through the communication module it carries, obtains the environmental point cloud data Ei through the lidar configured in the supermarket, and the user inputs the waiting time T_user and the specified commodities to Robot A through the operation interface. Robot A locates the positions of the commodities through the supermarket map data Map;

[0013] Among them, the status data ORS_B of Robot B includes the original pose information POSE_B of Robot B, the speed information VEL_B of Robot B, the acceleration vector ACC_B of Robot B, and the short-term intention path ITP_B of Robot B.

[0014] Preferably, S2 includes S21 and S22;

[0015] S21. Generate the supermarket environment into a map grid coordinate system Gcs based on the supermarket map data Map, perform point cloud downsampling processing on the original point cloud data set PCD_A of Robot A in the current cycle to obtain the point cloud data PCD_Apro of Robot A, perform Matlab distortion correction processing on the original image frame data IMG_A of Robot A in the current cycle to obtain the image frame data IMG_Apro of Robot A, and run the SLAM positioning algorithm in combination with the point cloud data PCD_Apro of Robot A, the image frame data IMG_Apro of Robot A in the current cycle, the original odometer reading ODO_A of Robot A in the current cycle, and the grid coordinate system Gcs to obtain the accurate pose information POSE_A of Robot A in the current map grid coordinate system Gcs;

[0016] The map grid coordinate system Gcs is composed of a set of grid cells C = {c1, c2,..., cm}, m represents the total number of grid cells. Combine the environmental point cloud data Ei to construct a pedestrian flow density estimation algorithm and calculate the pedestrian flow density ρ_ci of the grid cell ci;

[0017] According to the status data ORS_B of Robot B, construct a trajectory prediction algorithm to predict the predicted position vector P_B(Δt) of Robot B within the future time window Δt.

[0018] Preferably, S22. Among them, the expression of the pedestrian flow density estimation algorithm is as follows:

[0019] ;

[0020] ;

[0021] In the formula, i represents the grid cell number, p represents the point cloud point of the environmental point cloud data Ei in the grid cell ci, v_p represents the average moving speed of the point p in the grid cell ci, A_ci represents the area of the grid cell ci, α represents a preset speed influence factor, w_h(p_z) represents the height weight function. If the value of the height weight function w_h(p_z) is 0, it means that the point p is not included in the calculation of the pedestrian flow density. If the value of the height weight function w_h(p_z) is 1, it means that the point p is included in the calculation of the pedestrian flow density. p_z represents the vertical coordinate of the point p, Hgro_max represents the maximum ground vertical height threshold, and Hhum_max represents the maximum high-altitude vertical height threshold;

[0022] Among them, the expression of the trajectory prediction algorithm is as follows:

[0023] ;

[0024] Combining the map grid coordinate system Gcs and the pedestrian flow density ρ_ci of the grid cell ci to generate a pedestrian flow density map CDM, and combining the short-term intention path ITP_B of the robot B and the predicted position vector POB_B(Δt) of the robot B within the future time window Δt to generate the predicted motion path TRJ_B(Δt) of the robot B within the future time window Δt.

[0025] Preferably, S3 includes S31 and S32;

[0026] S31. Based on the map grid coordinate system Gcs, combining the accurate pose information POSE_A of the robot A and the commodity positioning to determine the starting grid cell Sc and the target grid cell Gc, initializing the number of path scheme strategies k = 3, and based on the number of path scheme strategies, the set of strategy parameters Params_k = {β_k, ζ_k, Dsg_k, δ_k} of the k-th type is set in advance by the designer. The three preset path scheme strategies are: a balanced scheme, a risk aversion scheme, and a speed priority scheme;

[0027] Among them, β_k represents the density influence coefficient set under the k-th strategy, ζ_k represents the risk penalty weight set under the k-th strategy, Dsg_k represents the safety distance threshold set under the k-th strategy, and δ_k represents the distance influence index set under the k-th strategy;

[0028] Based on the crowd density map CDM and the predicted movement path TRJ_B(Δt) of robot B within the future time window Δt, a strategic spatio-temporal risk incremental cost algorithm is constructed to calculate the spatio-temporal risk incremental cost DC_k of each grid cell based on the k-th strategy, and record the spatio-temporal risk incremental costs DC_k of all grid cells in the grid spatio-temporal risk incremental cost set DCth_k of the k-th strategy;

[0029] Based on the grid spatio-temporal risk incremental cost set DCth_k of the k-th strategy, use the dijkstra algorithm to find the optimal candidate path Path_k of the k-th strategy, and generate a candidate path set Path by combining the optimal candidate paths of all strategies.

[0030] Preferably, S32, the strategic spatio-temporal risk incremental cost algorithm consists of a basic time cost algorithm and a spatio-temporal risk factor algorithm;

[0031] Among them, the expression of the basic time cost algorithm is as follows:

[0032] ;

[0033] In the formula, T_k(ci) represents the basic time cost of moving in the grid cell ci in the k-th strategy, L_seg represents the geometric distance of moving in the grid cell ci, V_max represents the maximum speed of the robot, ρ_ci represents the crowd density of the grid cell ci, exp represents the natural exponential function, and m represents the total number of grid cells;

[0034] Among them, the expression of the spatio-temporal risk factor algorithm is as follows:

[0035] ;

[0036] In the formula, R_k(ci) represents the spatio-temporal risk factor of the robot B when moving in the grid cell ci in the k-th strategy, POS_ci represents the position where the simulated robot A walks to the grid cell ci, T_ci represents the predicted time when the robot A reaches the grid cell ci in the simulation, POS_B(T_ci) represents the position of the robot B in the predicted movement path TRJ_B(Δt) at the time T_ci, and dist(POS_ci, POS_B(T_ci)) represents the Euclidean distance between the simulated robot A and the robot B at the predicted time T_ci;

[0037] Among them, the expression of the spatio-temporal risk incremental cost algorithm is as follows;

[0038] ;

[0039] Wherein, DC_k(ci) represents the spatio-temporal risk incremental cost of moving in the grid cell ci under the k-th strategy.

[0040] Preferably, S4 includes S41 and S42;

[0041] S41. Based on the generated candidate path set Path, smooth the paths in the candidate path set Path and create a visualization interface. The information fed back by the visualization interface includes a path map, the pedestrian flow density ρ on the path, and the path prediction time T obtained by accumulating the basic path grid time costs of the path grid cells. The user selects a path and marks the selected path as Path_obt. By recording the path map of the path Path_obt and the path prediction passing time TPath_obt obtained by accumulating the basic time costs of the grid cells on the path Path_obt, a path planning report AS is generated.

[0042] Preferably, S42. The robot A departs according to the path Path_obt selected by the user, and observes the path in real time through the lidar, camera and wheel odometer carried by the robot A. When it detects an impassable obstacle on the front path, based on the grid cell where the current position is located as the new starting grid cell Sc_new, the strategy parameter set Params of the path strategy solution selected by the user is used to re-plan the path, record the actual path passing time T, the number of times of re-planning the path Ren and the position of re-planning the path Rep, and generate a feedback data set Fh.

[0043] Preferably, S5 includes S51;

[0044] S51. Based on the path planning report As and the feedback data set Fh, perform iterative optimization;

[0045] If the actual path passing time T minus the path prediction passing time TPath_obt ≥ 30s, then the professional staff adjusts the preset speed influence factor α and the density influence coefficient β set under the current strategy;

[0046] If the actual path passing time T minus the path prediction passing time TPath_obt < 30s, then no adjustment is required;

[0047] If the number of times of planning the path Ren ≥ 2 times, then the professional staff adjusts the risk penalty weight ζ set under the current strategy, the safety distance threshold Dsg set under the current strategy, and the distance influence index δ set under the current strategy;

[0048] If the number of times of planning the path Ren < 2 times, then no adjustment is required;

[0049] Add the path planning report As and the feedback data set Fh to the historical database His.

[0050] A supermarket shopping robot control system adopting intelligent path planning, including a data acquisition module, a data analysis module, a path generation module, an execution module and an iterative optimization module;

[0051] The data acquisition module obtains environmental data Ei through sensors configured in the supermarket, and through sensors and communication modules carried by robot A, it collects image data around robot A in real time and receives the status data ORS_B of robot B. The user inputs specified commodities to robot A through the operation interface and locates the commodity positions;

[0052] The data analysis module divides the supermarket environment into a grid coordinate system Gcs based on the supermarket map data Map, calculates the precise pose information POSE_A of robot A, generates a crowd density map CDM in combination with the environmental data Ei, and generates a predicted motion path TRJ_B(Δt) of robot B within the future time window Δt based on the status data ORS_B of robot B;

[0053] The path generation module calculates the spatio-temporal risk increment cost DC based on the precise pose information POSE_A of robot A, the crowd density map CDM and the predicted motion path TRJ_B(Δt) of robot B within the future time window Δt, and generates a candidate path set Path based on the spatio-temporal risk increment cost DC;

[0054] The execution module selects a path from the generated candidate path set Path, marks the selected path as Path_obt, generates a path planning report AS, and robot A departs along the path Path_obt and observes the path in real time, records the travel data, and generates a feedback data set Fh;

[0055] The iterative optimization module performs iterative optimization based on the path planning report As and the feedback data set Fh, and adds the path planning report As and the actual feedback data set Fh to the historical database His.

[0056] The present invention provides a supermarket shopping robot control method and system adopting intelligent path planning, having the following beneficial effects:

[0057] (1) By introducing real-time perception and quantitative evaluation of the environmental pedestrian flow density, and combining with the prediction and collaborative consideration of the motion intentions of other collaborative robots, the present invention constructs a more comprehensive and intelligent dynamic path planning and coordination mechanism. This mechanism no longer simply treats pedestrians as isolated obstacles for passive avoidance, nor does it let multiple robots plan paths independently. Instead, through a unified optimization framework, it simultaneously seeks the optimal solutions for avoiding crowded areas and preventing potential conflicts among robots. This makes the operation of a single robot and even the entire robot group smoother and more efficient, significantly reducing situations such as waiting, sudden stops, and ineffective detours caused by congestion or conflicts, thereby improving the overall task execution efficiency and customer satisfaction in the supermarket scenario. At the same time, it also enhances the safety and system robustness of multi-robot collaborative operations, laying a solid foundation for the large-scale application of service robots in real complex environments.

[0058] (2) The environmental data Ei is obtained through the sensors configured in the supermarket, and the image data around robot A and the status data ORS_B of robot B are collected in real time through the sensors and communication module carried by robot A. Among them, the status data ORS_B of robot B includes the original pose information POSE_B, speed information VE_LB, acceleration vector ACC_B, and short-term intended path ITP_B of robot B. The user can also input the specified commodity and locate the commodity position through the operation interface. Based on the supermarket map data Map, the supermarket environment is divided into a grid coordinate system Gcs, and the accurate pose information POSE_A of robot A is calculated. More importantly, combined with the environmental data Ei, the system can generate a pedestrian flow density map CDM to quantify the pedestrian flow density ρci of each grid cell ci. At the same time, based on the status data ORS_B of robot B, the system can generate the predicted motion path TRJ_B(Δt) of robot B within the future time window Δt. This comprehensive perception and modeling of the pedestrian flow density and the future paths of collaborative robots overcome the shortcomings of insufficient information in traditional methods, providing accurate and dynamic environmental information for subsequent intelligent and collaborative path planning, and significantly improving the depth and breadth of environmental perception.

[0059] (3) Based on the precise pose information POSEA of robot A, the crowd density map CDM, and the predicted motion path TRJ_B(Δt) of robot B within the future time window Δt, calculate the spatio-temporal risk increment cost DC. This cost takes into account the basic time cost and spatio-temporal risk factors, comprehensively evaluating the potential risks and efficiencies of different path segments. Based on the spatio-temporal risk increment cost DC, the system can generate a candidate path set Path. The user can select a path Path_obt from the candidate path set Path, and the system generates a path planning report AS. Robot A departs along the selected path and observes and records the travel data in real time, generating a feedback data set Fh, including the actual path travel time T, the number of times of replanning the path Ren, and the replanning path position Rep. Finally, perform iterative optimization based on the path planning report AS and the feedback data set Fh. By comparing the actual path travel time T with the accumulated path prediction time TPath_obt, and analyzing the number of times of replanning the path Ren, the system can adjust the preset speed influence factor α and the set of policy parameters Params set by the current policy. At the same time, add the path planning report AS and the feedback data set Fh to the historical database His to achieve the continuous learning and optimization of the system. This combination of dynamic programming and iterative optimization enables the robot to quickly adapt to the new environment and continuously improve the efficiency and safety of path planning. Description of the Drawings

[0060] Figure 1 Schematic diagram of the steps of a supermarket shopping robot control method using intelligent path planning according to the present invention;

[0061] Figure 2 Schematic block diagram of a supermarket shopping robot control system using intelligent path planning according to the present invention;

[0062] Figure 3 Visualization display diagram of candidate path information. Detailed Embodiment

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0064] Embodiment 1

[0065] The present invention provides a supermarket shopping robot control method using intelligent path planning. Please refer to Figure 1 , including the following steps:

[0066] S1. Obtain environmental data Ei through sensors configured in the supermarket. Through the sensors and communication module carried by robot A, collect the image data around robot A in real time and receive the status data ORS_B of robot B. The user inputs the specified goods to robot A through the operation interface and locates the positions of the goods.

[0067] S2. Divide the supermarket environment into a grid coordinate system Gcs based on the supermarket map data Map. Calculate the precise pose information POSE_A of robot A. Combine the environmental data Ei to generate a crowd density map CDM. Based on the status data ORS_B of robot B, generate the predicted motion path TRJ_B(Δt) of robot B within the future time window Δt.

[0068] S3. Based on the precise pose information POSE_A of robot A, the crowd density map CDM, and the predicted motion path TRJ_B(Δt) of robot B within the future time window Δt, calculate the spatio-temporal risk incremental cost DC. Generate a candidate path set Path based on the spatio-temporal risk incremental cost DC.

[0069] S4. Based on the generated candidate path set Path, the user selects a path and marks the selected path as Path_obt. Generate a path planning report AS. Robot A departs along the path Path_obt and observes the path in real time, records the travel data, and generates a feedback data set Fh.

[0070] S5. Based on the path planning report As and the feedback data set Fh, perform iterative optimization and add the path planning report As and the actual feedback data set Fh to the historical database His.

[0071] In this embodiment, by fusing multi-source perception data with dynamic environment modeling, the navigation efficiency and collaborative safety of the robot in complex supermarket scenarios are significantly improved. Based on the grid coordinate system Gcs generated from the supermarket map data Map, combined with the environmental point cloud data Ei, the crowd density map CDM is constructed in real time, the crowd density ρ_ci of each grid cell ci is quantified, and through the state data ORS_B of robot B, including the pose information POSE_B, speed VEL_B, and short-term intention path ITP_B, the movement path TRJ_B(△t) of robot B within the future time window △t is predicted. On this basis, the system generates a diverse set of candidate paths Path including balanced, risk-averse, and speed-priority types through the set of policy parameters Params_k, and the user can select the optimal path Path_obt according to the needs. This method not only effectively avoids the path congestion problem caused by traditional technologies ignoring the crowd density ρ_ci, but also realizes multi-robot collaborative obstacle avoidance by predicting the TRJ_B(△t) of robot B, avoiding collisions and ineffective detours. At the same time, based on the iterative optimization mechanism of the path planning report As and the feedback data set Fh, the speed influence factor α and the set of policy parameters Params set by the current policy are dynamically adjusted, solving the defect of insufficient adaptability of the existing system, and finally significantly reducing the spatio-temporal risk of multi-robot collaborative operations while improving the path efficiency.

[0072] Embodiment 2

[0073] This embodiment is an explanatory note based on Embodiment 1. Please refer to Figure 1 , specifically: S1 includes S11;

[0074] S11. Robot A scans the surrounding environment through the mounted lidar to obtain the original point cloud data set PCD_A of the current cycle of robot A, captures the images within the field of view through the mounted camera to obtain the original image frame data IMG_A of the current cycle of robot A, records the wheel rolling information through the mounted wheel odometer to obtain the original odometer reading ODO_A of the current cycle of robot A, listens to and decodes the state data ORS_B broadcast by robot B in the network through the mounted communication module, obtains the environmental point cloud data Ei through the lidar configured in the supermarket, and the user inputs the waitable time T_user and the specified commodities to robot A through the operation interface. Robot A locates the commodity positions through the supermarket map data Map;

[0075] Among them, the state data ORS_B of robot B includes the original pose information POSE_B of robot B, the speed information VEL_B of robot B, the acceleration vector ACC_B of robot B, and the short-term intention path ITP_B of robot B;

[0076] S2 includes S21 and S22;

[0077] S21. Generate the supermarket environment as a map grid coordinate system Gcs based on the supermarket map data Map, perform point cloud downsampling on the original point cloud data set PCD_A of robot A in the current cycle to obtain the point cloud data PCD_Apro of robot A, perform Matlab distortion correction on the original image frame data IMG_A of robot A in the current cycle to obtain the image frame data IMG_Apro of robot A, and run the SLAM positioning algorithm in combination with the point cloud data PCD_Apro of robot A, the image frame data IMG_Apro of robot A in the current cycle, the original odometer reading ODO_A of robot A in the current cycle, and the grid coordinate system Gcs to obtain the accurate pose information POSE_A of robot A in the current map grid coordinate system Gcs;

[0078] The map grid coordinate system Gcs consists of a set of grid cells C = {c1, c2,..., cm}, where m represents the total number of grid cells. Combine the environmental point cloud data Ei to construct a pedestrian flow density estimation algorithm to calculate the pedestrian flow density ρ_ci of the grid cell ci;

[0079] According to the state data ORS_B of robot B, construct a trajectory prediction algorithm to predict the predicted position vector P_B(Δt) of robot B within the future time window Δt;

[0080] S22. Among them, the expression of the pedestrian flow density estimation algorithm is as follows:

[0081] ;

[0082] ;

[0083] In the formula, i represents the grid cell number, p represents the point cloud point in the grid cell ci in the environmental point cloud data Ei, v_p represents the average moving speed of the point p in the grid cell ci, A_ci represents the area of the grid cell ci, α represents the preset speed influence factor, w_h(p_z) represents the height weight function. If the value of the height weight function w_h(p_z) is 0, it means that the point p is not included in the calculation of the pedestrian flow density. If the value of the height weight function w_h(p_z) is 1, it means that the point p is included in the calculation of the pedestrian flow density. p_z represents the vertical coordinate of the point p, Hgro_max represents the maximum ground vertical height threshold, and Hhum_max represents the maximum high-altitude vertical height threshold;

[0084] Among them, the expression of the trajectory prediction algorithm is as follows:

[0085] ;

[0086] Generate a crowd density map CDM by combining the crowd density ρ_ci of the map grid coordinate system Gcs and the grid cell ci, and generate a predicted motion path TRJ_B(Δt) of robot B within the future time window Δt by combining the short-term intention path ITP_B of robot B and the predicted position vector POB_B(Δt) of robot B within the future time window Δt.

[0087] In this embodiment, the dynamic environment perception and prediction capabilities are enhanced through multi-modal data fusion and precise modeling. Specifically, robot A collects the original point cloud data set PCD_A through lidar, captures the original image frame data IMG_A through a camera, and obtains the original odometer reading ODO_A through a wheeled odometer. Combining with the point cloud data Ei of the supermarket environment, through point cloud downsampling, Matlab distortion correction, and SLAM positioning algorithm, precise pose information POSE_A is generated in the grid coordinate system Gcs. At the same time, based on the height weight function w_h(p_z) and the speed influence factor α, a crowd density estimation algorithm is constructed to accurately quantify the crowd density ρ_ci of the grid cell ci, avoiding the density misjudgment caused by the traditional method ignoring the height and moving speed of pedestrians. In addition, through the trajectory prediction algorithm, the predicted position vector P_B(Δt) within its future time window Δt is calculated in real time, and a predicted motion path TRJ_B(Δt) with high confidence is generated. These technical means not only solve the problem that the existing system relies on local perception and lacks global cooperation in dynamic target tracking, but also significantly reduce the path planning deviation caused by sensor data noise or pedestrian behavior randomness through accurate pose positioning and multi-dimensional environment modeling, so as to achieve more reliable obstacle avoidance decision-making and path robustness in the multi-robot cooperation scenario.

[0088] Embodiment 3

[0089] This embodiment is an explanatory note for Embodiment 2. Please refer to Figure 1 and Figure 3 , specifically: S3 includes S31 and S32;

[0090] S31. Based on the map grid coordinate system Gcs, combine the precise pose information POSE_A of robot A and the commodity positioning to determine the starting grid cell Sc and the target grid cell Gc, initialize the number of path plan strategies k = 3, and based on the number of path plan strategies, the set of strategy parameters Params_k = {β_k, ζ_k, Dsg_k, δ_k} of the k-th type is set in advance by the designer. The three preset path plan strategies are: a balanced plan, a risk aversion plan, and a speed priority plan;

[0091] Among them, β_k represents the density influence coefficient set under the k-th strategy, ζ_k represents the risk penalty weight set under the k-th strategy, Dsg_k represents the safety distance threshold set under the k-th strategy, and δ_k represents the distance influence exponent set under the k-th strategy;

[0092] Based on the crowd density map CDM and the predicted motion path TRJ_B(Δt) of robot B within the future time window Δt, construct a strategic spatio-temporal risk incremental cost algorithm, calculate the spatio-temporal risk incremental cost DC_k of each grid cell based on the k-th strategy, and record the spatio-temporal risk incremental costs DC_k of all grid cells in the grid spatio-temporal risk incremental cost set DCth_k of the k-th strategy;

[0093] Based on the grid spatio-temporal risk incremental cost set DCth_k of the k-th strategy, use the dijkstra algorithm to find the optimal candidate path Path_k of the k-th strategy, and generate a candidate path set Path by combining the optimal candidate paths of all strategies;

[0094] S32. The strategic spatio-temporal risk incremental cost algorithm consists of a basic time cost algorithm and a spatio-temporal risk factor algorithm;

[0095] Among them, the expression of the basic time cost algorithm is as follows:

[0096] ;

[0097] In the formula, T_k(ci) represents the basic time cost of moving in the grid cell ci under the k-th strategy, L_seg represents the geometric distance of moving in the grid cell ci, V_max represents the maximum speed of the robot, ρ_ci represents the crowd density of the grid cell ci, exp represents the natural exponential function, and m represents the total number of grid cells;

[0098] Among them, the expression of the spatio-temporal risk factor algorithm is as follows:

[0099] ;

[0100] In the formula, R_k(ci) represents the spatio-temporal risk factor of the robot B when moving in the grid cell ci under the k-th strategy, POS_ci represents the position where the simulated robot A walks to the grid cell ci, T_ci represents the predicted time when the simulated robot A reaches the grid cell ci, POS_B(T_ci) represents the position of the robot B in the predicted motion path TRJ_B(Δt) at the time T_ci, and dist(POS_ci, POS_B(T_ci)) represents the Euclidean distance between the simulated robot A and the robot B at the predicted time T_ci;

[0101] Among them, the algorithm expression of the spatio-temporal risk incremental cost is as follows;

[0102] ;

[0103] In the formula, DC_k(ci) represents the spatio-temporal risk incremental cost of moving in the grid cell ci under the k-th strategy.

[0104] Specific example of calculating the spatio-temporal risk incremental cost of grid cells:

[0105] For the balanced plan k = 1, the set of strategy parameters Params_1 is as follows:

[0106] The density influence coefficient β1 = 1.0;

[0107] The risk penalty weight ζ1 = 1.0, the safety distance threshold Dsg1 = 4.0 m, and the distance influence index δ1 = 2.0;

[0108] Set the calculated grid cell to c27, with coordinates (5, 6) in the map grid coordinate system Gcs;

[0109] The size of the grid cell in the supermarket environment is 1.0 * 1.0, and the pedestrian flow density ρ_c27 of grid cell c27 = 3.0;

[0110] The geometric distance of grid cell movement L_seg = 1.0, and the maximum speed of the robot V_max = 1;

[0111] Calculation of the basic time cost of grid cell c27:

[0112] ;

[0113] When exploring grid cell c27, the predicted time T_ci for robot A to reach grid cell ci = 1.5 s, and the position of robot B at time T_ci, POS_B(T_ci) = (6, 7);

[0114] Calculation of the spatio-temporal risk factor of grid cell c27:

[0115] ;

[0116] Algorithm for the empty risk incremental cost of grid cell c27:

[0117] .

[0118] In this embodiment, by constructing a strategic spatio-temporal risk incremental cost algorithm, the method can dynamically calculate the spatio-temporal risk incremental cost DC_k of each grid cell based on different strategies based on a preset path plan strategy and a set of strategy parameters Params_k, in combination with the crowd density map CDM and the predicted motion path TRJ_B(Δt) of robot B within a future time window Δt. This cost calculation integrates the basic time cost T_k(curr, next) and the spatio-temporal risk factor R_k(curr, next). The basic time cost T_k takes into account the maximum speed V_max of the robot and the crowd density ρ_next of the next grid cell, and the spatio-temporal risk factor R_k quantifies the spatio-temporal conflict risk between the robot and robot B at the predicted time T_next through the distance influence index δ_k and the safety distance threshold Dsg_k. This enables the system to generate a set of candidate paths Path containing multiple optimal candidate paths, providing diverse path choices that balance efficiency, safety, and real-time performance, solving the problem of path rigidity caused by single-objective optimization in traditional path planning, and significantly enhancing the ability and decision-making flexibility of the robot to perform adaptive path decision optimization in complex dynamic environments. This is different from the overall improvement of path planning efficiency and safety achieved by integrating multi-source perception described in claim 1.

[0119] Embodiment 4

[0120] This embodiment is an explanatory description carried out in Embodiment 3. Please refer to Figure 1 and Figure 3 , specifically: S4 includes S41 and S42;

[0121] S41. Based on the generated set of candidate paths Path, smooth the paths in the set of candidate paths Path and create a visualization interface. The information fed back by the visualization interface includes the path map, the crowd density ρ on the path, and the path prediction time T obtained by accumulating the basic path grid time costs of the path grid cells. The user selects a path and marks the selected path as Path_obt. By recording the path map of the path Path_obt and the path prediction passing time TPath_obt obtained by accumulating the basic time costs of the grid cells on the path Path_obt, a path planning report AS is generated.

[0122] S42. The robot A departs according to the selected path Path_obt, and uses the lidar, camera, and wheel odometer mounted on the robot A to observe the path in real time. When an impassable obstacle appears on the detected front path, based on the current grid cell where the robot is located as the new starting grid cell Sc_new, the strategy parameter set Params of the path strategy solution selected by the user is used to re-plan the path, record the actual path passing time T, the number of times of re-planning the path Ren, and the re-planning path position Rep, and generate a feedback data set Fh;

[0123] S5 includes S51;

[0124] S51. Based on the path planning report As and the feedback data set Fh, perform iterative optimization;

[0125] If the actual path passing time T minus the predicted path passing time TPath_obt ≥ 30s, then the professional personnel adjust the preset speed influence factor α and the density influence coefficient β set under the current strategy;

[0126] If the actual path passing time T minus the predicted path passing time TPath_obt < 30s, then no adjustment is required;

[0127] If the number of times of planning the path Ren ≥ 2 times, then the professional personnel adjust the risk penalty weight ζ, the safety distance threshold Dsg, and the distance influence index δ set under the current strategy;

[0128] If the number of times of planning the path Ren < 2 times, then no adjustment is required;

[0129] Add the path planning report As and the feedback data set Fh to the historical database His.

[0130] In this embodiment, by observing and recording real-time travel data based on the path Path_obt selected by the user, the system can generate a feedback data set Fh including the actual path travel time T, the number of times of replanning the path Ren, and the replanning path position Rep. Based on the iterative optimization of the path planning report As and the feedback data set Fh, the system can, according to the difference between the actual path travel time T and the accumulated path prediction time TPath_obt, as well as the number of times of planning the path Ren, enable professionals to dynamically adjust the set of policy parameters Params such as the preset speed influence factor α, the density influence coefficient β set under the current policy, the risk penalty weight ζ, the safety distance threshold Dsg set under the current policy, and the distance influence index δ set under the current policy, and add the relevant data to the historical database His. This continuous learning and parameter self-adaptation ability based on real-world execution feedback significantly enhances the path planning efficiency and safety of the robot in unknown or changing environments, realizes the self-improvement of system performance and rapid environmental adaptation, which is different from the benefits brought by Claim 1 that focuses on integrating the pedestrian flow density and the predicted trajectory of the robot in the planning stage to optimize the initial path.

[0131] Embodiment 5

[0132] A control method and system for a supermarket shopping robot adopting intelligent path planning, please refer to Figure 2 , specifically: including a data acquisition module, a data analysis module, a path generation module, an execution module, and an iterative optimization module;

[0133] The data acquisition module obtains environmental data Ei through sensors configured in the supermarket, and through sensors and communication modules carried by robot A, real-time collects image data around robot A and receives the status data ORS_B of robot B. The user inputs specified commodities to robot A through the operation interface and locates the commodity positions.

[0134] The data analysis module divides the supermarket environment into a grid coordinate system Gcs based on the supermarket map data Map, calculates the precise pose information POSE_A of robot A, generates a pedestrian flow density map CDM in combination with the environmental data Ei, and generates a predicted motion path TRJ_B(Δt) of robot B within the future time window Δt based on the status data ORS_B of robot B.

[0135] The path generation module calculates the spatio-temporal risk increment cost DC based on the precise pose information POSE_A of robot A, the pedestrian flow density map CDM, and the predicted motion path TRJ_B(Δt) of robot B within the future time window Δt, and generates a candidate path set Path based on the spatio-temporal risk increment cost DC.

[0136] The execution module generates a path planning report AS by allowing the user to select a path from the generated set of candidate paths Path and marking the selected path as Path_obt. Robot A then departs along path Path_obt, observes the path in real time, records the travel data, and generates a set of feedback data Fh.

[0137] The iterative optimization module performs iterative optimization based on the path planning report As and the set of feedback data Fh, and adds the path planning report As and the actual set of feedback data Fh to the historical database His.

[0138] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A control method for a supermarket shopping robot using intelligent path planning, characterized in that: Including the following steps: S1. Obtain environmental data Ei through sensors configured in the supermarket. Through sensors and communication modules carried by robot A, real-time collect image data around robot A and receive the status data ORS_B of robot B. The user inputs a specified commodity to robot A through the operation interface and locates the position of the commodity; S2. Divide the supermarket environment into a grid coordinate system Gcs based on the supermarket map data Map, calculate the precise pose information POSE_A of robot A, generate a crowd density map CDM in combination with the environmental data Ei, and generate a predicted motion path TRJ_B(Δt) of robot B within a future time window Δt based on the status data ORS_B of robot B; S3. Based on the precise pose information POSE_A of robot A, the crowd density map CDM, and the predicted motion path TRJ_B(Δt) of robot B within a future time window Δt, calculate the spatio-temporal risk incremental cost DC, and generate a candidate path set Path based on the spatio-temporal risk incremental cost DC; S4. Based on the generated candidate path set Path, the user selects a path and marks the selected path as Path_obt, generates a path planning report AS, robot A departs along the path Path_obt and observes the path in real time, records the travel data, and generates a feedback data set Fh; S5. Based on the path planning report As and the feedback data set Fh, perform iterative optimization and add the path planning report As and the actual feedback data set Fh to the historical database His.

2. The control method of a supermarket shopping robot adopting intelligent path planning according to claim 1, wherein: S1 includes S11; S11. Robot A scans the surrounding environment through the carried lidar to obtain the original point cloud data set PCD_A of the current cycle of robot A, captures images within the field of view through the carried camera to obtain the original image frame data IMG_A of the current cycle of robot A, records the wheel rolling information through the carried wheel odometer to obtain the original odometer reading ODO_A of the current cycle of robot A, listens to and decodes the status data ORS_B broadcast by robot B from the network through the carried communication module, obtains the environmental point cloud data Ei through the lidar configured in the supermarket, the user inputs the waitable time T_user and the specified commodity to robot A through the operation interface, and robot A locates the position of the commodity through the supermarket map data Map; Among them, the status data ORS_B of robot B includes the original pose information POSE_B of robot B, the speed information VEL_B of robot B, the acceleration vector ACC_B of robot B, and the short-term intention path ITP_B of robot B.

3. The control method of a supermarket shopping robot using intelligent path planning according to claim 1, wherein: S2 includes S21 and S22; S21. Based on the supermarket map data Map, generate the supermarket environment into a map grid coordinate system Gcs. Perform point cloud downsampling on the original point cloud data set PCD_A of robot A in the current cycle to obtain the point cloud data PCD_Apro of robot A. Perform Matlab distortion correction on the original image frame data IMG_A of robot A in the current cycle to obtain the image frame data IMG_Apro of robot A. Combine the point cloud data PCD_Apro of robot A, the image frame data IMG_Apro of robot A in the current cycle, the original odometer reading ODO_A of robot A in the current cycle, and the grid coordinate system Gcs to run the SLAM positioning algorithm to obtain the accurate pose information POSE_A of robot A in the current map grid coordinate system Gcs; The map grid coordinate system Gcs is composed of a set of grid cells C = {c1, c2,..., cm}, where m represents the total number of grid cells. Combine the environmental point cloud data Ei to construct a pedestrian flow density estimation algorithm to calculate the pedestrian flow density ρ_ci of the grid cell ci; According to the state data ORS_B of robot B, construct a trajectory prediction algorithm to predict the predicted position vector P_B(Δt) of robot B within the future time window Δt.

4. A control method for a supermarket shopping robot using intelligent path planning according to claim 3, characterized in that: S22. Among them, the expression of the pedestrian flow density estimation algorithm is as follows: ； ； In the formula, i represents the grid cell number, p represents the point cloud point in the grid cell ci in the environmental point cloud data Ei, v_p represents the average moving speed of the point p in the grid cell ci, A_ci represents the area of the grid cell ci, α represents the preset speed influence factor, w_h(p_z) represents the height weight function. If the value of the height weight function w_h(p_z) is 0, it means that the point p is not included in the calculation of the pedestrian flow density. If the value of the height weight function w_h(p_z) is 1, it means that the point p is included in the calculation of the pedestrian flow density. p_z represents the vertical coordinate of the point p, Hgro_max represents the maximum ground vertical height threshold, and Hhum_max represents the maximum high-altitude vertical height threshold; Among them, the expression of the trajectory prediction algorithm is as follows: ； Combine the map grid coordinate system Gcs and the pedestrian flow density ρ_ci of the grid cell ci to generate a pedestrian flow density map CDM. Combine the short-term intention path ITP_B of robot B and the predicted position vector POB_B(Δt) of robot B within the future time window Δt to generate the predicted motion path TRJ_B(Δt) of robot B within the future time window Δt.

5. The control method of a supermarket shopping robot using intelligent path planning according to claim 4, characterized in that: S3 includes S31 and S32; S31. Based on the map grid coordinate system Gcs, combine the accurate pose information POSE_A of robot A and the commodity positioning to determine the starting grid cell Sc and the target grid cell Gc. Initialize the number of path plan strategies k = 3, and based on the number of path plan strategies, the set of strategy parameters Params_k = {β_k, ζ_k, Dsg_k, δ_k} of the k-th type are set in advance by the designer. The three preset path plan strategies are: the balanced plan, the risk aversion plan, and the speed priority plan; Among them, β_k represents the density influence coefficient set under the k-th strategy, ζ_k represents the risk penalty weight set under the k-th strategy, Dsg_k represents the safety distance threshold set under the k-th strategy, and δ_k represents the distance influence index set under the k-th strategy; Based on the crowd density map CDM and the predicted motion path TRJ_B(Δt) of robot B within the future time window Δt, a strategized spatio-temporal risk incremental cost algorithm is constructed to calculate the spatio-temporal risk incremental cost DC_k of each grid cell based on the k-th strategy, and record the spatio-temporal risk incremental costs DC_k of all grid cells in the grid spatio-temporal risk incremental cost set DCth_k of the k-th strategy; Based on the grid spatio-temporal risk incremental cost set DCth_k of the k-th strategy, use the dijkstra algorithm to find the optimal candidate path Path_k of the k-th strategy, and generate a candidate path set Path by combining the optimal candidate paths of all strategies.

6. A control method for a supermarket shopping robot using intelligent path planning according to claim 5, characterized in that: S32. The strategized spatio-temporal risk incremental cost algorithm consists of a basic time cost algorithm and a spatio-temporal risk factor algorithm; Among them, the expression of the basic time cost algorithm is as follows: ； In the formula, T_k(ci) represents the basic time cost of moving in grid cell ci under the k-th strategy, L_seg represents the geometric distance of moving in grid cell ci, V_max represents the maximum speed of the robot, ρ_ci represents the crowd density of grid cell ci, exp represents the natural exponential function, and m represents the total number of grid cells; Among them, the expression of the spatio-temporal risk factor algorithm is as follows: ； In the formula, R_k(ci) represents the spatio-temporal risk factor of robot B when moving in grid cell ci under the k-th strategy, POS_ci represents the position where the simulated robot A walks to grid cell ci, T_ci represents the predicted time when the simulated robot A reaches grid cell ci, POS_B(T_ci) represents the position of robot B in the predicted motion path TRJ_B(Δt) at time T_ci, and dist(POS_ci, POS_B(T_ci)) represents the Euclidean distance between the simulated robot A and robot B at the predicted time T_ci; Among them, the expression of the spatio-temporal risk incremental cost algorithm is as follows; ； In the formula, DC_k(ci) represents the spatio-temporal risk incremental cost of moving in grid cell ci under the k-th strategy.

7. A control method for a supermarket shopping robot using intelligent path planning according to claim 6, characterized in that: S4 includes S41 and S42; S41. Based on the generated candidate path set Path, smooth the paths in the candidate path set Path and create a visualization interface. The information fed back by the visualization interface includes the path map, the crowd density ρ on the path, and the path prediction time T obtained by accumulating the basic path grid time costs of the grid cells on the path. The user selects a path and marks the selected path as Path_obt, and generates a path planning report AS by recording the path map of Path_obt and the path prediction passing time TPath_obt obtained by accumulating the basic time costs of the grid cells on the path Path_obt.

8. A control method for a supermarket shopping robot using intelligent path planning according to claim 7, characterized in that: S42. The robot A departs according to the selected path Path_obt. It observes the path in real time through the lidar, camera, and wheel odometer carried by the robot A. When an impassable obstacle appears on the detected front path, based on the grid cell where the current position is located as the new starting grid cell Sc_new, it uses the set of policy parameters Params of the path strategy solution selected by the user to re-plan the path, records the actual path passing time T, the number of times of re-planning the path Ren, and the re-planning path position Rep, and generates a feedback data set Fh.

9. The control method of a supermarket shopping robot adopting intelligent path planning according to claim 8, characterized in that: S5 includes S51; S51. Based on the path planning report As and the feedback data set Fh, perform iterative optimization; If the actual path passing time T minus the predicted path passing time TPath_obt ≥ 30s, then professionals adjust the preset speed influence factor α and the density influence coefficient β set under the current policy; If the actual path passing time T minus the predicted path passing time TPath_obt < 30s, then no adjustment is required; If the number of times of planning the path Ren ≥ 2 times, then professionals adjust the risk penalty weight ζ, the safety distance threshold Dsg, and the distance influence index δ set under the current policy; If the number of times of planning the path Ren < 2 times, then no adjustment is required; Add the path planning report As and the feedback data set Fh to the historical database His.

10. A supermarket shopping robot control system using intelligent path planning, which is applied to a supermarket shopping robot control method using intelligent path planning according to any one of claims 1 to 9, and is characterized in that: It includes a data acquisition module, a data analysis module, a path generation module, an execution module, and an iterative optimization module; The data acquisition module obtains environmental data Ei through the sensors configured in the supermarket, and through the sensors and communication module carried by the robot A, it collects the image data around the robot A in real time and receives the status data ORS_B of the robot B. The user inputs the specified goods to the robot A through the operation interface and locates the goods position; The data analysis module divides the supermarket environment into a grid coordinate system Gcs based on the supermarket map data Map, calculates the accurate pose information POSE_A of the robot A, generates a crowd density map CDM in combination with the environmental data Ei, and generates a predicted motion path TRJ_B(△t) of the robot B within the future time window △t based on the status data ORS_B of the robot B; The path generation module calculates the spatio-temporal risk incremental cost DC based on the accurate pose information POSE_A of the robot A, the crowd density map CDM, and the predicted motion path TRJ_B(△t) of the robot B within the future time window △t, and generates a candidate path set Path based on the spatio-temporal risk incremental cost DC; The execution module selects a path from the generated candidate path set Path, marks the selected path as Path_obt, generates a path planning report AS, the robot A departs along the path Path_obt and observes the path in real time, records the travel data, and generates a feedback data set Fh; The iterative optimization module performs iterative optimization based on the path planning report As and the feedback data set Fh, and adds the path planning report As and the actual feedback data set Fh to the historical database His.

Images