Robot obstacle avoidance control method and system

Abstract

The application provides a robot obstacle avoidance control method and system, and relates to the technical field of robot control, and the method comprises the following steps: when the robot needs to avoid obstacles with a crowd of people, the robot is controlled to avoid obstacles to minimize the influence on the crowd of people continuing to walk together, including: step a, whenever the limit is updated, steps b to d are repeatedly executed until the crowd of people starts to move away from the robot; step b, predicting the crowd distribution of the crowd of people at multiple future unit time points within the predicted limit after the update; step c, for each future unit time point, the influence area of the crowd distribution at the future unit time point is demarcated based on the influence area demarcation rule, and is associated with the future unit time point; step d, the robot is controlled to avoid obstacles, so that for each future unit time point, the robot does not enter the influence area associated with the future unit time point. The moving trend of the crowd of people is predicted in advance, and the robot path planning is optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot control, in particular to a robot obstacle avoidance control method and system. BACKGROUND

[0002] Currently, with the development of robot technology, robots are increasingly used in public places such as hotels, shopping malls, and airports. These robots are usually used to perform tasks such as guiding guests, transporting goods, or providing information. However, in these complex human-robot coexistence environments, robot navigation faces significant challenges, especially when there are dynamic obstacles, especially groups of people moving together (co-moving crowds). Traditional robot obstacle avoidance methods usually treat all obstacles as homogeneous objects, lacking targeted processing of the dynamic characteristics of co-moving crowds. This approach often leads to inefficient or disruptive robot behavior when facing group movements, such as sudden stops or improper choice of detour paths, thereby disrupting the natural movement trajectory of the crowd and causing inconvenience or even chaos.

[0003] In scenarios such as hotel lobbies or shopping malls with high human traffic, maintaining the continuity and smoothness of crowd movement is crucial for operational efficiency and user experience. Any interruption caused by robots can lead to delays, reduced comfort, and even potential safety hazards. At the same time, predicting and managing the movement of co-moving crowds is a technical challenge. Compared to the behavior of individual pedestrians, group movement has higher complexity, with direction, speed, and formation changing dynamically at any time. Existing methods often struggle to accurately predict the future distribution of the crowd and effectively delineate critical areas that the robot should avoid, making it difficult to balance safety and minimization of impact on the crowd during obstacle avoidance.

[0004] To address the above problems, an innovative robot obstacle avoidance control method is needed that can prospectively predict the movement trend of co-moving crowds and optimize robot path planning based on it. SUMMARY

[0005] One of the objectives of the present application is to provide a robot obstacle avoidance control method and system to solve the problems identified in the background.

[0006] In a first aspect, the robot obstacle avoidance control method provided by the embodiments of the present application comprises:

[0007] When the robot needs to avoid the co-moving crowd, the robot is controlled to avoid obstacles to minimize the impact on the co-moving crowd continuing to move together.

[0008] Optionally, the robot is controlled to avoid obstacles to minimize the impact on the co-moving crowd continuing to move together, comprising:

[0009] Step a, whenever the prediction limit is updated, repeat steps b to d until the crowd starts to move away from the robot;

[0010] Step b, predict the crowd distribution of the same crowd at multiple future unit time points within the prediction limit after the update;

[0011] Step c, for each future unit time, based on the influence area delineation rule, delineate the influence area of the crowd distribution at the future unit time and associate it with the future unit time;

[0012] Step d, the robot is controlled to avoid obstacles, so that for each future unit time, the robot does not enter the influence area associated with the future unit time.

[0013] Optionally, the prediction limit includes: the highest time limit to ensure the credibility of the predicted crowd distribution.

[0014] Optionally, the step b, predicting the crowd distribution of the same crowd at multiple future unit time points within the prediction limit after the update, comprises:

[0015] Based on the pre-trained crowd movement prediction model, according to the historical movement information of the same crowd, the crowd distribution of the same crowd at multiple future unit time points within the prediction limit after the update is predicted.

[0016] Optionally, the influence area delineation rule comprises:

[0017] For each non-terminal future unit time, on the environmental perception map of the robot, the ground positions of the same person in the crowd distribution at the future unit time and the crowd distribution at the next future unit time are connected to obtain multiple position connections, the target person is determined from the crowd distribution at the future unit time, the minimum rectangle enclosing all position connections and the area of the circle with all target person position connections as the diameter are determined, and the influence area of the crowd distribution at the future unit time is obtained after de-duplication; Wherein, the minimum distance between the target person and other persons in the crowd distribution at the future unit time is less than the preset distance threshold or the position connection of the target person and the position connection of other persons produces intersection;

[0018] For the last future unit time, on the environmental perception map, the minimum enclosing circle enclosing the ground positions of all persons in the crowd distribution at the future unit time is determined, the minimum enclosing circle is enlarged according to the preset radius multiple to obtain the influence area of the crowd distribution at the future unit time.

[0019] Optionally, in step d, when the robot needs to enter the influence area associated with each remaining future unit time due to limited avoidance space, the following steps are further included:

[0020] On the environmental perception map, a plurality of movement paths of the robot are planned, so that the robot enters the respective associated influence area in turn at each remaining future unit time;

[0021] Based on the movement path with the minimum quantified value of the influence on the same group of people continuing to walk together, the robot is controlled to move.

[0022] Optionally, the obtaining step of the quantified value of the influence on the same group of people continuing to walk together comprises:

[0023] For each remaining future unit time, a local path executed by the robot at the future unit time is intercepted from the movement path, a movement direction of the local path is extracted, a connection direction of a plurality of position connections of the influence area associated with the future unit time at the time of demarcation is traced back, the number of connection directions falling within a preset standard direction relationship range between the movement direction and the connection direction is counted, and is associated with the future unit time; wherein the preset standard direction relationship range contains a plurality of standard direction relationships representing that the personnel will produce a reaction of preparing to avoid the robot.

[0024] Based on the respective preset weights corresponding to the different remaining future unit times, the respective associated numbers are weighted calculated to obtain the quantified value; wherein the preset weights and the early or late degree of the remaining future unit times are in a negative correlation relationship.

[0025] Optionally, the robot obstacle avoidance control method further comprises:

[0026] The robot is controlled to record an obstacle avoidance history of avoiding obstacles for the same group of people;

[0027] Based on the obstacle avoidance history, a same group of people walking time rule in the working scene of the robot is mined;

[0028] Based on the same group of people walking time rule, a daily working route of the robot is optimized.

[0029] In a second aspect, the robot obstacle avoidance control system provided by the embodiments of the present application comprises:

[0030] The obstacle avoidance control module is configured to control the robot to avoid obstacles when the robot needs to avoid obstacles for the same group of people, so as to minimize the influence on the same group of people continuing to walk together.

[0031] Optionally, the obstacle avoidance control module controls the robot to avoid obstacles so as to minimize the influence on the same group of people continuing to walk together, comprising:

[0032] Step a, whenever the prediction limit is updated, steps b to d are repeatedly executed until the same group of people starts to move away from the robot;

[0033] Step b, predicting the crowd distribution of the same crowd at multiple future unit time moments within the prediction limit after the update;

[0034] Step c, for each future unit time moment, based on the influence area demarcation rule, demarcating the influence area of the crowd distribution at the future unit time moment, and associating with the future unit time moment;

[0035] Step d, obstacle avoidance control is performed on the robot, so that for each future unit time moment, the robot does not enter the influence area associated with the future unit time moment.

[0036] The present application has the following beneficial effects:

[0037] The present application realizes efficient and safe obstacle avoidance of the same crowd in a complex environment by dynamically predicting crowd movement, demarcating the influence area and implementing precise control. Specifically, through forward-looking prediction and area division, the robot avoids the key area of crowd movement and interaction, maintains a safe distance from the crowd, reduces the risk of collision, dynamically adjusts the strategy to adapt to different scenes, and ensures the efficiency of task completion. It is suitable for various application scenarios such as hotel greeting and meal delivery robots, and provides an intelligent solution for human-machine comfortable coexistence.

[0038] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or will be learned by practice of the present application. The objects and other advantages of the present application can be achieved and obtained by the structure specifically pointed out in the written description and the accompanying drawings.

[0039] The technical solutions of the present application will be further described in detail below with the help of the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0040] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation of the present application. In the drawings:

[0041] Figure 1 A flowchart of the robot obstacle avoidance control method in the embodiment of the present application;

[0042] Figure 2 Another flowchart of the robot obstacle avoidance control method in the embodiment of the present application;

[0043] Figure 3 Another flowchart of the robot obstacle avoidance control method in the embodiment of the present application;

[0044] Figure 4 A schematic diagram of the robot obstacle avoidance control system in the embodiment of the present application. DETAILED DESCRIPTION

[0045] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0046] The research and development approach of this application is to predict the future distribution of a group of people traveling together and delineate the influence zone that the robot should avoid entering, thereby achieving obstacle avoidance control of the robot and minimizing the impact on the group's continued travel. The influence zone refers to the area where entering it would interfere with the group's continued travel. This method is based on a pre-trained crowd movement prediction model, which uses historical movement data to predict the future position of the crowd. Through dynamically updated prediction limits and influence zone delineation rules, it ensures that the robot can safely navigate around obstacles without interrupting the movement or interaction of the crowd. This method is particularly suitable for complex environments where robots and humans coexist, such as hotels and shopping malls, aiming to improve the robot's task efficiency while ensuring the comfort and safety of the crowd.

[0047] Figure 1 A flowchart of a robot obstacle avoidance control method is provided for embodiments of this application, such as... Figure 1 As shown, the method includes:

[0048] 101. When the robot needs to avoid obstacles from a group of people traveling together, obstacle avoidance control should be implemented on the robot to minimize the impact on the group's continued travel.

[0049] In this step, the robots can be hotel-grade welcoming robots, food delivery robots, and takeout robots, among others. These robots automatically detect objects that need to be avoided (these detection tasks are within the scope of existing technology and will not be elaborated upon). When the robot detects a group of people moving together, it triggers an obstacle avoidance control mechanism. Through path planning and dynamic adjustments, it ensures that the robot's movement does not interfere with the group's collective movement trajectory or individual interactions. A group of people moving together refers to a group of people moving in a group, such as a group of tourists or colleagues. The robot uses sensors (such as LiDAR or cameras) to detect obstacles in the environment in real time. When it identifies a group of people moving together, it prioritizes minimizing the impact on their movement continuity.

[0050] In one implementation example, in a hotel lobby, a welcoming robot is guiding guests to the front desk when it detects a group of tourists moving towards the exit. The robot uses sensors to determine the crowd's location, initiates obstacle avoidance control, and adjusts its path to bypass the crowd's expected movement route, rather than directly crossing through the crowd or suddenly stopping, thus avoiding disrupting the tourists' collective movement or causing confusion.

[0051] like Figure 2 As shown, in step 101, obstacle avoidance control is performed on the robot to minimize the impact on the group of people continuing to walk together. This specifically includes the following sub-steps:

[0052] Step a, repeat steps b to d whenever the prediction horizon is updated until the crowd starts moving away from the robot. The prediction horizon includes the maximum time limit that ensures the predicted crowd distribution is reliable.

[0053] In this step, the prediction horizon refers to the longest future time range that ensures the crowd distribution prediction result is reliable. To ensure the accuracy of obstacle avoidance control, the system continuously monitors the changes in the prediction horizon and re-executes the subsequent steps at each update until the crowd movement direction indicates that its distance from the robot starts to increase (i.e., moving away from the robot). This cyclic mechanism ensures that the robot always bases its obstacle avoidance on the latest prediction data, adapting to the dynamic changes in the crowd.

[0054] In one implementation example, in a hotel lobby scenario, a welcoming robot detects a group of tourists moving towards the exit. The initial prediction horizon is 5 seconds, updated every 1 second. When the prediction horizon is adjusted to 4 seconds due to changes in the crowd movement pattern (e.g., acceleration), the robot re-executes steps b to d to update the obstacle avoidance strategy. If the tourists then turn in the direction away from the robot (e.g., into the corridor), the cycle terminates and the robot resumes the normal task path.

[0055] Step b, predict the crowd distribution of the crowd at multiple future unit times within the prediction horizon after the update. Step b includes predicting the crowd distribution of the crowd at multiple future unit times within the prediction horizon after the update based on the historical movement information of the crowd according to the pre-trained crowd movement prediction model.

[0056] In this step, a pre-trained crowd movement prediction model is used to predict the position distribution of the crowd at multiple future time points (e.g., every second) within a prediction limit based on the historical movement information (e.g., position, speed, trajectory) of the crowd. The prediction limit is determined by the model confidence, and the process includes: the model continuously attempts to predict further future time, until the confidence of the prediction result is below the preset threshold, and the maximum time range that is currently reliable is the prediction limit. A large amount of historical data is required to train the model, covering typical patterns of crowd movement (e.g., straight walking, turning, dispersing, etc.). The specific process of training the crowd movement prediction model includes: using a long short-term memory network (LSTM) as the core architecture, and using the PyTorch deep learning framework for training. The training data comes from the historical data of crowd movement in multiple scenes (e.g., hotels, shopping malls, stations, parks, etc.), including continuous time series data collected by laser radar, camera or GPS, each data including timestamp, position coordinates, speed, direction and crowd density, covering typical movement patterns such as straight walking, turning, dispersing, etc., and the data volume needs to be sufficient to ensure the model generalization ability. The training process uses supervised learning, the input is the past 10 seconds of trajectory sequence, and the output is the predicted position distribution at future seconds (one time point per second). The model generates a confidence score for each prediction time point through an additional confidence output layer (based on Dropout or Bayesian neural network), and terminates the prediction when the confidence is below the preset threshold (e.g., 0.7), determining the prediction limit. During training, mean squared error (MSE) is used as the loss function for position prediction, and the confidence loss is optimized, with an Adam optimizer learning rate of 0.001, a batch size of 64, and a training of 100 epochs. After training, the model can output the position distribution and confidence of each second in the future based on the input real-time trajectory data, which is used for the robot obstacle avoidance application scenario of this application.

[0057] In one implementation example, in a shopping mall, a delivery robot detects a group of customers walking from a restaurant to an elevator. The robot obtains the movement trajectory of the crowd in the past 10 seconds (coordinate points recorded by laser radar), and inputs the pre-trained model. The model predicts the position distribution of the crowd every second in the next 4 seconds, and finds that the confidence drops below the threshold at the 5th second, so the prediction limit is 4 seconds. The prediction result shows that the crowd will move along a straight line to the elevator entrance, and the robot plans an obstacle avoidance path accordingly.

[0058] Step c, for each future unit time, based on the influence area delimiting rule, delimit the influence area of the crowd distribution at the future unit time, and associate it with the future unit time.

[0059] In this step, the influence area refers to the area that will interfere with the crowd to continue walking if the robot enters. For each future time point, the system applies specific rules to delineate these areas based on the crowd distribution prediction results and associates them with the corresponding time to ensure that the robot avoids them.

[0060] The delineation rules are processed in two cases for non-final and final time points to take into account the crowd movement path and uncertainty.

[0061] Specifically, the influence area delineation rules include:

[0062] Rule 1: For each non-final future unit time, connect the ground positions of the same personnel in the crowd distribution at the future unit time and the crowd distribution at the next future unit time on the robot's environment perception map to obtain multiple position lines. Determine the target personnel from the crowd distribution at the future unit time, determine the smallest rectangle that encloses all position lines and the area of the circle with all target personnel position lines as the diameter, and perform deduplication to obtain the influence area of the crowd distribution at the future unit time; Where the minimum distance between the target personnel and other personnel in the crowd distribution at the future unit time is less than the preset distance threshold or the position line intersects with the position line of other personnel.

[0063] In rule 1, when the robot detects the environment, it models the surrounding environment in real time to form an environment perception map. The line connecting the ground positions of the same person in the crowd distribution of two adjacent future time units represents the further movement of the person. Considering that the people in the crowd continue to walk together, the people near the boundary will only interact with the people inside them. Therefore, the robot needs to avoid the area of the smallest rectangle that encloses all the position lines. The preset distance threshold represents the distance between two people that is easy to interact. When the minimum distance is less than the threshold, it means that the corresponding person is likely to interact with other people and is a target person. In addition, when the position line of a person intersects with the position line of another person, it means that the person may change his moving direction greatly to interact with a specific person and is likely to interact with other people, so he is also a target person. However, when the target person actually interacts, he is less likely to interact with other people at the head and tail of his moving position. More often, he focuses on the range near him during the moving process. To address this situation, the robot needs to focus on avoiding the area of the circle with the position lines of all target people as the diameter. Therefore, the area of the smallest rectangle that encloses all the position lines and the area of the circle with the position lines of all target people as the diameter are determined, and the duplicate areas are removed to obtain the influence area of the crowd distribution at the future time unit. The implementation of rectangular coverage of the overall moving range and the circular area focusing on the interaction point ensures the avoidance of key areas. In addition, special attention is paid to target people to prevent the robot from interfering with the social behavior of the crowd. Secondly, the use of the de-duplication operation simplifies the definition of the area and can avoid a certain degree of redundant calculation.

[0064] In a practical case: In a hotel lobby, the robot predicts the crowd distribution at the 2nd and 3rd seconds. Connecting the positions of the same person between the two seconds, the moving path is obtained. Assuming that the preset distance threshold is 0.5 meters, it is found that the distance between two people is 0.3 meters, and they are marked as target people; another person's path intersects with him, and he is also marked as a target person. The system calculates the minimum rectangle of all paths (5 meters long and 2 meters wide), and generates a circular area for the target person with a diameter equal to the path length (about 1 meter in diameter). After merging and de-duplicating, the influence area at the 2nd second is obtained. Rule 2: For the last future time unit, on the environment perception map, determine the smallest enclosing circle that encloses the ground positions of all people in the crowd distribution at the future time unit. Enlarge the smallest enclosing circle by a preset radius multiple to obtain the influence area of the crowd distribution at the future time unit.

[0065] In rule 2, the future unit time is at the end, and the crowd movement at the next future unit time is unknown, therefore, for safety considerations, the minimum enclosing circle of the ground position of all people in the crowd distribution at the future unit time is enlarged according to a preset radius multiple, and the influence area of the crowd distribution at the future unit time is obtained, to cope with the uncertainty of future movement. The preset radius multiple can be set as needed, for example, set to 1.5 times, and the radius of the minimum enclosing circle is increased by the multiple to form a new enclosing circle. The uncertainty after the prediction limit is compensated by enlarging the area, and the risk of omission is avoided. In addition, the single circle has low calculation complexity and is suitable for fast response in the scenario of the end unit time. In addition, it can adapt to possible sudden changes of the crowd, such as sudden stop or turning.

[0066] In an actual case: in the above shopping mall scenario, the 4th second is the prediction limit end. The system calculates the minimum enclosing circle (diameter 3 meters) of all customer positions, and enlarges it by 1.5 times to a diameter of 4.5 meters to form an influence area, ensuring that even if the crowd spreads slightly, the robot will not enter the potential interference range.

[0067] Step d, obstacle avoidance control is performed on the robot, so that for each future unit time, the robot does not enter the influence area associated with the future unit time.

[0068] In this step d, finally, the robot is controlled to avoid obstacles, so that it does not enter the corresponding associated influence area at each future unit time, thereby achieving the purpose of minimizing the influence on the continued travel of the same crowd. According to the influence area drawn in step c, the robot calculates an obstacle avoidance path through a path planning algorithm (the robot path planning algorithm belongs to the existing technical category and is not described in detail) to ensure that at each future time point, its position is located outside the corresponding influence area, thereby realizing minimal interference with crowd movement. The robot action is synchronized with the crowd movement, avoiding abrupt stop or acceleration. In addition, under the premise of ensuring obstacle avoidance, path optimization reduces unnecessary detours. Secondly, the crowd does not need to adjust its own behavior, and the robot is "invisible" and integrated into the environment.

[0069] In an actual case: in the hotel lobby, the robot plans a detour path based on the foregoing prediction and influence area: avoid the rectangular corridor at the 1st second, avoid the circular interaction area of the target person at the 2nd-3rd second, and avoid the last circular area after the lobby at the 4th second. The robot moves along the planned path at a smooth speed, successfully detours the crowd without entering any influence area.

[0070] The robot obstacle avoidance control method provided in the application realizes efficient and safe obstacle avoidance of the same crowd in a complex environment by dynamically predicting the movement of the crowd, demarcating the influence area and implementing precise control. Specifically, through forward-looking prediction and area division, the robot avoids the key area of crowd movement and interaction, maintains a safe distance from the crowd, reduces the risk of collision, dynamically adjusts the strategy to adapt to different scenarios, and ensures the efficiency of task completion. It is suitable for various application scenarios such as hotel greeting and meal delivery robots, and provides an intelligent solution for human-machine comfortable coexistence.

[0071] In some embodiments, when the robot needs to enter the influence area associated with each remaining future unit time due to limited avoidance space, the step d further comprises the following steps:

[0072] 201. On the environment perception map, a plurality of movement paths of the robot are planned, so that the robot enters the respective associated influence area at each remaining future unit time in turn.

[0073] In this step, the robot finds through path planning that no matter how the movement path is adjusted, it cannot completely avoid the corresponding influence area at some future unit times (remaining future unit times). In this case, the robot needs to plan a movement path so that it enters the respective associated influence area at each remaining future unit time in turn, while ensuring that the influence on the same crowd continues to be minimized. The degree of influence is measured by a quantitative value, and the goal of path planning is to minimize this quantitative value.

[0074] 202. Based on the movement path with the minimum quantitative value of the influence on the same crowd continuing to walk, the robot is controlled to move.

[0075] In this step, in some scenarios of robot obstacle avoidance control, due to limited avoidance space in the environment, the robot cannot completely avoid all influence areas at future unit times. At this time, in order to ensure that the robot can continue to perform tasks while minimizing the influence on the same crowd continuing to walk, the system will trigger a backup solution. This solution plans the movement path of the robot so that when it has to enter the influence area, it selects the path with the minimum influence on the crowd, and evaluates and optimizes this influence based on the quantitative value.

[0076] The acquisition step of the quantitative value of the influence on the same crowd continuing to walk in step 201 includes:

[0077] 301、For each of the remaining future unit time, the local path of the robot performing at the future unit time is intercepted from the moving path, the moving direction of the local path is extracted, the connection direction of the multiple position connection lines associated with the future unit time is traced, the number of connection directions falling into the preset standard direction relationship range is counted, and the future unit time is associated. Wherein, the preset standard direction relationship range contains multiple standard direction relationships representing the reactions of the personnel ready to avoid the robot.

[0078] In this step, the moving route is divided into the local path of the robot at each of the remaining future unit time. The moving direction of the local path is extracted, which is the direction of the robot moving. The influence area associated with the future unit time is connected to form multiple position connection lines when it is delimited, and the connection direction of these position connection lines is traced. The connection direction is the direction of the ground position of the personnel at the previous future unit time to the ground position of the personnel at the next future unit time, representing the moving direction of the corresponding personnel.

[0079] In reality, when the personnel sees the robot, if he perceives that there may be a collision risk or affect his continued progress, etc., he will produce a reaction ready to avoid the robot, which makes the robot affect the personnel. In addition, when the personnel sees the robot, when the personnel perceives that there may be a collision risk or affect his continued progress, etc., at the corresponding unit time, the respective moving directions of the robot and the personnel will produce a certain relationship, which can be determined by experiment in advance as a standard direction relationship, for example: face-to-face relationship (two direction angles close to 180°, such as 170°-180°), about to cross relationship (two direction angles close to 90°, such as 80°-100°) and oblique relationship (two direction angles 30°-80° or 100°-170°). Then, the standard direction relationship range is obtained by integrating these standard direction relationships.

[0080] The number of connection directions falling into the preset standard direction relationship range is counted, the larger the number, the greater the impact on the same crowd continuing to walk in the future unit time. The moving direction and the connection direction can be represented by vectors.

[0081] 302、Based on the different preset weights corresponding to the different remaining future unit times, the numbers associated with each other are weighted and calculated to obtain a quantitative value. Wherein, the preset weight and the early or late degree of the remaining future unit time are negatively related.

[0082] In this step, the earlier the future unit time, the earlier the representative person of the crowd is affected by the robot, which may cause a chain reaction to the overall movement of the crowd, and the greater the influence on the quantitative value, the greater the preset weight; the later the future unit time, the shorter the duration of the influence, the smaller the preset weight; therefore, the preset weight and the early or late degree of the remaining future unit time are negatively correlated. Based on the respective preset weights corresponding to the different remaining future unit times, the respective associated numbers are weighted and calculated to obtain the quantitative value.

[0083] On the environment perception map, based on the above quantitative value calculation method, a moving path is planned to minimize the quantitative value. When the robot moves along the path, it enters the corresponding influence area at each remaining future unit time, but by optimizing the direction relationship and timing selection, the influence on the crowd is minimized. According to the planned moving path, the robot is controlled to move at a smooth speed and direction, ensuring that it enters the influence area as planned at each remaining future unit time while minimizing the influence.

[0084] In a simple practical case, a hotel delivery robot encounters a group of 10 tourists moving east-northeast in a 3-meter-wide corridor and cannot completely avoid in 2 seconds. The standby path planning is triggered.

[0085] The moving path A and the moving path B of the robot are planned.

[0086] Analysis of the moving path A shows that there are 3 direction relationships between the moving direction of the local path in the future 1st second and the direction of the line connecting multiple positions of the line drawn when the influence area is drawn, which fall within the preset standard direction relationship range, and there are 0 in the future 2nd second. The preset weight of the future 1st second is 0.8, and the preset weight of the future 2nd second is 0.2. Quantitative value = 0.8 x 3 + 0.2 x 0 = 2.4.

[0087] The moving path B is analyzed in the same way to determine the quantitative value as 1.2.

[0088] Therefore, based on the moving path B, the robot is controlled to move.

[0089] The above method quantifies the impact and optimizes the path, so that even if the robot enters the impact area, it can choose a path that is significantly different from the direction of the crowd movement, avoiding triggering the avoidance reaction and maintaining the natural movement state of the crowd. In scenarios where avoidance space is limited, the robot can still dynamically plan a path and continue to perform tasks, avoiding efficiency decline or task interruption due to stagnation. Detailed direction relationship analysis ensures that the robot will not force the crowd to change behavior, reducing the risk of collision and improving the comfort of human-robot coexistence. The weighting mechanism prioritizes early impact and combines real-time prediction to dynamically adjust the path, ensuring that the optimal strategy is maintained when the crowd movement changes. This method is suitable for high-density crowd scenarios such as hotels, shopping malls, and airports, especially during peak hours or in narrow passages, effectively managing the interaction between robots and crowds.

[0090] As shown in Figure 3 some embodiments, the robot obstacle avoidance control method further comprises:

[0091] 102. The robot control method records the obstacle avoidance history of the same crowd.

[0092] In this step, when the robot performs obstacle avoidance operations, the system automatically records key information related to obstacle avoidance through sensors and internal computing modules, forming obstacle avoidance history data. These data provide a basis for subsequent analysis of crowd behavior patterns and optimization of robot paths. The purpose of recording obstacle avoidance history is to accumulate enough data samples to understand when, where, and under what circumstances the robot needs to avoid the same crowd, thereby providing support for prediction and optimization. Specifically, the robot will record information including: obstacle avoidance time, obstacle avoidance location, crowd characteristics (crowd size, average moving speed, moving direction, etc.), obstacle avoidance strategy, obstacle avoidance duration, and robot transfer to management, etc. These data are collected in real time by the robot's sensors (such as lidar, camera) and built-in processors, and stored in local memory or cloud database. The storage format can be a structured log file (such as JSON or CSV) for subsequent analysis.

[0093] 103. Based on the obstacle avoidance history, the same crowd time regularity in the robot's working scene is mined.

[0094] In this step, after collecting sufficient obstacle avoidance history data, the system extracts the co-presence time patterns of the crowd within the robot's working environment through data analysis techniques. These patterns include the frequency, duration, and movement patterns (such as path preferences) of the crowd appearing at specific time periods and locations. The goal of mining is to discover repeatable patterns from historical data to provide decision-making basis for robot path optimization. For example: calculate the number of times the crowd appears, the average duration, and other indicators within different time periods (such as every hour) or specific time periods (such as lunch time 12:00-13:00). Use machine learning algorithms (such as K-means clustering or time series analysis) to identify the crowd's clustering patterns in time and space.

[0095] 104、Based on the co-presence time patterns of the crowd, optimize the robot's daily work route.

[0096] In this step, using the mined co-presence time patterns of the crowd, the robot optimizes its daily work route by adjusting the task plan and path planning. The goal of optimization is to reduce conflicts with the crowd, improve task completion efficiency, and ensure that the movement does not interfere with the crowd. For example: in time periods with high crowd frequency, the robot prioritizes tasks in low-crowd areas or reduces activities in high-crowd areas. When planning the path, prefer to avoid crowded areas and use alternative routes. According to the crowd movement pattern, adjust the task execution order and time to make the robot move as much as possible when the crowd is less.

[0097] By recording obstacle avoidance history, the robot accumulates detailed data basis; by mining the co-presence time patterns of the crowd, the system reveals the spatiotemporal patterns of crowd activity; by optimizing the daily work route, the robot realizes efficient and safe task execution.

[0098] Figure 4 A schematic diagram of a robot obstacle avoidance control system is provided for the embodiments of the present application, as shown in Figure 4 The system comprises:

[0099] The obstacle avoidance control module 100 is used to control the robot to avoid obstacles when the robot needs to avoid the co-presence crowd, so as to minimize the impact on the co-presence crowd continuing to co-presence.

[0100] The obstacle avoidance control module controls the robot to avoid obstacles to minimize the impact on the co-presence crowd continuing to co-presence, comprising:

[0101] Step a, whenever the prediction limit is updated, repeat steps b to d until the co-presence crowd begins to move away from the robot;

[0102] Step b, predict the crowd distribution at multiple future unit times within the updated prediction limit of the co-presence crowd;

[0103] Step c, for each future unit time, based on the influence area demarcation rule, demarcate the influence area of the crowd distribution at the future unit time, and associate with the future unit time;

[0104] Step d, the robot is controlled to avoid obstacles, so that for each future unit time, the robot does not enter the influence area associated with the future unit time.

[0105] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application also intends to include these modifications and variations.

Claims

1. A method for controlling obstacle avoidance of a robot, characterized by, The method comprises: when the robot needs to avoid obstacles from the crowd, the robot is controlled to avoid obstacles to minimize the impact on the crowd continuing to walk together; the robot is controlled to avoid obstacles to minimize the impact on the crowd continuing to walk together, comprising: step a, repeating steps b to d whenever the prediction limit is updated until the crowd starts to move away from the robot; step b, predicting the crowd distribution of the crowd at multiple future unit time points within the prediction limit after the update; step c, for each future unit time point, the influence area of the crowd distribution at the future unit time point is delineated based on the influence area delineation rule, and is associated with the future unit time point; step d, the robot is controlled to avoid obstacles, so that for each future unit time point, the robot does not enter the influence area associated with the future unit time point, wherein the influence area delineation rule comprises: for each non-terminal future unit time point, the ground positions of the same person in the crowd distribution at the future unit time point and the crowd distribution at the next future unit time point are connected on the environmental perception map of the robot, a plurality of position connections are obtained, target persons are determined from the crowd distribution at the future unit time point, a minimum rectangle enclosing all position connections and a circular area with all target person position connections as diameters are determined, and de-duplication is performed to obtain the influence area of the crowd distribution at the future unit time point; wherein the minimum distance between the target person and other persons in the crowd distribution at the future unit time point is less than a preset distance threshold or the position connection of the target person intersects with the position connection of other persons; for the last future unit time point, the minimum enclosing circle enclosing the ground positions of all persons in the crowd distribution at the future unit time point is determined on the environmental perception map, and the minimum enclosing circle is enlarged by a preset radius multiple to obtain the influence area of the crowd distribution at the future unit time point.

2. The robot obstacle avoidance control method of claim 1, wherein, The prediction limit comprises: the highest time limit to ensure that the predicted crowd distribution is reliable. 3.The robot obstacle avoidance control method of claim 1, wherein, The step b of predicting the crowd distribution of the crowd at multiple future unit time points within the prediction limit after the update comprises: based on the pre-trained crowd movement prediction model, the crowd distribution of the crowd at multiple future unit time points within the prediction limit after the update is predicted according to the historical movement information of the crowd.

4. The robot obstacle avoidance control method of claim 1, wherein, In the step d, when the robot needs to enter the influence area associated with each remaining future unit time point due to limited avoidance space, the following steps are further included: on the environmental perception map, multiple movement paths of the robot are planned to enter the respective associated influence areas at each remaining future unit time point in turn; based on the movement path with the minimum quantitative value of the impact on the crowd continuing to walk together, the robot is controlled to move.

5. The robot obstacle avoidance control method of claim 4, wherein, The step of obtaining the quantitative value of the impact on the crowd continuing to walk together comprises: For each remaining future unit time, a local path executed by the robot at the future unit time is intercepted from the moving path, a moving direction of the local path is extracted, a connection direction of a plurality of position connections of a connection line associated with the future unit time in the connection line is traced, a number of connection directions falling into a preset standard direction relationship range between the moving direction is counted, and the future unit time is associated; wherein the preset standard direction relationship range contains a plurality of standard direction relationships in which the representative person will produce a reaction to avoid the robot. Based on the preset weights corresponding to the different remaining future unit times, the numbers associated with the respective future unit times are weighted and calculated to obtain a quantitative value; wherein the preset weights and the early or late degree of the remaining future unit times are in a negative correlation. 6.The robot obstacle avoidance control method of claim 1, wherein, Further comprising: controlling the robot to record an obstacle avoidance history of avoiding the same group of people; based on the obstacle avoidance history, mining a group of people time regularity in the robot working scene; based on the group of people time regularity, optimizing the daily working route of the robot.

7. A robot obstacle avoidance control system characterized by, comprising: an obstacle avoidance control module for controlling the robot to avoid the same group of people when the robot needs to avoid the same group of people, so as to minimize the impact on the same group of people continuing to walk together; the obstacle avoidance control module controls the robot to avoid the same group of people, so as to minimize the impact on the same group of people continuing to walk together, including: step a, whenever the prediction limit is updated, steps b to d are repeatedly executed until the same group of people begins to move away from the robot; step b, predicting the distribution of the same group of people at a plurality of future unit times within the prediction limit after the update; step c, for each future unit time, based on the influence area delimiting rule, delimiting the influence area of the distribution of the same group of people at the future unit time, and associating the future unit time; step d, controlling the robot to avoid the same group of people, so that for each future unit time, the robot does not enter the influence area associated with the future unit time, wherein the influence area delimiting rule comprises: for each non-terminal future unit time, on the environmental perception map of the robot, the distribution of the same group of people at the future unit time and the ground positions of the same people in the distribution of the same group of people at the next future unit time are connected to obtain a plurality of position connections, target persons are determined from the distribution of the same group of people at the future unit time, a minimum rectangle surrounding all position connections is determined, and a region of a circle with all target person position connections as diameters is determined, and de-duplication is performed to obtain the influence area of the distribution of the same group of people at the future unit time; wherein the minimum distance between the target person and other persons in the distribution of the same group of people at the future unit time is less than a preset distance threshold or the position connection of the target person intersects with the position connection of other persons; for the terminal future unit time, on the environmental perception map, a minimum enclosing circle enclosing the ground positions of all persons in the distribution of the same group of people at the future unit time is determined, the minimum enclosing circle is enlarged according to a preset radius multiple to obtain the influence area of the distribution of the same group of people at the future unit time.

Images