Dynamic obstacle avoidance and path optimization method based on intelligent robot

Abstract

The invention discloses a dynamic obstacle avoidance and path optimization method based on a body intelligence robot, and relates to the technical field of robot path planning, and the method comprises the following steps: collecting map data and task constraint data; performing path planning analysis on the map data and the task constraint data based on a set path optimization algorithm to generate initial planning path data; according to the intelligent robot dynamic obstacle avoidance and path optimization method, a robot actual driving path prediction model is constructed, and a target function is constructed for the deviation between the robot actual driving path prediction model and the real-time path planning data of the next time period to carry out minimization analysis; and the real-time path planning data of the next time period is optimized, so that the robot runs according to the optimized real-time path planning data to obtain an actual running path which fits the original real-time path planning data, and the running error of the robot is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot path planning, in particular to a robot dynamic obstacle avoidance and path optimization method based on embodied intelligence. BACKGROUND

[0002] Embodied intelligence refers to an intelligent system with a physical body (physical entity or virtual avatar) that can understand the environment, learn and complete tasks through perception, interaction and execution. The application of embodied intelligence in the field of robot dynamic obstacle avoidance and path optimization can achieve intelligent obstacle avoidance and path optimization based on the robot's own state and parameters combined with the perceived environmental parameters, so that the planned path meets the limitations of the robot's own state and parameters. For example, the passable width of the planned obstacle avoidance path needs to meet the volume limitation of the robot, and the turning radius and the inside-outside wheel difference of the robot need to be met at the turning point.

[0003] However, the precision of the dynamic obstacle avoidance and path optimization of the embodied intelligent robot not only depends on the planning of the algorithm itself, but also depends on the precision of the sensor and the actuator. The jitter of the robot during driving may cause deviation in the data collected by the sensor, and the execution error of the actuator in controlling the robot driving process may cause the robot driving path to deviate from the planned path. SUMMARY

[0004] The purpose of the present application is to provide a robot dynamic obstacle avoidance and path optimization method based on embodied intelligence to solve the above problems in the prior art.

[0005] In order to achieve the above purpose, the present application provides the following technical scheme: a robot dynamic obstacle avoidance and path optimization method based on embodied intelligence, comprising the following steps:

[0006] S1, collecting map data and task constraint data; wherein the map data can be a grid map, a point cloud map or a topological map, etc. The structure information of the target area can be collected by laser radar, depth camera, etc. Then the grid map or point cloud map can be generated by processing algorithms such as Gmapping, Hector SLAM, LOAM, Cartographer or ORB-SLAM3. Then the topological map can be extracted from the grid map or point cloud map based on region segmentation or key point extraction technology.

[0007] S2, based on setting path optimization algorithm, the map data and task constraint data are analyzed for path planning to generate initial planning path data; for example, the path optimization algorithm can select A* algorithm, RRT* (rapidly-exploring random tree star) algorithm, Hybrid A* algorithm, etc. according to needs.

[0008] S3, the robot travels according to the initial planning path data, and monitors and analyzes its own travel state and the distance between the robot and the obstacle and the path boundary in real time to generate real-time monitoring data; for example, for a differential drive robot, its own travel state can include the monitored linear velocity, angular velocity, battery voltage, left wheel motor current and right wheel motor current, etc.; for an Ackerman steering robot, i.e. a robot similar to a car steering geometry, its own travel state can include the monitored actual vehicle speed, front wheel steering angle, yaw rate, longitudinal acceleration and lateral acceleration, etc. The distance between the robot and the obstacle and the path boundary can be detected and identified by sensors such as laser radar, millimeter wave radar or depth camera, and the distance between the robot and the obstacle or boundary can be calculated, and the distance between the robot and the obstacle and the path boundary can be calculated in combination with the map data. The linear velocity, angular velocity, actual vehicle speed, yaw rate, longitudinal acceleration and lateral acceleration can be obtained by IMU (inertial sensor) measurement and analysis.

[0009] S4, real-time update the map data and the initial planning path data based on the real-time monitoring data respectively to generate real-time map data and real-time planning path data, and the update process is the same as S1 and S2.

[0010] S5, analyze and process the path actually traveled by the robot based on the real-time monitoring data and the real-time map data to generate robot actual travel path data;

[0011] S6, based on the real-time path planning data, real-time monitoring data and robot actual travel path data in the sliding window set in the historical data and task constraint data, train a deep network model to obtain a robot actual travel path prediction model, and based on the robot actual travel path prediction model, analyze and process the real-time path planning data of the current sliding window and the next sliding window, and the real-time monitoring data and robot actual travel path data in the current sliding window to output the robot actual travel path prediction data of the next sliding window;

[0012] S7, based on the real-time planning path data and the robot actual travel path prediction data, perform path deviation analysis and processing to generate real-time path deviation analysis data, and based on the real-time deviation analysis data, correct the real-time planning path data to generate optimized path data.

[0013] Further, the task constraint data includes pose data of the starting point and the ending point, speed limit data, acceleration limit data, jerk limit data (acceleration rate limit) and robot geometric size data, etc.

[0014] Further, the initial planning path data includes the horizontal coordinate, the vertical coordinate and the heading angle of each path point corresponding to each time point.

[0015] Further, the S5 includes the following steps:

[0016] ID codes are set for the boundary points of each obstacle and the path boundary points in the real-time map data, to obtain boundary point ID data;

[0017] Each boundary point ID data is associated with the coordinate data of the boundary points of the corresponding obstacle or the path boundary points in the real-time map data, to generate boundary point ID coordinate data;

[0018] The distance information between the robot and the obstacles and the distance information between the robot and the path boundaries are extracted from the real-time monitoring data, to obtain the distance data between the robot at each time point and each boundary point ID data monitored, and to generate robot relative coordinate data; for example, the robot is 50 cm away from boundary point 1, 60 cm away from boundary point 2, and 55 cm away from boundary point 3;

[0019] The boundary point ID coordinate data corresponding to the boundary point ID data in the robot relative coordinate data is searched out in the boundary point ID coordinate data, and then combined one by one with the corresponding robot relative coordinate data, to generate reference boundary point ID coordinate and distance data pairs;

[0020] Based on the EKF (Extended Kalman Filter) algorithm, the robot actual coordinate data at the previous time point and the velocity data of the robot in each coordinate component (such as the velocity components in the x-axis and the y-axis), as well as the velocity data of the robot in each coordinate component at the current time point and the reference boundary point ID coordinate and distance data pairs are analyzed and processed, to estimate the optimal robot actual coordinate data at the current time point and the velocity data of the robot in each coordinate component; wherein the robot actual coordinate data at the starting point can be obtained through the pose data of the starting point in the task constraint data, and the velocity data of the robot in each coordinate component at the starting point is 0.

[0021] The robot actual coordinate data is collected and arranged in the corresponding time sequence, to generate robot actual driving path data.

[0022] Further, the S6 includes the following steps to train the deep network model and obtain the robot actual driving path prediction model:

[0023] The time stamps of the real-time path planning data, the real-time monitoring data and the robot actual driving path data in the historical data are aligned;

[0024] Based on the S-shaped speed curve algorithm, the actual driving speed corresponding to each path planning coordinate is analyzed based on the task constraint data and real-time path planning data, and path planning coordinate target speed data is obtained.

[0025] The path planning coordinate target speed data is combined with the real-time path planning data at the corresponding path planning coordinate to generate a real-time path planning vector at the corresponding time;

[0026] Based on the real-time monitoring data, a real-time monitoring vector at each time is generated, and based on the robot actual driving path data, a robot actual driving path vector at each time is generated;

[0027] The real-time path planning vector, real-time monitoring vector and robot actual driving path vector in one setting sliding window, and the real-time path planning vector in the next setting sliding window are collected to generate a model input vector sequence; wherein the time periods corresponding to the two setting sliding windows are continuous on the time axis;

[0028] The robot actual driving path vector in the next setting sliding window is collected to generate a model output vector sequence;

[0029] Each dimension of each vector in the model input vector sequence and model output vector sequence is subjected to Z-score standardization processing to generate a model input standard vector sequence and a model output standard vector sequence, respectively; when performing Z-score standardization processing, the mean and standard deviation of each dimension on all training samples are recorded, facilitating subsequent restoration of the standardized data value to the data value before standardization;

[0030] Based on the model input standard vector sequence and model output standard vector sequence, a deep neural network model is trained, and each dimension of each model output standard vector in the model output standard vector sequence is inversely restored to a robot actual driving path vector in the output layer, obtaining a robot actual driving path prediction model, which is used to output a predicted robot actual driving path vector sequence of the next sliding window based on the model input standard vector sequence of the current setting sliding window, generating robot actual driving path prediction data. Wherein, when selecting a deep neural network model, LSTM (Long Short Term Memory Network), GRU (Gated Recurrent Unit), TCN (Temporal Convolutional Network) and TransformerEncoder can be selected for training according to needs.

[0031] Further, the S7 comprises the following steps:

[0032] The robot's actual driving path prediction data is subtracted from the corresponding dimension data in the real-time path planning data in the corresponding sliding window to obtain the real-time path deviation analysis data. That is, the real-time path deviation analysis data = robot's actual driving path prediction data - real-time path planning data. Furthermore, the value of each dimension of the real-time path deviation analysis data at each moment is equal to the value of each dimension (such as horizontal coordinate / vertical coordinate / heading angle) in each robot's actual driving path vector sequence corresponding to the robot's actual driving path prediction data, minus the corresponding feature (horizontal coordinate / vertical coordinate / heading angle, corresponding to the feature dimension selected by the robot's actual driving path vector) of the real-time path planning data at the same moment.

[0033] The real-time path deviation analysis data is used to compensate the real-time path planning data in the corresponding sliding window to obtain optimized path data, i.e., optimized path data = real-time path planning data + real-time path deviation analysis data.

[0034] Furthermore, S7 comprises the following steps:

[0035] The real-time path planning data in the next sliding window of the current sliding window is used as the variable to be optimized to obtain the future path planning vector sequence. The future path planning vector sequence is composed of the future path planning vectors corresponding to all times in the corresponding sliding window arranged in chronological order. The dimension and feature type of each dimension of the future path planning vector are the same as the actual driving path vector of the robot.

[0036] Based on the robot's actual driving path prediction model, the robot's actual driving path prediction data at each time step is calculated as the sum of the squared L2 norms of the differences between the predicted data and the future path planning vector at the same time step, weighted by a weighted matrix. An optimization model for predicting the robot's actual driving path is then constructed. For example, the weighted matrix W can be expressed as: , , and These represent the differences between the robot's actual driving path prediction data predicted by the robot's actual driving path prediction model at each time point and the future path planning vector at the same time point in the horizontal, vertical, and heading angle dimensions, respectively.

[0037] Set the constraints for the robot's actual driving path prediction optimization model;

[0038] The robot actual driving path prediction optimization model is minimized to obtain the future path planning vector sequence of the robot actual driving path prediction optimization model minimum, and the optimization path data is generated, and the gradient descent method, sequence quadratic programming (SQP), memory limited quasi-Newton method (L-BFGS-B), differential evolution (DE) and the like can be used in the minimization solving.

[0039] Further, the S6 further comprises collecting robot actual driving path data, constructing a new model input vector sequence and a model output vector sequence, generating a new model input vector sequence and a new model output vector sequence, and regularly updating the robot actual driving path prediction model based on the new model input vector sequence and the new model output vector sequence, comprising the following steps:

[0040] Collecting robot actual driving path data and corresponding real-time path planning data and real-time monitoring data and aligning the time stamps, constructing a new model input vector sequence and a new model output vector sequence;

[0041] Every setting update time, randomly extracting a setting new sample number of pairs from all new model input vector sequences and new model output vector sequences, and randomly extracting a setting old sample number of pairs from all historical model input vector sequences and model output vector sequences, and putting them into an update sample set.

[0042] Updating the robot actual driving path prediction model based on the update sample set.

[0043] Advantages:

[0044] 1. Compared with the prior art, the robot dynamic obstacle avoidance and path optimization method based on embodied intelligence provided by the application can predict the robot actual driving path prediction model according to the real-time monitoring data, planning path data, robot actual driving path data of the current period and the planning path data of the next period, and can compensate the planning path by analyzing the deviation of each dimension feature of the actual path and the planning path at each time, so that the actual driving path and the planning path of the robot in the next period are consistent, and the robot driving error is reduced.

[0045] 2. Compared with the prior art, the robot dynamic obstacle avoidance and path optimization method based on embodied intelligence provided by the application minimizes the deviation of the actual driving path prediction model of the robot and the real-time path planning data of the next period by constructing a target function, optimizes the real-time path planning data of the next period, and makes the robot drive according to the optimized real-time path planning data, so that the actual driving path is consistent with the original real-time path planning data, and the robot driving error is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description only represent some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0047] Figure 1 The method step diagram provided for the embodiments of the present application is provided.

[0048] Figure 2 Another embodiment step diagram of S7 provided for the embodiments of the present application is provided. DETAILED DESCRIPTION

[0049] In order to make those skilled in the art better understand the technical solutions of the present application, the present application will be further described in detail with reference to the drawings.

[0050] In the description of the present application, the terms "first", "second" are only used for description purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited. In addition, the terms "mounting", "connecting", "connecting" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be directly connected, or indirectly connected through intermediate medium, or the communication between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0051] In the following, the example embodiments will be described more fully with reference to the accompanying drawings, but the example embodiments can be embodied in different forms and should not be interpreted as being limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the present disclosure thorough and complete, and to enable those skilled in the art to fully understand the scope of the present disclosure.

[0052] In the case of no conflict, the embodiments of the present disclosure and the features in the embodiments can be combined with each other.

[0053] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0054] The terms used herein are only used to describe specific embodiments and are not intended to limit the present disclosure. As used herein, the singular forms "a" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the terms "comprise" and / or "consist of", when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0055] The embodiments described herein can be described with reference to plan views and / or cross-sectional views by virtue of the idealized schematic illustrations of the present disclosure. Thus, the example illustrations can be modified according to manufacturing techniques and / or tolerances. Therefore, the embodiments are not limited to the illustrated embodiments in the drawings, but include modifications based on manufacturing processes. Thus, the regions illustrated in the drawings have schematic properties and the shapes of the regions shown in the drawings do not necessarily illustrate the specific shape of the regions of the elements, but are intended to represent the exemplary nature of the embodiments.

[0056] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and the present disclosure.

[0057] Please refer to Figure 1 A robot dynamic obstacle avoidance and path optimization method based on embodied intelligence, comprising the following steps:

[0058] S1, collecting map data and task constraint data;

[0059] The map data can be a grid map, a point cloud map, or a topological map, etc. The structure information of the target area can be collected by a laser radar, a depth camera, etc., and then a grid map or a point cloud map can be generated by processing algorithms such as Gmapping, Hector SLAM, LOAM, Cartographer, or ORB-SLAM3, and then a topological map can be extracted from the grid map or the point cloud map based on region segmentation or key point extraction technology;

[0060] The task constraint data includes pose data of the starting point and the ending point, speed limit data, acceleration limit data, jerk limit data (acceleration rate limit), and robot geometric size data, etc.

[0061] S2, path planning analysis is performed on the map data and task constraint data based on a path optimization algorithm to generate initial planning path data; for example, the path optimization algorithm can select an A* algorithm, an RRT* (rapidly-exploring random tree star) algorithm, a Hybrid A* algorithm, etc. according to needs. The initial planning path data includes the horizontal coordinate, the vertical coordinate, and the heading angle of the path point corresponding to each time point.

[0062] S3, the robot travels according to the initial planning path data, and real-time monitoring and analysis are performed on the self-traveling state and the distance between the robot and the obstacle and the path boundary to generate real-time monitoring data; for example, for a differential drive robot, the self-traveling state can include the line speed, the angular speed, the battery voltage, the left wheel motor current, and the right wheel motor current monitored; for an Ackerman steering robot, i.e., a robot similar to a car steering geometry, the self-traveling state can include the actual vehicle speed, the front wheel steering angle, the yaw rate, the longitudinal acceleration, and the lateral acceleration monitored. The distance between the robot and the obstacle and the path boundary can be detected and identified by a laser radar, a millimeter wave radar, or a depth camera, etc. sensor, and the distance between the robot and the obstacle or the boundary is calculated, and the distance between the robot and the obstacle and the path boundary is calculated in combination with the map data. The line speed, the angular speed, the actual vehicle speed, the yaw rate, the longitudinal acceleration, and the lateral acceleration can be obtained by measuring and analyzing an IMU (inertial sensor).

[0063] S4, real-time updating is performed on the map data and the initial planning path data based on the real-time monitoring data to generate real-time map data and real-time planning path data, and the updating process is the same as that of S1 and S2.

[0064] S5, the path actually traveled by the robot is analyzed and processed based on the real-time monitoring data and the real-time map data to generate robot actual travel path data, including the following steps:

[0065] S5.1, ID codes are set for the boundary points of each obstacle and the path boundary points in the real-time map data to obtain boundary point ID data;

[0066] S5.2, each boundary point ID data is associated with the coordinate data of the boundary points of the obstacle or the path boundary points corresponding to the real-time map data to generate boundary point ID coordinate data;

[0067] S5.3, extract the distance between the robot and the obstacle and the distance between the robot and the path boundary from the real-time monitoring data, obtain the distance data between the robot at each time and each boundary point ID data monitored, and generate the relative coordinate data of the robot; for example, the distance between the robot and boundary point 1 is 50 cm, the distance between the robot and boundary point 2 is 60 cm, and the distance between the robot and boundary point 3 is 55 cm;

[0068] S5.4, search for the boundary point ID coordinate data corresponding to the boundary point ID data in the relative coordinate data of the robot in the boundary point ID coordinate data, and then one-to-one corresponding combination with the corresponding relative coordinate data of the robot, to generate the reference boundary point ID coordinate and distance data pair;

[0069] S5.5, based on the EKF (extended Kalman filter) algorithm, analyze and process the robot actual coordinate data at the last time and the speed data of the robot in each coordinate component (such as the speed component in the x-axis and y-axis), and the speed data of the robot in each coordinate component at the current time and the reference boundary point ID coordinate and distance data pair, to estimate the optimal robot actual coordinate data at the current time and the speed data of the robot in each coordinate component; wherein the robot actual coordinate data at the starting point can be obtained by the pose data of the starting point in the task constraint data, and the speed data of the robot in each coordinate component at the starting point is 0;

[0070] S5.6, collect and arrange the robot actual coordinate data in the corresponding time sequence to generate the robot actual driving path data.

[0071] S6, based on the real-time path planning data, real-time monitoring data and robot actual driving path data in the sliding window set in the task constraint data and historical data, train the deep network model to obtain the robot actual driving path prediction model, and based on the robot actual driving path prediction model, analyze and process the real-time path planning data of the current sliding window and the next sliding window, and the real-time monitoring data and robot actual driving path data in the current sliding window, and output the robot actual driving path prediction data of the next sliding window;

[0072] Wherein, to train the deep network model and obtain the robot actual driving path prediction model, the steps are as follows:

[0073] S6.1, align the time stamps of the real-time path planning data, real-time monitoring data and robot actual driving path data in the historical data;

[0074] S6.2, based on the S-shaped speed curve algorithm, analyze the actual driving speed corresponding to each path planning coordinate in the task constraint data and the real-time path planning data, to obtain the path planning coordinate target speed data;

[0075] S6.3, combine the path planning coordinate target speed data with the real-time path planning data at the corresponding path planning coordinate to generate a real-time path planning vector corresponding to the time point;

[0076] S6.4, generate a real-time monitoring vector at each time point based on the real-time monitoring data, and generate a robot actual driving path vector at each time point based on the robot actual driving path data;

[0077] S6.5, collect the real-time path planning vector, the real-time monitoring vector and the robot actual driving path vector in one setting sliding window, and the real-time path planning vector in the next setting sliding window, to generate a model input vector sequence; wherein the time periods corresponding to the two setting sliding windows are continuous on the time axis;

[0078] S6.6, collect the robot actual driving path vector in the next setting sliding window to generate a model output vector sequence;

[0079] S6.7, Z-score standardization processing is performed on each dimension of each vector in the model input vector sequence and the model output vector sequence to generate a model input standard vector sequence and a model output standard vector sequence respectively; the mean and standard deviation of each dimension on all samples participating in training are recorded when performing Z-score standardization processing, to facilitate subsequent restoration of the standardized data value to the data value before standardization;

[0080] S6.8, based on the model input standard vector sequence and the model output standard vector sequence, a deep neural network model is trained, and each dimension of each model output standard vector in the model output standard vector sequence is inversely restored to the robot actual driving path vector in the output layer to obtain a robot actual driving path prediction model, which is used to output the predicted robot actual driving path vector sequence of the next sliding window according to the model input standard vector sequence of the current setting sliding window, to generate robot actual driving path prediction data. When selecting a deep neural network model, LSTM (Long Short-Term Memory Network), GRU (Gated Recurrent Unit), TCN (Temporal Convolutional Network) and TransformerEncoder can be selected for training according to needs.

[0081] Further, robot actual driving path data can also be collected to construct new model input vector sequences and model output vector sequences, generate new model input vector sequences and new model output vector sequences, and periodically update the robot actual driving path prediction model based on the new model input vector sequences and the new model output vector sequences, including the following steps;

[0082] S6.8.1, collect the robot actual driving path data and the corresponding real-time path planning data and real-time monitoring data and align the time stamps, construct a new model input vector sequence and a new model output vector sequence;

[0083] S6.8.2, every set update time, randomly extract a set of new sample pairs from all new model input vector sequences and new model output vector sequences, and randomly extract a set of old sample pairs from all historical model input vector sequences and model output vector sequences, and put them into the update sample set.

[0084] S6.8.3, update the robot actual driving path prediction model based on the update sample set.

[0085] S7, based on the real-time planning path data and the robot actual driving path prediction data, perform path deviation analysis processing to generate real-time path deviation analysis data, based on the real-time deviation analysis data, correct the real-time planning path data to generate optimized path data, including the following steps:

[0086] S7.1, subtract each dimension data in the robot actual driving path prediction data from the corresponding real-time path planning data in the sliding window to obtain the real-time path deviation analysis data, that is, real-time path deviation analysis data = robot actual driving path prediction data - real-time path planning data, further, the value of each dimension of the real-time path deviation analysis data at each time is equal to the numerical value of each dimension (such as horizontal coordinate / vertical coordinate / heading angle) in each robot actual driving path vector in the corresponding robot actual driving path vector sequence of the robot actual driving path prediction data minus the corresponding feature (horizontal coordinate / vertical coordinate / heading angle, corresponding to the selected feature dimension of the robot actual driving path vector) data of the corresponding real-time path planning data at the same time;

[0087] S7.2, use the real-time path deviation analysis data to compensate the real-time path planning data in the corresponding sliding window to obtain the optimized path data, that is, optimized path data = real-time path planning data + real-time path deviation analysis data.

[0088] Please refer to Figure 2 , in another embodiment, S7 can also be the following steps:

[0089] S7.a, use the real-time path planning data in the next sliding window of the current sliding window as the to-be-optimized variable to obtain a future path planning vector sequence, the future path planning vector sequence is composed of all corresponding future path planning vectors in the corresponding sliding window in time sequence, the dimension and each dimension feature type of the future path planning vector are the same as those of the robot actual driving path vector;

[0090] S7.b, the difference between the data of the robot actual driving path prediction data predicted by the robot actual driving path prediction model at each time and the future path planning vector at the same time is weighted by a set weight matrix, and the sum of the square of the L2 norm is used to construct a robot actual driving path prediction optimization model; the robot actual driving path prediction optimization model can be represented as:

[0091]

[0092] wherein U represents the future path planning vector sequence to be optimized, is the number of times corresponding to the next sliding window in the prediction time domain, k is the kth time in the next sliding window, is the robot actual driving path prediction data of the next sliding window predicted by the robot actual driving path prediction model, represents the future path planning vector at the kth time in the next sliding window, W is a set weight matrix W, which can be represented as: and respectively represent the difference between the data of the robot actual driving path prediction data predicted by the robot actual driving path prediction model at each time and the future path planning vector at the same time in the horizontal coordinate, vertical coordinate and heading angle dimension; represents the square of the L2 norm with the weight matrix W;

[0093] S7.c, setting the constraint condition of the robot actual driving path prediction optimization model; the constraint condition can be: the acceleration of the optimized output future path planning vector sequence (optimized path data) is less than or equal to the set maximum acceleration, the curvature of the corresponding path is less than or equal to the set maximum curvature, the speed of the robot does not exceed the set speed limit data, etc.

[0094] S7.d, the robot actual driving path prediction optimization model is minimized to satisfy the constraint condition, and the future path planning vector sequence at the minimum of the robot actual driving path prediction optimization model is obtained, and the optimized path data is generated; the gradient descent method, the sequential quadratic programming (SQP), the limited memory quasi-Newton method (L-BFGS-B), the differential evolution (DE) and the like can be used in the minimization solving.

[0095] The above only describes certain exemplary embodiments of the present application in a descriptive manner, and it is needless to say that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present application. Therefore, the above figures and description are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the present application.​​

Claims

1. A method for dynamic obstacle avoidance and path optimization of robots based on embodied intelligence, characterized in that, Includes the following steps: S1. Collect map data and task constraint data; S2. Based on the path optimization algorithm, perform path planning analysis on the map data and task constraint data to generate initial planned path data; S3. The robot travels according to the initial planned path data, and monitors and analyzes its own travel status and the distance between the robot and obstacles and path boundaries in real time, generating real-time monitoring data. S4. Based on the real-time monitoring data, update the map data and the initial planned path data in real time to generate real-time map data and real-time planned path data; S5. Based on the real-time monitoring data and real-time map data, analyze and process the actual path traveled by the robot to generate actual robot travel path data. S6. Based on the task constraint data and historical data, set the real-time path planning data, real-time monitoring data and robot actual driving path data in the sliding window, train a deep network model to obtain the robot actual driving path prediction model, and analyze and process the real-time path planning data of the current sliding window and the next sliding window, as well as the real-time monitoring data and robot actual driving path data in the current sliding window, based on the robot actual driving path prediction model, and output the robot actual driving path prediction data for the next sliding window. S7. Based on the real-time planned path data and the robot's actual driving path prediction data, perform path deviation analysis processing to generate real-time path deviation analysis data. Based on the real-time deviation analysis data, perform correction processing on the real-time planned path data to generate optimized path data.

2. The method for dynamic obstacle avoidance and path optimization of a robot based on embodied intelligence according to claim 1, characterized in that, The task constraint data includes pose data of the start and end points, velocity limit data, acceleration limit data, jerk limit data, and robot geometric dimension data.

3. The method for dynamic obstacle avoidance and path optimization of a robot based on embodied intelligence according to claim 1, characterized in that, The initial planned path data includes the x-coordinate, y-coordinate, and heading angle of the path point at each time point.

4. The method for dynamic obstacle avoidance and path optimization of a robot based on embodied intelligence according to claim 1, characterized in that, S5 includes the following steps: Set ID codes for the boundary points of each obstacle and the path boundary points in the real-time map data to obtain boundary point ID data; Associate each boundary point ID data with the coordinate data of the boundary points of obstacles or path boundary points in the real-time map data to generate boundary point ID coordinate data; The distance information between the robot and obstacles and between the robot and the path boundary is extracted from the real-time monitoring data to obtain the distance data between the robot and the ID data of each monitored boundary point at each time moment, and to generate the robot's relative coordinate data. Search for the boundary point ID coordinate data that corresponds to the boundary point ID data in the robot relative coordinate data in the boundary point ID coordinate data, and then combine it with the corresponding robot relative coordinate data one by one to generate a reference boundary point ID coordinate and distance data pair. Based on the EKF algorithm, the robot's actual coordinate data and velocity data in each coordinate component at the previous time step, as well as the robot's velocity data in each coordinate component at the current time step and the coordinates and distance data of the reference boundary point ID, are analyzed and processed to estimate the optimal robot's actual coordinate data and velocity data in each coordinate component at the current time step. The robot's actual coordinate data is collected and arranged in chronological order according to the corresponding time, and the robot's actual driving path data is generated.

5. The method for dynamic obstacle avoidance and path optimization of a robot based on embodied intelligence according to claim 1, characterized in that, The deep network model trained in S6 is used to obtain the robot's actual driving path prediction model, including the following steps: Align the timestamps of real-time path planning data, real-time monitoring data, and actual robot driving path data in historical data; Based on the S-shaped speed curve algorithm, the actual driving speed corresponding to each path planning coordinate is analyzed based on the task constraint data and real-time path planning data to obtain the target speed data of the path planning coordinates. The target velocity data at the path planning coordinates is combined with the real-time path planning data at the corresponding path planning coordinates to generate the real-time path planning vector at the corresponding time. Real-time monitoring vectors are generated at each time based on the real-time monitoring data, and actual robot travel path vectors are generated at each time based on the actual robot travel path data. The real-time path planning vector, real-time monitoring vector, and robot actual driving path vector in one setting sliding window, as well as the real-time path planning vector in the next setting sliding window, are collected to generate a model input vector sequence. Collect the actual robot path vectors in the next setting sliding window and generate a model output vector sequence; Z-score normalization is performed on each dimension of each vector in the model input vector sequence and the model output vector sequence to generate the model input standard vector sequence and the model output standard vector sequence, respectively. Based on the standard vector sequence of model input and model output, a deep neural network model is trained. In the output layer, each dimension of each model output standard vector in the model output standard vector sequence is inversely converted into the robot's actual driving path vector, resulting in a robot actual driving path prediction model. This model is used to predict the robot's actual driving path vector sequence for the next sliding window based on the model input standard vector sequence of the currently set sliding window, thereby generating robot actual driving path prediction data.

6. The method for dynamic obstacle avoidance and path optimization of a robot based on embodied intelligence according to claim 1, characterized in that, S7 includes the following steps: The actual driving path prediction data of the robot is subtracted from the corresponding data of each dimension in the real-time path planning data in the corresponding sliding window to obtain the real-time path deviation analysis data. The real-time path deviation analysis data is used to compensate the real-time path planning data in the corresponding sliding window to obtain optimized path data.

7. The method for dynamic obstacle avoidance and path optimization of a robot based on embodied intelligence according to claim 1, characterized in that, S7 comprises the following steps: The real-time path planning data in the next sliding window of the current sliding window is used as the variable to be optimized to obtain the future path planning vector sequence. The future path planning vector sequence is composed of the future path planning vectors corresponding to all times in the corresponding sliding window arranged in chronological order. Based on the difference between the robot's actual driving path prediction data at each time point and the future path planning vector at the same time point, the robot's actual driving path prediction optimization model is constructed by summing the squares of the L2 norms weighted by the set weight matrix. Set the constraints for the robot's actual driving path prediction optimization model; The robot's actual driving path prediction optimization model is minimized to obtain the future path planning vector sequence when the robot's actual driving path prediction optimization model is minimized, and the optimized path data is generated.

8. The method for dynamic obstacle avoidance and path optimization of a robot based on embodied intelligence according to claim 4, characterized in that, S6 further includes collecting actual robot driving path data, constructing new model input vector sequence and model output vector sequence, generating new model input vector sequence and new model output vector sequence, and periodically updating the robot actual driving path prediction model based on the new model input vector sequence and new model output vector sequence, including the following steps; Collect the robot's actual driving path data, corresponding real-time path planning data, and real-time monitoring data, align the timestamps, and construct a new model input vector sequence and a new model output vector sequence; Every set update time, randomly select a new number of sample pairs from all new model input vector sequences and new model output vector sequences, and randomly select an old number of sample pairs from all historical model input vector sequences and model output vector sequences, and put them into the update sample set; The prediction model for the robot's actual driving path is updated based on the updated sample set.

Images