Safety trajectory tracking method and system for ground differential drive robot

Abstract

The invention discloses a safe trajectory tracking method and system for a ground differential drive robot, and particularly relates to the technical field of robot operation management. Comprehensive means of multi-source data fusion, dynamic safe distance setting, trajectory prediction uncertainty quantification and trajectory deviation cost optimization are introduced; the path planning capability of the robot in a complex dynamic environment is improved; the behavior characteristics and environment complexity of the dynamic obstacle are evaluated in real time, the safe distance is dynamically adjusted, and the dynamic obstacle area is accurately quantified in combination with the trajectory prediction model, so that the problems of low trajectory tracking precision and insufficient obstacle avoidance robustness in the background technology are solved; the path optimization is realized through the minimization strategy of the trajectory deviation cost function, the operation safety of the robot in diversified scenes is ensured, and meanwhile, the efficiency and stability of path planning are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of robot operation management, and more specifically, to a method and system for tracking the safe trajectory of a ground differential drive robot. Background Art

[0002] Ground differential drive robots occupy an important position in industrial logistics, intelligent warehousing and path planning and obstacle avoidance tasks in complex dynamic environments due to their simple structure and flexible steering. Such robots need to achieve safe and efficient trajectory tracking under conditions where dynamic obstacles are dense and the environment is complex and changeable. This requires the robots to have the ability to perceive the environment in real time, accurately predict the motion trajectory of dynamic obstacles, and flexibly adjust their own paths according to changes in the environment.

[0003] However, existing trajectory prediction and planning methods have obvious deficiencies in many aspects. In terms of trajectory prediction accuracy, traditional methods often rely on fixed safety distance models or simple assumptions about obstacle motion characteristics, ignoring the complexity of the environment and the diversity of dynamic obstacle behavior changes. This causes the robot's trajectory to easily deviate from expectations in actual operation, limiting its obstacle avoidance capabilities. In addition, in setting the dynamic safety distance, existing methods also lack comprehensive consideration of environmental changes and obstacle behavior, making it difficult for robots to respond safely in a timely manner when faced with emergencies. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a safe trajectory tracking method and system for a ground differential drive robot, which solves the problems raised in the above-mentioned background technology by introducing a comprehensive approach of multi-source data fusion, dynamic safety distance setting, trajectory prediction uncertainty quantification and trajectory deviation cost optimization.

[0005] To achieve the above object, the present invention provides the following technical solution: a ground differential drive robot safe trajectory tracking method, comprising the following steps: Step S01, data collection and preprocessing: collect the position information, speed, acceleration and steering angle data of the dynamic obstacle through the sensor, and output the preprocessed obstacle motion trajectory history sequence; Step S02, obstacle trajectory prediction and uncertainty quantification: input the preprocessed obstacle motion trajectory history sequence into the obstacle trajectory prediction model, output the movement trajectory and the first probability distribution map in the short future time; combine with the Bayesian algorithm to output the predicted dynamic obstacle area range and the second probability distribution map; Step S03, robot trajectory planning and dynamic adjustment: according to the type of obstacle and motion characteristics, define a dynamic safety distance for each obstacle, determine the safety distance constraint area based on the dynamic safety distance, and record the overlapping area of ​​the safety distance constraint area and the target trajectory as a risk area; if there is a risk area, update the target trajectory of the robot in real time, output the corrected trajectory, and minimize the trajectory deviation cost function that deviates from the target; Step S04, real-time execution and feedback optimization: according to the corrected trajectory, the robot motion instructions are adjusted and executed in real time, the obstacle avoidance effect is evaluated, and the obstacle trajectory prediction model is optimized according to the obstacle avoidance effect.

[0006] Preferably, the preprocessed obstacle motion trajectory history sequence is obtained to obtain the dynamic characteristics of the obstacle motion trajectory and form a feature matrix F; the feature matrix F is input into the obstacle trajectory prediction model, and the long-term and short-term dependencies are captured through the time gating mechanism, and the mobile prediction trajectory and the first probability distribution map in the short future time are output; combined with the Bayesian inference method, the uncertainty of the mobile prediction trajectory is quantified.

[0007] Preferably, the safety distance constraint area is obtained based on the dynamic obstacle area and the dynamic safety distance, and the robot is safe if it does not enter the safety distance constraint area; the dynamic safety distance is positively correlated with the speed and acceleration characteristics of the obstacle.

[0008] Preferably, the trajectory deviation cost function satisfies the following calculation model:

[0009] in, Refers to the cumulative rate of change of acceleration after executing the corrected trajectory; It refers to the energy consumed to execute the corrected trajectory; It refers to the cumulative distance that the trajectory deviates from the target path after correction; α, β, γ are weight coefficients, and .

[0010] Preferably, the setting method of the dynamic safety distance includes: collecting environmental state parameters during the operation of the robot, the environmental state parameters at least including light intensity, ground reflectivity, sensor noise level, obstacle speed v and acceleration a, calculating an environmental complexity index based on the difference between the environmental state parameters and preset values, and adjusting the dynamic safety distance based on the environmental complexity index; Collect environmental state parameters and calculate the environmental complexity index ECI through the following model:

[0011] in, Indicates the light intensity deviation, represents the ground reflectivity deviation, represents the sensor noise level, represents the dynamic characteristics of the obstacle, Indicates the rate of environmental change, using the cumulative change in environmental status It is expressed as a ratio of the time span; , , , They represent the reference values ​​of illumination intensity, ground reflectivity deviation, sensor noise level, and obstacle dynamic characteristics respectively; , , , , Respectively represent the weights of each item, and c represents the basic complexity constant; Dynamically adjust the safety distance based on the environmental complexity index ECI; The dynamic safety distance Satisfies the following formula:

[0012] in, represents the basic safety distance, which is usually the minimum safety distance of the robot in a static environment and is set based on the obstacle type; k is the environment complexity weight, It is the dynamic obstacle motion characteristic adjustment item, which is positively correlated with the obstacle speed and acceleration, and represents the impact of dynamic motion on the safety distance.

[0013] Preferably, the obstacle avoidance effect quality coefficient is used to quantify the obstacle avoidance effect. If the obstacle avoidance effect quality index is lower than a threshold, it indicates that the obstacle avoidance effect is abnormal, and the obstacle trajectory prediction model is optimized; the obstacle avoidance effect quality coefficient The way to obtain is: Obtaining an obstacle avoidance effect evaluation index, and dividing the obstacle avoidance effect evaluation index into a positive index and a negative index; the positive index refers to an index whose larger value indicates a better obstacle avoidance effect, and the negative index refers to an index whose smaller value indicates a better obstacle avoidance effect;

[0014]

[0015] Among them, N is the number of obstacle avoidance effect evaluation indicators, is the weight of the i-th obstacle avoidance effect evaluation index, satisfying , represents the actual value of the i-th obstacle avoidance effect evaluation index, Represents the normalized nonlinear mapping function; in the normalized nonlinear mapping function represents the ideal value of the i-th obstacle avoidance effect evaluation index; Used to control the normalized sensitivity of the i-th obstacle avoidance effect evaluation index.

[0016] Preferably, the obstacle trajectory prediction model is optimized as follows: Perform data augmentation on existing datasets to generate more diverse training samples; Use a multi-task learning framework to share the feature layer of the obstacle trajectory prediction model and enhance the global understanding ability of the obstacle trajectory prediction model; During the operation of the robot, the actual trajectory and obstacle behavior of the failed obstacle avoidance are collected in real time, and the training data is added to improve the processing capability of the obstacle trajectory prediction model through online updates.

[0017] Preferably, the construction of the obstacle trajectory prediction model includes: Step S11, data set division and preprocessing: obtain the preprocessed obstacle motion trajectory history sequence, divide it into a training set and a test set, extract dynamic features, and normalize the feature values; generate a time series input window; Step S12, model design and construction: design a deep learning model, use time series modeling and multi-layer networks to extract temporal dependencies, set the output layer, and predict points and uncertainty distribution with future trajectories; Step S13, model training and optimization: define a joint loss function, combine trajectory error and uncertainty quantification, use the Adam optimizer to adjust model parameters, optimize hyperparameters through the validation set, record the training process, monitor the loss function and performance, and obtain the trained deep learning model weights; Step S14, model evaluation and testing: using the test set to evaluate the trajectory prediction accuracy, uncertainty quantification effect and generalization ability of the model; Step S15: Export the trained obstacle trajectory prediction model into a deployable format.

[0018] Preferably, the method includes a path planning dynamic adjustment step based on a trajectory deviation cost function, including: Divide the robot's operating environment into several areas according to the grid; Collect real-time data to obtain the real-time environmental complexity index of each area; Based on the starting and ending points of the task and the real-time environmental complexity index of each area, a path planning algorithm is used to generate several candidate paths. The candidate paths meet the priority rules, which include: giving priority to avoiding areas with high environmental complexity index; giving priority to avoiding areas that directly threaten the safety of robot operation; Generate a trajectory deviation cost function for each candidate path; divide the candidate path into several sub-segments according to equal distances; perform weighted summation of the trajectory deviation cost and weight of the sub-segments to obtain the path cost index of each candidate path; The candidate path with the lowest path cost index is selected as the robot execution instruction.

[0019] To achieve the purpose of the present invention, a ground differential drive robot safety trajectory tracking system is provided, comprising: The data acquisition and preprocessing module collects dynamic obstacle information in the robot's operating environment through multiple sensors, including at least position information, speed, acceleration and steering angle data. After the data is timestamp aligned and processed by Kalman filtering, a historical trajectory sequence is generated and the preprocessed obstacle motion trajectory history sequence is output; The safety distance adjustment module obtains the environmental state parameters during the operation of the robot, calculates the environmental complexity index based on the difference between the environmental state parameters and the preset values, and adjusts the dynamic safety distance based on the environmental complexity index; The trajectory prediction and quantification module uses the obstacle trajectory history sequence to predict the obstacle trajectory in the short term in the future through the obstacle trajectory prediction model, and generates a first probability distribution map; combined with the Bayesian inference method, it quantifies the uncertainty of trajectory prediction and outputs the dynamic obstacle area range and the second probability distribution map; The trajectory planning and dynamic adjustment module defines the dynamic safety distance and generates the safety distance constraint area. The overlapping area between the safety distance constraint area and the target trajectory is recorded as the risk area. If there is a risk area, the robot's target trajectory is updated in real time, and the corrected trajectory is output, while minimizing the trajectory deviation cost function that deviates from the target. The real-time execution and feedback module generates and executes motion instructions in real time according to the corrected trajectory, monitors the obstacle avoidance effect through sensors, calculates the obstacle avoidance quality coefficient, and optimizes the obstacle trajectory prediction model if the obstacle avoidance quality coefficient is lower than the threshold.

[0020] Technical effects and advantages of the present invention: (1) The safe trajectory tracking method for a ground differential drive robot provided by the present invention significantly improves the robot's path planning ability in a complex dynamic environment by introducing a comprehensive approach of multi-source data fusion, dynamic safety distance setting, trajectory prediction uncertainty quantification, and trajectory deviation cost optimization. It also evaluates the behavioral characteristics of dynamic obstacles and the complexity of the environment in real time, dynamically adjusts the safety distance, and accurately quantifies the dynamic obstacle area in combination with the trajectory prediction model, thereby solving the problems of low trajectory tracking accuracy and insufficient obstacle avoidance robustness in the background technology.

[0021] (2) The safe trajectory tracking method for a ground differential drive robot provided by the present invention realizes path optimization through a strategy of minimizing the trajectory deviation cost function, thereby ensuring the safe operation of the robot in a variety of scenarios. At the same time, it greatly improves the efficiency and stability of path planning, and provides a new solution for the robot's autonomous navigation in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the overall structure of the present invention.

[0023] Figure 2 This is a structural block diagram of the robot safety trajectory tracking system of the present invention. DETAILED DESCRIPTION

[0024] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0025] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0026] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present application, its application, or uses.

[0027] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0028] Example 1, see Figure 1 A flow chart of a method for safely tracking a trajectory of a ground differential drive robot is provided. The embodiment of the present invention provides the following Figure 1 A method for safely tracking a ground differential drive robot is shown, comprising the following steps: Step S01, data collection and preprocessing: collect the position information, speed, acceleration and steering angle data of dynamic obstacles through sensors (including but not limited to laser radar, camera, IMU), align multi-source data based on timestamps, and use Kalman filtering to smooth dynamic data; output the preprocessed obstacle motion trajectory history sequence; Step S02, obstacle trajectory prediction and uncertainty quantification: input the preprocessed obstacle motion trajectory history sequence into the obstacle trajectory prediction model, output the movement trajectory and the first probability distribution map in the short future time; combine with the Bayesian algorithm to output the predicted dynamic obstacle area range and the second probability distribution map; Explanation: Obtain the preprocessed obstacle motion trajectory history sequence, obtain the dynamic characteristics of the obstacle motion trajectory, and construct the feature matrix F; input the feature matrix F into the obstacle trajectory prediction model, capture the long-term and short-term dependencies through the time gating mechanism, and focus on the global relationship between trajectory features; output the mobile prediction trajectory and the first probability distribution map in the short future time (such as 1-5 seconds); combine the Bayesian inference method to quantify the uncertainty of the mobile prediction trajectory; construct the uncertainty range of the trajectory prediction, and generate the second probability distribution map of the dynamic obstacle area; the first probability distribution map reflects the confidence of the obstacle trajectory prediction model in predicting the future obstacle position; the high probability area indicates that the obstacle trajectory prediction model is more certain about the obstacle position, while the low probability area indicates that the prediction uncertainty is large; the second probability distribution map shows the uncertainty of the dynamic obstacle area range, reflecting the confidence of the obstacle trajectory prediction model in predicting the spatial area occupied by the future obstacle; the first probability distribution map focuses on the uncertainty of the predicted mobile trajectory, while the second probability distribution map focuses on the uncertainty of the dynamic obstacle area range; Step S03, robot trajectory planning and dynamic adjustment: according to the type of obstacle and motion characteristics, define a dynamic safety distance for each obstacle; determine the safety distance constraint area based on the dynamic safety distance, and record the overlapping area of ​​the safety distance constraint area and the target trajectory as the risk area; if there is a risk area, update the target trajectory of the robot in real time, output the corrected trajectory, ensure that the robot avoids the risk area, and minimize the trajectory deviation cost function that deviates from the target; Explanation: According to the type of obstacle and the motion characteristics, a dynamic safety distance is defined for each obstacle. The safety distance constraint area is obtained based on the dynamic obstacle area and the dynamic safety distance. If the robot does not enter the safety distance constraint area, it indicates safety. The dynamic safety distance is positively correlated with the speed and acceleration characteristics of the obstacle. In a possible embodiment, the environmental sensor data is classified through a semantic segmentation model to identify the types of obstacles and behavior patterns (such as people, animals, and vehicles), set the priority and safety distance of each type of obstacle, and generate a semantically enhanced dynamic obstacle dataset.

[0029] It is necessary to further explain in the embodiment of the present invention that the goal of the trajectory deviation cost function is to minimize the deviation cost and optimize the motion performance while ensuring that the robot avoids the risk area; the trajectory deviation cost function satisfies the following calculation model:

[0030] in, Refers to the cumulative rate of change of acceleration after executing the corrected trajectory; It refers to the energy consumed to execute the corrected trajectory; It refers to the cumulative distance that the trajectory deviates from the target path after correction; α, β, γ are weight coefficients, and ; It is necessary to further explain in the embodiment of the present invention that the environmental state parameters during the operation of the robot are collected, the environmental complexity index is calculated based on the difference between the environmental state parameters and the preset value, and the dynamic safety distance is adjusted based on the environmental complexity index; It needs to be further explained in the embodiment of the present invention that the dynamic safety distance is obtained in the following manner: Collect environmental state parameters, which include at least light intensity, ground reflectivity, sensor noise level, obstacle speed v and acceleration a, and calculate the environmental complexity index ECI through the following model:

[0031] in, Indicates the light intensity deviation, represents the ground reflectivity deviation, represents the sensor noise level, represents the dynamic characteristics of the obstacle, Indicates the rate of environmental change, using the cumulative change in environmental status It is expressed as a ratio of the time span; , , , They represent the reference values ​​of illumination intensity, ground reflectivity deviation, sensor noise level, and obstacle dynamic characteristics respectively; , , , , Respectively represent the weights of each item, and c represents the basic complexity constant; The safety distance is dynamically adjusted based on the environment complexity index ECI; the dynamic safety distance Dsafe satisfies the following formula:

[0032] in, represents the basic safety distance, which is usually the minimum safety distance of the robot in a static environment and is set based on the obstacle type; k is the environment complexity weight, It is the dynamic obstacle motion characteristic adjustment item, which is positively correlated with the obstacle speed and acceleration, and represents the impact of dynamic motion on the safety distance; In one possible embodiment, , is the influence of speed on safety distance, is the influence of acceleration on the safety distance; Step S04, real-time execution and feedback optimization: According to the corrected trajectory, the robot motion instructions are adjusted and executed in real time, the obstacle avoidance effect is evaluated, and the obstacle trajectory prediction model is continuously optimized according to the obstacle avoidance effect and the newly collected data to enhance the adaptability of the obstacle trajectory prediction model and form a dynamic closed-loop control system.

[0033] The explanation states that the feedback loop refers to comparing the real-time data monitored by the sensor with the robot's current motion state to evaluate the obstacle avoidance effect; if the sensor data indicates that the robot is approaching an obstacle or is about to collide, the feedback loop will immediately trigger the obstacle avoidance mechanism, such as deceleration, turning or stopping; the feedback loop records key data in the robot's obstacle avoidance process, such as the time point of success / failure of obstacle avoidance, the speed and acceleration during obstacle avoidance.

[0034] It is necessary to further explain in the embodiment of the present invention that the obstacle avoidance effect is quantified by using the obstacle avoidance effect quality coefficient. The way to obtain is:

[0035]

[0036] Among them, N is the number of obstacle avoidance effect evaluation indicators, is the weight of the i-th obstacle avoidance effect evaluation index, satisfying , represents the actual value of the i-th obstacle avoidance effect evaluation index, Represents the normalized nonlinear mapping function; in the normalized nonlinear mapping function represents the ideal value of the i-th obstacle avoidance effect evaluation index; Used to control the normalized sensitivity of the i-th obstacle avoidance effect evaluation index; What needs to be further explained in the embodiments of the present invention is that the positive indicator refers to an indicator whose larger value is, the better the obstacle avoidance effect is, and the positive indicator at least includes the obstacle avoidance success rate and the dynamic reaction efficiency; the negative indicator refers to an indicator whose smaller value is, the better the obstacle avoidance effect is, and the negative indicator at least includes the obstacle avoidance stability quantitative index, the distance redundancy quantitative index, the obstacle avoidance path deviation quantitative index, and the average obstacle avoidance time; obstacle avoidance stability quantitative index, distance redundancy quantitative index, obstacle avoidance path deviation quantitative index, and the average obstacle avoidance time.

[0037] In the embodiment of the present invention, the obstacle avoidance stability quantification index, the distance redundancy quantification index, and the obstacle avoidance path deviation quantification index are respectively the obstacle avoidance stability index, the distance redundancy index, and the obstacle avoidance path deviation index in the embodiment of the present invention; It is necessary to further explain in the embodiment of the present invention that the obstacle avoidance stability index measures the stability of the robot's trajectory during the obstacle avoidance process and is calculated based on the rate of change of trajectory curvature; the obstacle avoidance stability index Satisfies the following formula:

[0038] in, is the trajectory curvature, T is the total obstacle avoidance time, Indicates the starting point of time. Indicates the end of time, is the second-order derivative of the trajectory curvature; It is necessary to further explain in the embodiment of the present invention that the distance redundancy index is used to measure whether the distance between the robot and the obstacle during the obstacle avoidance process is too large; the distance redundancy index Satisfies the following formula:

[0039] in, is the real-time distance between the robot and the obstacle, For a safe distance, Indicates the remaining part beyond the safety distance; The obstacle avoidance path deviation index is used to measure the degree of deviation of the robot's obstacle avoidance path from the planned path. In the embodiment of the present invention, it is necessary to further explain that the obstacle avoidance path deviation index Satisfies the following formula:

[0040] in, is the actual trajectory during obstacle avoidance. For the ideal trajectory of planning, is the total length of the planned trajectory.

[0041] The obstacle avoidance time refers to the time difference between detecting an obstacle and starting to avoid it; It is necessary to further explain in the embodiment of the present invention that if the obstacle avoidance effect quality index is not lower than the threshold, it indicates that the obstacle avoidance effect is normal and no measures are taken; if the obstacle avoidance effect quality index is lower than the threshold, it indicates that the obstacle avoidance effect is abnormal and the obstacle trajectory prediction model is optimized. The obstacle trajectory prediction model is optimized in the following manner: Perform data augmentation on existing datasets to generate more diverse training samples; for example: Add transformations such as rotation, translation, and scaling to obstacle trajectories; Simulate more complex scenarios (such as high-speed moving obstacles, nonlinear trajectories, etc.); Add extreme scenarios (such as sudden acceleration of obstacles and crowded areas); use a multi-task learning framework to jointly train obstacle trajectory prediction with other related tasks (such as obstacle classification and environment semantic segmentation), share feature layers, and enhance the global understanding ability of the obstacle trajectory prediction model; Modify the model's loss function to dynamically adjust the weights of prediction error and uncertainty quantification; During the operation of the robot, the actual trajectory and obstacle behavior of obstacle avoidance failures are collected in real time, and training data is added to improve the obstacle trajectory prediction model's ability to handle abnormal scenarios through online updates; It needs to be further explained in the embodiment of the present invention that the construction of the obstacle trajectory prediction model includes: Step S11, data set division and preprocessing: obtain the preprocessed obstacle motion trajectory history sequence, divide it into a training set and a test set, extract dynamic features (such as speed, acceleration, steering angle, acceleration change rate, angular velocity change rate), and normalize the feature values; (using sliding window technology) generate a time series input window for training and evaluation of the obstacle trajectory prediction model; Step S12, model design and construction: Design a deep learning model (such as LSTM or Transformer), use time series modeling and multi-layer networks to extract temporal dependencies, clarify the model architecture, including the type of network layers used, the number of layers, the configuration of each layer (such as the number of neurons, activation function, etc.), set the output layer, and predict the points and uncertainty distribution of future trajectories; Step S13, model training and optimization: define a joint loss function, combine trajectory error (such as MSE or MAE) and uncertainty quantification (expressed as entropy or variance of the predicted distribution), use the Adam optimizer to adjust model parameters, optimize hyperparameters (such as learning rate and batch size) through the validation set, record the training process, monitor the loss function and performance, and obtain the trained deep learning model weights; Step S14, model evaluation and testing: Use the test set to evaluate the trajectory prediction accuracy (such as MAE or EDE), uncertainty quantification effect and generalization ability of the model; Step S15: Export the trained obstacle trajectory prediction model into a deployable format to implement real-time prediction function and support online update and dynamic optimization.

[0042] In a possible embodiment, the method includes a path planning dynamic adjustment step based on a trajectory deviation cost function, including: Divide the robot's operating environment into several areas according to the grid; Collect real-time data to obtain the real-time environmental complexity index of each area; Based on the starting and ending points of the task and the real-time environmental complexity index of each area, a path planning algorithm is used to generate several candidate paths. The candidate paths meet the priority rules, which include: giving priority to avoiding areas with high environmental complexity index; giving priority to avoiding areas that directly threaten the safety of robot operation; Generate a trajectory deviation cost function for each candidate path; divide the candidate path into several sub-segments according to equal distances; perform weighted summation of the trajectory deviation cost and weight of the sub-segments to obtain the path cost index of each candidate path; Explanation: By assigning a larger weight to the sub-path close to the robot's position, the impact of future segments on the trajectory deviation cost can be reduced; The candidate path with the lowest path cost index is selected as the robot execution instruction.

[0043] It needs to be further explained in the embodiment of the present invention that the path cost index CL_j of the j-th candidate path is obtained as follows: Assume that there are Q sub-segments, and use p to represent the sub-segment number. The weight coefficient of the sub-segment is set according to the time when the robot arrives at the sub-path. The weight coefficient of the p-th sub-segment of the j-th candidate path is:

[0044] in, represents the time when the robot is expected to arrive at the pth sub-segment of the jth candidate path, Used to avoid zero denominator; The trajectory deviation cost function value of the p-th sub-segment of the j-th candidate path is recorded as ; Through the following formula

[0045] Calculate and obtain CL_j.

[0046] Example 2, see Figure 2 The robot safety trajectory tracking system structure block diagram, the embodiment of the present invention provides a ground differential drive robot safety trajectory tracking system, including: The data acquisition and preprocessing module collects dynamic obstacle information in the robot's operating environment through multiple sensors, including at least position information, speed, acceleration and steering angle data. After the data is timestamp aligned and processed by Kalman filtering, a historical trajectory sequence is generated and the preprocessed obstacle motion trajectory history sequence is output; The safety distance adjustment module obtains the environmental state parameters during the operation of the robot, calculates the environmental complexity index based on the difference between the environmental state parameters and the preset values, and adjusts the dynamic safety distance based on the environmental complexity index; The trajectory prediction and quantification module uses the obstacle trajectory history sequence to predict the obstacle trajectory in the short term in the future through the obstacle trajectory prediction model, and generates a first probability distribution map; combined with the Bayesian inference method, the uncertainty of the trajectory prediction is quantified, and the dynamic obstacle area range and the second probability distribution map are output; the second probability distribution map is used to generate a safe distance constraint area and identify risk areas; In a possible embodiment, the second probability distribution map is used for obstacle avoidance effect evaluation and model optimization; The trajectory planning and dynamic adjustment module defines the dynamic safety distance and generates the safety distance constraint area. The overlapping area between the safety distance constraint area and the target trajectory is recorded as the risk area. If there is a risk area, the robot's target trajectory is updated in real time, and the corrected trajectory is output, while minimizing the trajectory deviation cost function that deviates from the target. The real-time execution and feedback module generates and executes motion instructions in real time according to the corrected trajectory, monitors the obstacle avoidance effect through sensors, calculates the obstacle avoidance quality coefficient, and optimizes the obstacle trajectory prediction model if the obstacle avoidance quality coefficient is lower than the threshold.

[0047] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A safe trajectory tracking method for a ground differential drive robot, characterized in that: The following steps are involved: Step S01, data collection and preprocessing: collect the position information, speed, acceleration and steering angle data of the dynamic obstacle through the sensor, and output the preprocessed obstacle motion trajectory history sequence; Step S02, obstacle trajectory prediction and uncertainty quantification: input the preprocessed obstacle motion trajectory history sequence into the obstacle trajectory prediction model, and output the movement trajectory in the short future time and the first probability distribution map; Combined with the Bayesian algorithm, the predicted dynamic obstacle area range and the second probability distribution map are output; The second probability distribution map is used to generate a safe distance constraint area and identify risk areas; Step S03, robot trajectory planning and dynamic adjustment: according to the type of obstacle and motion characteristics, a dynamic safety distance is defined for each obstacle, a safety distance constraint area is determined based on the dynamic safety distance, and the overlapping area between the safety distance constraint area and the target trajectory is recorded as a risk area; If there is a risk area, the robot's target trajectory is updated in real time, and the corrected trajectory is output, while minimizing the trajectory deviation cost function that deviates from the target; Step S04, real-time execution and feedback optimization: according to the corrected trajectory, the robot motion instructions are adjusted and executed in real time, the obstacle avoidance effect is evaluated, and the obstacle trajectory prediction model is optimized according to the obstacle avoidance effect.

2. The method for safe trajectory tracking of a ground differential drive robot according to claim 1, characterized in that: Obtain the preprocessed obstacle motion trajectory history sequence, obtain the dynamic characteristics of the obstacle motion trajectory, and construct the feature matrix F; The feature matrix F is input into the obstacle trajectory prediction model, and the long-term and short-term dependencies are captured through the time gating mechanism. The mobile prediction trajectory and the first probability distribution map in the short future time are output; combined with the Bayesian inference method, the uncertainty of the mobile prediction trajectory is quantified.

3. The method for safe trajectory tracking of a ground differential drive robot according to claim 1, characterized in that: The safety distance constraint area is obtained based on the dynamic obstacle area and the dynamic safety distance. If the robot does not enter the safety distance constraint area, it indicates safety. The dynamic safety distance is positively correlated with the speed and acceleration characteristics of the obstacle.

4. The method for safe trajectory tracking of a ground differential drive robot according to claim 1, characterized in that: The trajectory deviation cost function satisfies the following calculation model: :in, Refers to the cumulative rate of change of acceleration after executing the corrected trajectory; It refers to the energy consumed to execute the corrected trajectory; It refers to the cumulative distance that the trajectory deviates from the target path after correction, α, β, γ are the weight coefficients of each item, and .

5. The method for safe trajectory tracking of a ground differential drive robot according to claim 3, characterized in that: The setting method of the dynamic safety distance includes: collecting environmental state parameters during the operation of the robot, the environmental state parameters at least including light intensity, ground reflectivity, sensor noise level, obstacle speed v and acceleration a, calculating an environmental complexity index based on the difference between the environmental state parameters and preset values, and adjusting the dynamic safety distance based on the environmental complexity index; Collect environmental state parameters and calculate the environmental complexity index ECI through the following model: :in, Indicates the light intensity deviation, represents the ground reflectivity deviation, represents the sensor noise level, represents the dynamic characteristics of the obstacle, Indicates the rate of environmental change, using the cumulative change in environmental status It is expressed as a ratio of the time span; , , , They represent the reference values ​​of illumination intensity, ground reflectivity deviation, sensor noise level, and obstacle dynamic characteristics respectively; , , , , Respectively represent the weights of each item, c represents the basic complexity constant; the safety distance is dynamically adjusted based on the environmental complexity index ECI; the dynamic safety distance Satisfies the following formula: :in, represents the basic safety distance, which is usually the minimum safety distance of the robot in a static environment and is set based on the obstacle type; k is the environment complexity weight, It is the dynamic obstacle motion characteristic adjustment item, which is positively correlated with the obstacle speed and acceleration, and represents the impact of dynamic motion on the safety distance.

6. The method for safe trajectory tracking of a ground differential drive robot according to claim 1, characterized in that: The obstacle avoidance effect quality coefficient is used to quantify the obstacle avoidance effect. If the obstacle avoidance effect quality index is lower than the threshold, it indicates that the obstacle avoidance effect is abnormal, and the obstacle trajectory prediction model is optimized; the obstacle avoidance effect evaluation index is obtained, and the obstacle avoidance effect evaluation index is divided into positive index and negative index; the obstacle avoidance effect quality coefficient The way to obtain is: : Where N is the number of obstacle avoidance effect evaluation indicators, is the weight of the i-th obstacle avoidance effect evaluation index, satisfying , represents the actual value of the i-th obstacle avoidance effect evaluation index, Represents the normalized nonlinear mapping function; in the normalized nonlinear mapping function represents the ideal value of the i-th obstacle avoidance effect evaluation index; Used to control the normalized sensitivity of the i-th obstacle avoidance effect evaluation index.

7. The method for safe trajectory tracking of a ground differential drive robot according to claim 6, characterized in that: The optimization method of the obstacle trajectory prediction model is: Perform data augmentation on existing datasets to generate more diverse training samples; Use a multi-task learning framework to share the feature layer of the obstacle trajectory prediction model and enhance the global understanding ability of the obstacle trajectory prediction model; During the operation of the robot, the actual trajectory and obstacle behavior of the failed obstacle avoidance are collected in real time, and the training data is added to improve the processing capability of the obstacle trajectory prediction model through online updates.

8. The method for safe trajectory tracking of a ground differential drive robot according to claim 1, characterized in that: The construction of the obstacle trajectory prediction model includes: Step S11, data set division and preprocessing: obtain the preprocessed obstacle motion trajectory history sequence, divide it into a training set and a test set, extract dynamic features, and normalize the feature values; generate a time series input window; Step S12, model design and construction: design a deep learning model, use time series modeling and multi-layer networks to extract temporal dependencies, set the output layer, and predict points and uncertainty distribution with future trajectories; Step S13, model training and optimization: define a joint loss function, combine trajectory error and uncertainty quantification, use the Adam optimizer to adjust model parameters, optimize hyperparameters through the validation set, record the training process, monitor the loss function and performance, and obtain the trained deep learning model weights; Step S14, model evaluation and testing: using the test set to evaluate the trajectory prediction accuracy, uncertainty quantification effect and generalization ability of the model; Step S15: Export the trained obstacle trajectory prediction model into a deployable format.

9. The method for safe trajectory tracking of a ground differential drive robot according to claim 1, characterized in that: The method includes a path planning dynamic adjustment step based on a trajectory deviation cost function, including: Divide the robot's operating environment into several areas according to the grid; Collect real-time data to obtain the real-time environmental complexity index of each area; Based on the starting and ending points of the task and the real-time environmental complexity index of each area, a path planning algorithm is used to generate several candidate paths. The candidate paths meet the priority rules, which include: giving priority to avoiding areas with high environmental complexity index; giving priority to avoiding areas that directly threaten the safety of robot operation; Generate a trajectory deviation cost function for each candidate path; divide the candidate path into several sub-segments according to equal distances; perform weighted summation of the trajectory deviation cost and weight of the sub-segments to obtain the path cost index of each candidate path; The candidate path with the lowest path cost index is selected as the robot execution instruction.

10. A ground differential drive robot safety trajectory tracking system, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The data acquisition and preprocessing module collects dynamic obstacle information in the robot's operating environment through multiple sensors, including at least position information, speed, acceleration and steering angle data. After the data is timestamp aligned and processed by Kalman filtering, a historical trajectory sequence is generated and the preprocessed obstacle motion trajectory history sequence is output; The safety distance adjustment module obtains the environmental state parameters during the operation of the robot, calculates the environmental complexity index based on the difference between the environmental state parameters and the preset values, and adjusts the dynamic safety distance based on the environmental complexity index; The trajectory prediction and quantification module uses the obstacle trajectory history sequence to predict the obstacle trajectory in the short term in the future through the obstacle trajectory prediction model, and generates a first probability distribution map; combined with the Bayesian inference method, it quantifies the uncertainty of trajectory prediction and outputs the dynamic obstacle area range and the second probability distribution map; The trajectory planning and dynamic adjustment module defines the dynamic safety distance and generates the safety distance constraint area. The overlapping area between the safety distance constraint area and the target trajectory is recorded as the risk area. If there is a risk area, the robot's target trajectory is updated in real time, and the corrected trajectory is output, while minimizing the trajectory deviation cost function that deviates from the target. The real-time execution and feedback module generates and executes motion instructions in real time according to the corrected trajectory, monitors the obstacle avoidance effect through sensors, calculates the obstacle avoidance quality coefficient, and optimizes the obstacle trajectory prediction model if the obstacle avoidance quality coefficient is lower than the threshold.

Images