Dynamic obstacle avoidance device for indoor mobile robots
By combining a time-series similarity search model and a hash table dictionary with an obstacle avoidance strategy classification model using LSTM, Transformer, and Softmax layers, the problem of insufficient accuracy and generalization ability of dynamic obstacle prediction and avoidance methods is solved, enabling efficient dynamic obstacle avoidance for mobile robots in indoor environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- FUDAN UNIVERSITY
- Filing Date
- 2023-08-25
- Publication Date
- 2026-07-07
Smart Images
Figure CN116909291B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of deep learning-based mobile robot navigation, and specifically relates to a dynamic obstacle avoidance device for an indoor mobile robot. Background Technology
[0002] Autonomous navigation of mobile robots in indoor environments has always been an important research topic in the fields of automation and robotics. In many applications, such as autonomous cleaning, logistics transportation, and personal assistance, mobile robots need to effectively avoid various dynamic obstacles, such as people, other robots, or moving objects, in complex indoor environments. This requires the robot to be able to predict the future trajectory of these obstacles and make decisions accordingly to achieve safe and efficient navigation.
[0003] Predicting the trajectory of dynamic obstacles is a significant challenge in mobile robot navigation. The process involves first predicting the trajectory of dynamic obstacles and adding it to a map, then using a local path planning algorithm for obstacle avoidance.
[0004] Numerous studies have presented various solutions to the problem of predicting the trajectory of dynamic obstacles.
[0005] Motion model-based methods, such as Kalman filtering and particle filtering, primarily rely on statistical learning and probabilistic models to predict the motion state and future trajectory of dynamic obstacles. For example, Kalman filtering is widely used in target tracking and state estimation tasks. Particle filtering is also a highly effective method; it generates multiple possible future states and then updates and corrects these states based on actual observations.
[0006] Deep learning-based methods, such as Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTMs), and multimodal methods, are used for prediction. These methods are typically data-driven, learning and predicting future trajectories directly from large amounts of motion trajectory data. For example, RNNs and LSTMs perform well in processing time-series data and predicting future states. Multimodal methods, on the other hand, are designed to handle the uncertainty and diversity of real-world environments; they attempt to predict multiple possible future trajectories and provide corresponding probability distributions.
[0007] Social force model-based approach: This approach mainly takes into account the social behavior and interactions of dynamic obstacles (such as pedestrians or other robots) and predicts future motion trajectories by simulating these social forces.
[0008] Reinforcement learning-based approach: This is a novel method that allows robots to autonomously learn how to avoid collisions and navigate safely by interacting with the environment in a simulated setting.
[0009] To integrate predicted motion trajectories into a map and make obstacle avoidance decisions, researchers typically use various local path planning algorithms. Examples include artificial potential field methods for dynamic environments and deep learning-based obstacle avoidance strategy classification methods.
[0010] Despite significant progress in some aspects of existing methods for predicting dynamic obstacle trajectories and making obstacle avoidance decisions, they still have some key shortcomings: motion model-based methods, such as Kalman filters and particle filters, may fail in nonlinear or abrupt motion; deep learning-based methods, while capable of predicting trajectories from large datasets, face challenges in generalization when encountering situations where the data differs greatly from the training data in real-world environments; the prediction uncertainty of multimodal methods may affect the precise control of mobile robots; social force models are also challenging to model in complex environments; reinforcement learning-based methods require substantial training time and resources, and the training process may be unstable; obstacle avoidance decision-making methods, such as artificial potential field methods, may get trapped in local optima.
[0011] In summary, existing obstacle avoidance methods for dynamic obstacles still suffer from low prediction accuracy and insufficient generalization ability. Summary of the Invention
[0012] This invention was made to solve the above-mentioned problems, and its purpose is to provide a dynamic obstacle avoidance device for indoor mobile robots.
[0013] This invention provides a dynamic obstacle avoidance device for an indoor mobile robot, mounted on the robot for controlling the robot to dynamically avoid obstacles based on dynamic obstacles. It features the following components: a data acquisition module for acquiring the motion trajectories of the dynamic obstacle and the mobile robot; a motion trajectory prediction module, including a time-series similarity search model, which predicts the motion trajectory of the dynamic obstacle based on a pre-stored hash table dictionary and multiple historical motion trajectories; an obstacle avoidance strategy generation module, including an obstacle avoidance strategy classification model, which generates an obstacle avoidance strategy from the predicted motion trajectory and the mobile robot's motion trajectory; a path planning module, storing a preset global map, which updates the mobile robot's local map based on the predicted motion trajectory and the global map, and then obtains the planned path based on TEB planning and the local map; and an obstacle avoidance execution module, which executes the path according to the rules. A path-planning control mobile robot performs dynamic obstacle avoidance. The motion trajectory prediction module includes at least a data storage unit and a first preprocessing unit. The data storage unit stores a hash table dictionary and historical motion trajectory data. The first preprocessing unit preprocesses the motion trajectory of the dynamic obstacle to obtain preprocessed data. A time series similarity search model processes the preprocessed data to obtain the predicted motion trajectory, specifically including the following steps: Step S1, extract features from the preprocessed data to obtain feature vectors; Step S2, perform binary hash encoding on the feature vectors to obtain feature binary codes; Step S3, retrieve the key-value pair corresponding to the binary code most similar to the feature binary code from the hash table dictionary; Step S4, perform ID retrieval on all historical motion trajectories based on the value of the key-value pair to obtain the corresponding historical motion trajectory; Step S5, extract the next time window of the time window corresponding to the key-value pair from the historical motion trajectories as the predicted motion trajectory.
[0014] The dynamic obstacle avoidance device for indoor mobile robots provided by this invention may also have the following features: during the training process of the time series similarity search model, a hash loss function is calculated based on the preprocessed data corresponding to the feature binary code and the corresponding reconstructed motion trajectory, and the parameters of the time series similarity search model are optimized based on the calculation results. The reconstructed motion trajectory is obtained by decoding the feature binary code.
[0015] The dynamic obstacle avoidance device for indoor mobile robots provided by the present invention may also have the following features: the obstacle avoidance strategy generation module includes at least a second preprocessing unit and an obstacle avoidance strategy generation unit. The second preprocessing unit is used to preprocess the predicted motion trajectory and the motion trajectory of the mobile robot respectively, and obtain the first preprocessed data and the second preprocessed data as inputs to the obstacle avoidance strategy classification model. The obstacle avoidance strategy generation unit is used to select the obstacle avoidance strategy category with the highest final probability from the final probabilities of each obstacle avoidance strategy category output by the obstacle avoidance strategy classification model as the obstacle avoidance strategy.
[0016] The dynamic obstacle avoidance device for indoor mobile robots provided by the present invention may also have the following features: wherein the preprocessing of the second preprocessing unit includes: min-max normalization processing operation and filling missing data using the time series difference method.
[0017] The dynamic obstacle avoidance device for indoor mobile robots provided by this invention may also have the following features: the obstacle avoidance strategy classification model includes an LSTM layer, a Transformer layer, a fully connected layer, and a Softmax layer. The LSTM layer is used to extract features from the first preprocessed data and the second preprocessed data to obtain obstacle avoidance features. The Transformer layer is used to process the obstacle avoidance features to obtain the obstacle avoidance strategy feature vector at each time point. The fully connected layer is used to obtain the probability of each obstacle avoidance strategy category at each time point based on the obstacle avoidance strategy feature vector. The Softmax layer is used to obtain the final probability of each obstacle avoidance strategy category based on the probabilities at all time points.
[0018] The dynamic obstacle avoidance device for indoor mobile robots provided by the present invention may also have the following feature: the obstacle avoidance strategy categories include obstacle avoidance, stopping, and retreating.
[0019] The dynamic obstacle avoidance device for indoor mobile robots provided by this invention may also have the following features: the process of obtaining the planned path in the path planning module is as follows: when the obstacle avoidance strategy is obstacle avoidance, the local map is continuously updated, and the planned path is obtained on the local map according to TEB planning; when the obstacle avoidance strategy is stop, the planned path is that the mobile robot stops in place; when the obstacle avoidance strategy is back, a large value is assigned to the back weight of TEB planning, and then TEB planning is performed according to the local map to obtain the back path as the planned path.
[0020] The dynamic obstacle avoidance device for indoor mobile robots provided by the present invention may also have the following feature: wherein the path planning module generates a planned path once in each control cycle, and the control cycle is 200 milliseconds.
[0021] The role and effect of invention
[0022] The dynamic obstacle avoidance device for indoor mobile robots according to the present invention achieves rapid and accurate prediction of the subsequent movement trajectory of dynamic obstacles by storing historical movement trajectories and constructing a corresponding hash table dictionary through a motion trajectory prediction module, calculating the time window of the historical movement trajectory most similar to the movement trajectory of the dynamic obstacle from the hash table dictionary through a time series similarity search model, and then using the movement trajectory corresponding to the next time window in the historical movement trajectory as the predicted movement trajectory. Therefore, the dynamic obstacle avoidance device for indoor mobile robots of the present invention enables mobile robots to achieve more accurate and efficient dynamic obstacle avoidance in indoor environments. Attached Figure Description
[0023] Figure 1 This is a schematic diagram of the frame of the dynamic obstacle avoidance device in an embodiment of the present invention;
[0024] Figure 2 This is a schematic diagram of the motion trajectory prediction module in an embodiment of the present invention;
[0025] Figure 3 This is a schematic diagram of the data processing flow of the time series similarity search model in an embodiment of the present invention;
[0026] Figure 4 This is a schematic diagram of the process of constructing a hash table dictionary in an embodiment of the present invention;
[0027] Figure 5 This is a schematic diagram of the obstacle avoidance strategy generation module in an embodiment of the present invention. Detailed Implementation
[0028] To make the technical means, creative features, objectives and effects of the present invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the dynamic obstacle avoidance device for the indoor mobile robot of the present invention.
[0029] In this embodiment, the dynamic obstacle avoidance device of the present invention is installed on a mobile robot and is used to control the mobile robot to perform dynamic obstacle avoidance based on dynamic obstacles.
[0030] Figure 1 This is a schematic diagram of the framework of the dynamic obstacle avoidance device in an embodiment of the present invention.
[0031] like Figure 1 As shown, the dynamic obstacle avoidance device 100 includes a data acquisition module 10, a motion trajectory prediction module 20, an obstacle avoidance strategy generation module 30, a path planning module 40, and an obstacle avoidance execution module 50.
[0032] The data acquisition module 10 is used to collect the motion trajectories of dynamic obstacles and mobile robots, respectively.
[0033] Figure 2This is a schematic diagram of the motion trajectory prediction module in an embodiment of the present invention.
[0034] like Figure 2 As shown, the motion trajectory prediction module 20 includes a data storage unit 201, a time series similarity search model storage unit 202, a first preprocessing unit 203, a prediction generation unit 204, and a motion trajectory prediction control unit 205 that controls the above units. The motion trajectory prediction module 20 predicts the motion trajectory of the dynamic obstacle based on a pre-stored hash table dictionary and multiple historical motion trajectories, using a time series similarity search model.
[0035] Data storage unit 201 is used to store hash table dictionary and historical motion trajectory data.
[0036] In this embodiment, the historical motion trajectory is obtained in the following way:
[0037] By manually controlling a simulated mobile robot's perspective and multiple dynamic vehicles acting as dynamic obstacles, the movement of a simulated mobile robot's vehicle within an indoor scene is achieved. The coordinates and time of each vehicle in the same global coordinate system are collected as its total trajectory. For example, the total trajectory of vehicle B is {[x1,y1,time1,IDB],[x2,y2,time2,IDB],...}, where IDB is the ID of the vehicle, x1 and y1 are the coordinates of vehicle B at time1, and x2 and y2 are the coordinates of vehicle B at timeti. The coordinates at me2 are used to divide the total motion trajectory of the dynamic car into multiple 1*n-dimensional matrices according to the length of the time window as the historical motion trajectory. For example, the historical motion trajectory of car B in the current time window is [Bx1,By1,Bt1,...,Bxn,Byn,Btn], where n is the length of the time window, Bx1 and By1 are the position coordinates of car B at Bt1 in the time window, and Bxn and Byn are the position coordinates of car B at Btn in the time window. In this embodiment, the time window length is 3 seconds, that is, n=3.
[0038] The time series similarity search model storage unit 202 stores the trained time series similarity search model.
[0039] The first preprocessing unit 203 is used to preprocess the motion trajectory of the dynamic obstacle to obtain preprocessed data. In this embodiment, the preprocessing operation includes normalization, and missing data is replaced with 0.
[0040] The prediction generation unit 204 is used to input preprocessed data into the time series similarity search model and combine it with a hash table dictionary and historical motion trajectories to obtain the predicted motion trajectory of dynamic obstacles.
[0041] Figure 3 This is a schematic diagram of the data processing flow of the time series similarity search model in an embodiment of the present invention.
[0042] like Figure 3 As shown, the time series similarity search model processes preprocessed data to obtain predicted motion trajectories, specifically including the following steps:
[0043] Step S1: Extract features from the preprocessed data to obtain feature vectors.
[0044] Step S2: Perform binary hash encoding on the feature vector to obtain the feature binary code.
[0045] Step S3: Retrieve the key-value pair corresponding to the binary code that is closest to the feature binary code in the hash table dictionary. In this embodiment, the binary code that is the same as or closest to the feature binary code is found in the hash dictionary by Hamming distance.
[0046] Figure 4 This is a schematic diagram of the process of constructing a hash table dictionary in an embodiment of the present invention.
[0047] like Figure 4 As shown, the construction process of the hash table dictionary includes the following steps T1 to T4:
[0048] Step T1: Build and initialize the Python dictionary.
[0049] Step T2: Input each preprocessed historical motion trajectory into the time series similarity search model for binary hash encoding to obtain the corresponding binary code.
[0050] Step T3: Use the binary code as the key of the dictionary and the ID of the historical motion trajectory corresponding to the binary code as the value of the dictionary. Then, the key and the value form a key-value pair.
[0051] Step T4: Add the key-value pairs corresponding to all historical motion trajectories to the Python dictionary to obtain the hash table dictionary.
[0052] Step S4: Retrieve the ID of all historical motion trajectories based on the value of the key-value pair to obtain the corresponding historical motion trajectory.
[0053] Step S5: In the historical motion trajectory, extract the next time window of the time window corresponding to the key-value pair as the predicted motion trajectory.
[0054] In this embodiment, the time series similarity search model is trained using existing historical motion trajectories. The hyperparameter settings of the time series similarity search model include: 4 layers in the Transformer, a learning rate of 2e-4, a batch size of 128, self-attention, and k = v = q = the historical motion trajectory under the current time window.
[0055] In this embodiment, the binary hash encoding performance of the time series similarity search model is trained using a hash loss function. The specific process is as follows:
[0056] The preprocessed data of the historical motion trajectory after being processed by the first preprocessing unit 203 is input into the time series similarity search model to obtain the corresponding feature binary code. The feature binary code is decoded to obtain the reconstructed motion trajectory. The hash loss function is calculated based on the preprocessed data and the corresponding reconstructed motion trajectory. The parameters of the time series similarity search model are optimized based on the calculation results, thereby improving the binary hash coding performance of the time series similarity search model.
[0057] The motion trajectory prediction control unit 205 stores a computer program for controlling the various constituent units of the motion trajectory prediction module 20.
[0058] Figure 5 This is a schematic diagram of the obstacle avoidance strategy generation module in an embodiment of the present invention.
[0059] like Figure 5 As shown, the obstacle avoidance strategy generation module 30 includes a second preprocessing unit 301, an obstacle avoidance strategy classification model storage unit 302, an obstacle avoidance probability generation unit 303, an obstacle avoidance strategy generation unit 304, and an obstacle avoidance strategy generation control unit 305, which are used to generate obstacle avoidance strategies by using the obstacle avoidance strategy classification model to generate strategies from the predicted motion trajectory and the motion trajectory of the mobile robot.
[0060] The second preprocessing unit 301 is used to preprocess the predicted motion trajectory and the motion trajectory of the mobile robot respectively, and obtain the first preprocessed data and the second preprocessed data as inputs to the obstacle avoidance strategy classification model.
[0061] The preprocessing in the second preprocessing unit includes: min-max normalization and filling missing data using time series interpolation.
[0062] The obstacle avoidance strategy classification model storage unit 302 stores a trained obstacle avoidance strategy classification model.
[0063] The obstacle avoidance probability generation unit 303 is used to input the first preprocessed data and the second preprocessed data into the obstacle avoidance strategy classification model to obtain the final probability of each obstacle avoidance strategy category.
[0064] The obstacle avoidance strategy classification model includes an LSTM layer, a Transformer layer, a fully connected layer, and a Softmax layer.
[0065] The LSTM layer is used to extract features from the first and second preprocessed data to obtain obstacle avoidance features.
[0066] The Transformer layer is used to process the obstacle avoidance features to obtain the obstacle avoidance strategy feature vector at each time point.
[0067] The fully connected layer is used to obtain the probability of each obstacle avoidance strategy category at each time point based on the obstacle avoidance strategy feature vector. In this embodiment, the number of nodes in the fully connected layer is equal to the total number of obstacle avoidance strategy categories, and self-attention is used, where k = v = q = matrix [first preprocessed data; second preprocessed data], the matrix has a dimension of 2*m, and m represents the length of the current time window.
[0068] The Softmax layer is used to obtain the final probability of each obstacle avoidance strategy category based on the probabilities at all time points.
[0069] In this embodiment, the historical motion trajectory of the existing dynamic car, the corresponding manual obstacle avoidance label, and the motion trajectory of the simulated car are used as training datasets to train the obstacle avoidance strategy classification model. The model is optimized by the cross-entropy loss function to obtain the trained obstacle avoidance strategy classification model. The parameters set include: 4 layers of LSTM, 1024 hidden units, 3 seconds sequence length, 4 layers of Transformer, 2e-4 learning rate, and 128 batch size.
[0070] The obstacle avoidance strategy generation unit 304 is used to select the obstacle avoidance strategy category with the highest final probability from all obstacle avoidance strategy categories as the obstacle avoidance strategy.
[0071] The obstacle avoidance strategies include obstacle avoidance, stopping, and backing up.
[0072] The obstacle avoidance strategy generation and control unit 305 stores a computer program for controlling the various constituent units of the obstacle avoidance strategy generation module 30.
[0073] The path planning module 40 stores a preset global map, which is used to update the local map of the mobile robot based on the predicted motion trajectory and the global map. Then, the planned path is obtained based on the TEB planning and the local map. In this embodiment, the predicted motion trajectory is mapped from the global map to the obstacle layer of the local map of the mobile robot to obtain the updated local map.
[0074] The process of obtaining the planned path in the path planning module 40 is as follows:
[0075] When the obstacle avoidance strategy is to avoid obstacles, the local map is continuously updated, and the planned path is obtained on the local map based on TEB planning.
[0076] When the obstacle avoidance strategy is to stop, the planned path is for the mobile robot to stop in place.
[0077] When the obstacle avoidance strategy is to back off, a large value is assigned to the back off weight of TEB planning, and then TEB planning is performed based on the local map to obtain the back off path as the planned path.
[0078] The path planning module 40 generates a planned path once in each control cycle, which is 200 milliseconds.
[0079] The obstacle avoidance execution module 50 is used to control the mobile robot to perform dynamic obstacle avoidance according to the planned path.
[0080] The role and effect of the embodiments
[0081] According to the dynamic obstacle avoidance device of the indoor mobile robot involved in this embodiment, the historical motion trajectory is stored by the motion trajectory prediction module and a corresponding hash table dictionary is constructed. Then, the time window of the historical motion trajectory that is most similar to the motion trajectory of the dynamic obstacle is calculated from the hash table dictionary by the time series similarity search model. Then, the motion trajectory corresponding to the next time window in the historical motion trajectory is used as the predicted motion trajectory, thereby realizing the rapid and accurate prediction of the subsequent motion trajectory of the dynamic obstacle.
[0082] Based on the preprocessed historical motion trajectories and the corresponding reconstructed motion trajectories, a hash loss function is calculated to optimize the binary hash coding performance of the time series similarity search model, thereby expanding the differences between the binary codes corresponding to different motion trajectories and making the binary codes better reflect the differences between motion trajectories.
[0083] The obstacle avoidance strategy assigns a larger backward weight to the TEB planning when moving backward, making the generated planned path more inclined to the backward path, thereby realizing the obstacle avoidance strategy of moving backward when encountering dynamic obstacles.
[0084] In summary, this method enables mobile robots to achieve more accurate and efficient dynamic obstacle avoidance in indoor environments.
[0085] The above embodiments are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention.
Claims
1. A dynamic obstacle avoidance device for an indoor mobile robot, mounted on the mobile robot, used to control the mobile robot to perform dynamic obstacle avoidance based on dynamic obstacles, characterized in that, include: The data acquisition module is used to collect the motion trajectories of dynamic obstacles and the mobile robot, respectively. The motion trajectory prediction module includes a time series similarity search model. Based on a pre-stored hash table dictionary and multiple historical motion trajectories, the time series similarity search model is used to predict the motion trajectory of the dynamic obstacle to obtain the predicted motion trajectory of the dynamic obstacle. An obstacle avoidance strategy generation module includes an obstacle avoidance strategy classification model, which is used to generate an obstacle avoidance strategy by using the obstacle avoidance strategy classification model to generate a strategy for the predicted motion trajectory and the motion trajectory of the mobile robot. The path planning module stores a preset global map, which is used to update the local map of the mobile robot based on the predicted motion trajectory and the global map, and then obtain the planned path based on TEB planning and the local map. The obstacle avoidance execution module is used to control the mobile robot to perform dynamic obstacle avoidance according to the planned path. The motion trajectory prediction module includes at least a data storage unit and a first preprocessing unit. The data storage unit is used to store the hash table dictionary and the historical motion trajectory data. The first preprocessing unit is used to preprocess the motion trajectory of the dynamic obstacle to obtain preprocessed data. The time series similarity search model processes the preprocessed data to obtain the predicted motion trajectory, specifically including the following steps: Step S1: Extract features from the preprocessed data to obtain feature vectors; Step S2: Perform binary hash encoding on the feature vector to obtain the feature binary code; Step S3: Retrieve the key-value pair corresponding to the binary code that is closest to the feature binary code from the hash table dictionary; Step S4: Perform ID retrieval on all the historical motion trajectories based on the value of the key-value pair to obtain the corresponding historical motion trajectory; Step S5: In the historical motion trajectory, extract the next time window of the time window corresponding to the key-value pair as the predicted motion trajectory.
2. The dynamic obstacle avoidance device for an indoor mobile robot according to claim 1, characterized in that: in, During the training process of the time series similarity search model, a hash loss function is calculated based on the preprocessed data and the corresponding reconstructed motion trajectory, and the parameters of the time series similarity search model are optimized based on the calculation results. The reconstructed motion trajectory is obtained by decoding the feature binary code corresponding to the preprocessed data.
3. The dynamic obstacle avoidance device for an indoor mobile robot according to claim 1, characterized in that: in, The obstacle avoidance strategy generation module includes at least a second preprocessing unit and an obstacle avoidance strategy generation unit. The second preprocessing unit is used to preprocess the predicted motion trajectory and the motion trajectory of the mobile robot, respectively, to obtain first preprocessed data and second preprocessed data, which are then used as inputs to the obstacle avoidance strategy classification model. The obstacle avoidance strategy generation unit is used to select the obstacle avoidance strategy category with the highest final probability from the final probabilities of each obstacle avoidance strategy category output by the obstacle avoidance strategy classification model as the obstacle avoidance strategy.
4. The dynamic obstacle avoidance device for an indoor mobile robot according to claim 3, characterized in that: in, The preprocessing of the second preprocessing unit includes: min-max normalization and filling missing data using time series interpolation.
5. The dynamic obstacle avoidance device for an indoor mobile robot according to claim 3, characterized in that: in, The obstacle avoidance strategy classification model includes an LSTM layer, a Transformer layer, a fully connected layer, and a Softmax layer. The LSTM layer is used to extract features from the first preprocessed data and the second preprocessed data to obtain obstacle avoidance features. The Transformer layer is used to process the obstacle avoidance features to obtain the obstacle avoidance strategy feature vector at each time point. The fully connected layer is used to obtain the probability of each obstacle avoidance strategy category at each time point based on the obstacle avoidance strategy feature vector; The Softmax layer is used to obtain the final probability of each obstacle avoidance strategy category based on the probabilities at all time points.
6. The dynamic obstacle avoidance device for an indoor mobile robot according to claim 1, characterized in that: in, The obstacle avoidance strategy categories include obstacle avoidance, stopping, and backing up.
7. The dynamic obstacle avoidance device for an indoor mobile robot according to claim 6, characterized in that: in, The process of obtaining the planned path in the path planning module is as follows: When the obstacle avoidance strategy involves avoiding obstacles, the local map is continuously updated, and the planned path is obtained on the local map based on the TEB plan. When the obstacle avoidance strategy is to stop, the planned path is for the mobile robot to stop in place. When the obstacle avoidance strategy is to retreat, a large value is assigned to the retreat weight of the TEB planning, and then the TEB planning is performed according to the local map to obtain the retreat path as the planned path.
8. The dynamic obstacle avoidance device for an indoor mobile robot according to claim 1, characterized in that: in, The path planning module generates the planned path once in each control cycle, which is 200 milliseconds.