Navigation method, device and equipment of mobile robot, medium and product
The highest matching trajectory sequence is selected as the robot navigation trajectory through diffusion model and auxiliary constraints, which solves the problem of navigation failure in the existing technology and achieves higher matching and security.
Patent Information
- Application Number
- CN202510621460.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-14
- Publication Date
- 2025-08-15
AI Technical Summary
The existing robot navigation method based on diffusion strategy cannot effectively quantify the degree of matching the trajectory sequence with the current observation state when encountering obstacles, resulting in navigation failure.
The matching degree between multiple candidate trajectory sequences and the current observation state is obtained through the diffusion model, and the trajectory sequence with the highest matching degree is selected as the navigation trajectory. Further screening is combined with auxiliary constraints to ensure that the trajectory sequence meets safety and feasibility.
Improves the matching degree of navigation trajectory with the current observed state, reduces the navigation failure rate, and enhances navigation success rate and security in unknown environments.
Smart Images
Figure CN120489129A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of robot navigation technology, and in particular to a navigation method, device, equipment, medium and product for a mobile robot. Background Art
[0002] In recent years, a series of "diffusion-based visual navigation methods" have emerged in the field of robotics. Specifically, these methods add noise to the action sequence space during the training phase and use the observation state and time step as guiding conditions to obtain the model parameters of the noise estimator through training; in the testing phase, such as Figure 8 As shown in Figure 1, after receiving the current observation state, the noise estimator starts with randomly sampled Gaussian noise and, after multiple iterations (e.g., K rounds), gradually removes the Gaussian noise, ultimately recovering a trajectory sequence that matches the current observation state. Compared to traditional imitation learning methods, the diffusion strategy-based approach has two main advantages: first, it can effectively model the multimodal distribution of trajectory sequences, and second, it is more suitable for predicting high-dimensional trajectory sequences.
[0003] However, the diverse trajectory sequences output by the model lack effective control and quantification mechanisms. For example, when a robot encounters an obstacle, while the diffusion strategy can generate multiple trajectory sequences that match the current observation state (e.g., multiple trajectories that avoid the obstacle by turning left or right), it cannot quantify the degree to which these trajectory sequences match the current observation state. Poor matching can lead to navigation failure. Summary of the Invention
[0004] The present application provides a navigation method, apparatus, device, medium and product for a mobile robot, aiming to solve the problem that navigation failure may occur when the predicted trajectory sequence and the current observation state are poorly matched.
[0005] In a first aspect, the present application provides a navigation method for a mobile robot, comprising: Using the current observation state as a guide condition, a first matching degree between multiple first candidate trajectory sequences and the current observation state is obtained based on a diffusion model. The diffusion model is used to obtain a prediction result corresponding to each first candidate trajectory sequence, and the prediction result is used to calculate the first matching degree. The first candidate trajectory sequence with the highest first matching degree is used as the navigation trajectory of the mobile robot.
[0006] As an embodiment, the navigation method of the mobile robot further includes: Based on the auxiliary constraint corresponding to the current observation state, at least one second candidate trajectory sequence is selected from the plurality of first candidate trajectory sequences, and a second matching degree corresponding to the second candidate trajectory sequence is obtained; wherein the second candidate trajectory sequence meets the auxiliary constraint, and the second matching degree is related to the first matching degree and a matching value of the second candidate trajectory sequence with the auxiliary constraint; The second candidate trajectory sequence with the second highest matching degree is used as the navigation trajectory of the mobile robot.
[0007] As an embodiment, for each first candidate trajectory sequence, obtaining a first matching degree between the first candidate trajectory sequence and the current observation state specifically includes: Adding known noise to each first candidate trajectory sequence to obtain a noisy candidate trajectory sequence; The noisy candidate trajectory sequence, the current observation state, and the time step are input into the diffusion model to obtain the predicted noise corresponding to the first candidate trajectory sequence; A first noise error corresponding to the first candidate trajectory sequence is calculated based on the predicted noise and the corresponding known noise, and the first noise error is used as a first matching degree between the first candidate trajectory sequence and the current observation state.
[0008] As an embodiment, the known noise includes random sampling noise and random sampling time step, and the first noise error is obtained by weighted calculation of the prediction error of the random sampling noise and the random sampling time step.
[0009] As an embodiment, multiple known noises are added to each first candidate trajectory sequence, and a first noise error corresponding to each first candidate trajectory sequence is obtained based on second noise errors of all known noises.
[0010] As an embodiment, the auxiliary constraint condition is obtained by performing data analysis on the current observation state.
[0011] As an embodiment, the plurality of first candidate trajectory sequences are obtained by preprocessing all trajectory sequences in the training dataset, where the preprocessing includes normalization and clustering.
[0012] The present application also provides a navigation device for a mobile robot, comprising a matching module and a navigation trajectory determination module; The matching module is used to obtain a first matching degree between multiple first candidate trajectory sequences and the current observation state based on a diffusion model, using the current observation state as a guide condition. The diffusion model is used to obtain a prediction result corresponding to each first candidate trajectory sequence, and the prediction result is used to calculate the first matching degree. The navigation trajectory determination module is used to use the first candidate trajectory sequence with the highest first matching degree as the navigation trajectory of the mobile robot.
[0013] As an embodiment, the navigation device further includes a screening module, the screening module being configured to screen at least one second candidate trajectory sequence from the plurality of first candidate trajectory sequences based on an auxiliary constraint condition corresponding to a current observation state, and obtain a second matching degree corresponding to the second candidate trajectory sequence; wherein the second candidate trajectory sequence satisfies the auxiliary constraint condition, and the second matching degree is correlated with the first matching degree and a matching value between the second candidate trajectory sequence and the auxiliary constraint condition; The navigation trajectory determination module is further configured to use the second candidate trajectory sequence with the second highest matching degree as the navigation trajectory of the mobile robot.
[0014] As an embodiment, the matching module is further configured to: Adding known noise to the first candidate trajectory sequence to obtain a noisy candidate trajectory sequence; Inputting the noisy candidate trajectory sequence, the current observation state, and the time step into the diffusion model to obtain the predicted noise corresponding to the first candidate trajectory sequence; A first noise error corresponding to the first candidate trajectory sequence is calculated based on the predicted noise and the corresponding known noise, and the first noise error is used as a first matching degree between the first candidate trajectory sequence and the current observation state.
[0015] In a third aspect, the present application also provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, it implements any of the above-mentioned mobile robot navigation methods.
[0016] In a fourth aspect, the present application also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, any of the above-mentioned navigation methods of the mobile robot is implemented.
[0017] In a fifth aspect, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements any of the above-mentioned navigation methods of the mobile robot. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the technical solutions in the present application or the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.
[0019] Figure 1 This is one of the flow charts of the navigation method of the mobile robot provided in this application; Figure 2This is the second flow chart of the mobile robot navigation method provided by this application; Figure 3 This is the effect diagram of the preprocessing of the trajectory sequence provided by this application; Figure 4 This is a flowchart of obtaining a first matching degree provided by this application; Figure 5 This is a graph showing the relationship between the time step and noise prediction accuracy obtained using different training data sets based on the navigation method of the mobile robot provided by the present application; Figure 6 This is the third flow chart of the mobile robot navigation method provided by this application; Figure 7 This is the fourth flow chart of the mobile robot navigation method provided by this application; Figure 8 It is a flow chart of the diffusion model of the prior art; Figure 9 This is an example of the passability of the mobile robot navigation method provided by this application in a single unknown scene; Figure 10 This is an example of the feasibility of the mobile robot navigation method provided by this application in multiple unknown scenarios; Figure 11 This is one of the structural diagrams of the navigation device of the mobile robot provided in this application; Figure 12 This is the second structural diagram of the navigation device of the mobile robot provided by this application; Figure 13 It is a structural diagram of the electronic device provided in this application. DETAILED DESCRIPTION
[0020] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.
[0021] It should be noted that, in the description of this application, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or also includes elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.
[0022] The terms "first," "second," and the like in this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that such terms are interchangeable where appropriate, so that embodiments of this application can be implemented in an order other than that illustrated or described herein. Furthermore, the terms "first," "second," and the like generally distinguish objects of a class and do not limit the number of objects; for example, the first object can be one or more. Furthermore, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates an "or" relationship between the connected objects.
[0023] The following combination Figures 1 to 13 Describe the navigation method, device, equipment, medium and product of the mobile robot provided in this application.
[0024] It should be noted that the navigation method of the mobile robot provided in the embodiment of the present application is implemented based on the navigation device of the mobile robot. The navigation method of the mobile robot takes the current observation state as the guidance condition, and selects the trajectory sequence with the highest matching degree from the representative candidate trajectory sequence as the navigation trajectory of the robot, thereby improving the navigation success rate.
[0025] In the embodiment of the present application, a navigation method of a mobile robot is described by taking a navigation device of a mobile robot as an execution subject as an example.
[0026] Figure 1 This is one of the flow charts of the navigation method of the mobile robot provided in this application. Figure 2 This is the second flow chart of the mobile robot navigation method provided in this application.
[0027] In one possible embodiment, Figure 1 and 2 As shown, the navigation method of the mobile robot provided by this application includes: S110: Taking the current observation state as a guide condition, obtaining first matching degrees between a plurality of first candidate trajectory sequences and the current observation state based on a diffusion model, and obtaining a sorted list of the plurality of first candidate trajectory sequences.
[0028] The multiple first candidate trajectory sequences are representative trajectory sequences selected from the diffusion model's training dataset. It should be noted that the acquisition of the multiple first candidate trajectory sequences is independent of the diffusion model's training and inference phases, and that in step S110, the diffusion model is trained based on the training dataset. The diffusion model is used to obtain a prediction result corresponding to each first candidate trajectory sequence, which is used to calculate the first matching degree.
[0029] The current observation state can be information such as images, point clouds, or local satellite maps acquired in the environment where the mobile robot is located.
[0030] S120: Using the first candidate trajectory sequence with the highest first matching degree as the navigation trajectory of the mobile robot.
[0031] Specifically, the navigation device of the mobile robot uses the current observation state as a guide condition and obtains a first matching degree between multiple first candidate trajectory sequences and the current observation state based on a diffusion model. The navigation device of the mobile robot then uses the first candidate trajectory sequence with the highest first matching degree as the navigation trajectory of the mobile robot.
[0032] Please combine Figure 2 In one embodiment, the diffusion model includes a condition editor and a noise estimator. The condition editor is used to encode the current observation state, and its output data is input into the noise estimator as a guiding condition. The input data of the noise estimator also includes multiple first candidate trajectory sequences and time steps. The output data of the noise estimator is input into a matching degree calculation module to obtain a first matching degree between each first candidate trajectory sequence and the current observation state. The first candidate trajectory sequence with the highest first matching degree is then determined as the navigation trajectory of the mobile robot.
[0033] This embodiment of the present application selects the first candidate trajectory sequence with the highest degree of match to the current observation state from multiple first candidate trajectory sequences and uses it as the navigation trajectory for the mobile robot. Compared to existing navigation trajectory determination schemes that lack candidate trajectory sequences, this significantly improves the degree of match between the navigation trajectory and the current observation state and reduces the navigation failure rate. Furthermore, inputting a representative first candidate trajectory sequence into a diffusion model reduces the model's spatial complexity and inference time, compared to calculating the degree of match between all trajectory sequences in the entire training dataset and the current observation state.
[0034] In step S110 , in a possible embodiment, the plurality of first candidate trajectory sequences are obtained by preprocessing all trajectory sequences in the training dataset, where the preprocessing includes normalization and clustering.
[0035] Specifically, combined Figure 3 , first of all the trajectory sequences in the training dataset (see Figure 3 Left figure) to obtain trajectory sequences of roughly the same length (see Figure 3 Then clustering is performed using a clustering method (such as the standard k-means algorithm, K-medoids algorithm, or Bisecting K-means algorithm) to obtain (e.g. 100) candidate trajectory sequences (see Figure 3 Right), where each candidate trajectory sequence is a cluster center. During the navigation process, given any trajectory sequence a, its corresponding cluster center can be found, i.e., one of the candidate trajectory sequences , The candidate trajectory sequences are marked as , .
[0036] In a possible embodiment, when normalizing all trajectory sequences, the mean lengths of all trajectory sequences are first calculated, and then a relative value of each trajectory sequence is calculated based on the mean length to obtain trajectory sequences with approximately the same length.
[0037] The embodiment of the present application eliminates the difficulties caused by the differences between different trajectory sequences through normalization, then extracts cluster centers from the normalized dataset and uses the representative features of the cluster centers to determine the first candidate trajectory sequence, thereby greatly improving the representative performance of the first candidate trajectory sequence.
[0038] In step S110 , in a possible embodiment, relevant features of the current observation state and each first candidate trajectory sequence are extracted, and the matching degree between the two is determined by calculating the features (eg, calculating the Euclidean distance, etc.).
[0039] In another possible embodiment, Figure 4 As shown, for each first candidate trajectory sequence, obtaining a first matching degree between the first candidate trajectory sequence and the current observation state specifically includes: S410: Add known noise to the first candidate trajectory sequence to obtain a noisy candidate trajectory sequence.
[0040] In one possible embodiment, the known noise is known Gaussian noise. It can be fixed Gaussian noise, that is, a randomly sampled noise is saved and used as the fixed known noise. When adding noise, all first candidate trajectory sequences are added with the known noise. Alternatively, it can be randomly sampled Gaussian noise, that is, different known noises are added to different first candidate trajectory sequences.
[0041] It should be noted that in the navigation task of a mobile robot, the trajectory sequence usually represents the actions that the mobile robot will perform in the future. waypoints. In general, the maximum value of the diffusion time step is set to , and select The trajectory sequence is a discrete sequence of future waypoints. Therefore, when adding noise to the trajectory sequence, it is necessary to offset and subtract the relative coordinates of adjacent waypoints in the trajectory sequence to obtain 8 relative coordinate increments to form an incremental sequence. This can be approximately considered continuous, so the incremental sequence can be used as the object for noise addition.
[0042] S420: Inputting the noisy candidate trajectory sequence, the current observation state, and the time step into a diffusion model to obtain predicted noise corresponding to the first candidate trajectory sequence output by the diffusion model.
[0043] Specifically, if Figure 4 and 7 As shown, the role of the noise estimator is to obtain the predicted noise from the noisy candidate trajectory sequence.
[0044] S430: Calculate a first noise error corresponding to the first candidate trajectory sequence based on the predicted noise of the first candidate trajectory sequence and the corresponding known noise, and use the first noise error as a first matching degree between the first candidate trajectory sequence and the current observation state.
[0045] Specifically, for each first candidate trajectory sequence, the navigation device of the mobile robot adds known noise to the first candidate trajectory sequence to obtain a noisy candidate trajectory sequence. The noisy candidate trajectory sequence, the current observation state, and the time step are then input into the diffusion model to obtain the predicted noise corresponding to the first candidate trajectory sequence. The first noise error corresponding to the first candidate trajectory sequence is then calculated based on the predicted noise of the first candidate trajectory sequence and the corresponding known noise, and the first noise error is used as the first matching degree between the first candidate trajectory sequence and the current observation state.
[0046] like Figure 8 As shown in , in the prior art, the noise estimator gradually removes the random Gaussian noise in the current observation state in K rounds of iterations to restore the executed trajectory sequence (i.e., generation capability). Figure 4 As shown, in this embodiment of the present application, the noise estimator is used to predict noise rather than remove it. This predicted noise is compared with the known noise added to the first candidate trajectory sequence to obtain a first noise error. The magnitude of the first noise error is used as the matching value between the first candidate trajectory sequence and the current observation state, thereby performing ranking (i.e., discriminability).
[0047] The embodiment of the present application transforms the generation capability of the diffusion model into the discrimination capability, which not only retains and inherits the diverse navigation skills of the diffusion model based on the current observation state, but also can obtain the matching degree between each candidate trajectory sequence and the current observation state.
[0048] In the above embodiment of obtaining the first matching degree by the first noise error, the accuracy of the first noise error is affected by the known noise itself and the time step. Based on this consideration, in a possible embodiment, the known noise Including random sampling noise and randomly sampled time steps ,Right now On this basis, the first noise error is obtained by weighted calculation of the prediction error of the random sampling noise and the random sampling time step.
[0049] In a possible embodiment, when a certain known noise is added to a certain first candidate trajectory sequence, the weight of the prediction error of the randomly sampled noise is the ratio of the first difference between the predicted noise and the known noise of the randomly sampled noise to the second difference between the predicted noise and the known noise corresponding to all first candidate trajectory sequences for which noise prediction is performed simultaneously.
[0050] In practice, the prediction error of the diffusion model is more dependent on the noise itself. For all candidate trajectory sequences, difficult-to-predict noise results in larger prediction errors, while easier-to-predict noise results in smaller errors. In other words, if a specific known noise has a large prediction error for a particular candidate trajectory sequence, the prediction error of that known noise for other trajectory sequences will also be large. Therefore, by fixing the noise and comparing the relative error sizes at the same error level, a more accurate ranking result can be achieved. Based on this, in a preferred embodiment, the randomly sampled noise is fixed.
[0051] From the perspective of time series, as the time step increases, the noise level in the first candidate trajectory sequence also gradually increases. Under this condition, the difficulty of distinguishing all first candidate trajectory sequences is the same. Time steps with higher prediction accuracy should have a higher weight in the final result, so the model needs to be able to distinguish different first candidate trajectory sequences at different time steps.
[0052] Figure 5 This is a graph showing the relationship between the time step and noise prediction accuracy obtained using different training data sets based on the mobile robot navigation method of this application, where Carla v1 and OMD v1 are two different training data sets. Figure 5 As shown in , using different time steps to predict the added known noise on the training dataset will result in different accuracy values, and the diffusion model provides accurate estimates at early and middle time steps. In other words, now the total The time step classifiers can integrate the results of all time step predictions according to their noise estimation capabilities, that is, normalize the prediction accuracy and use it as the weight of the time step : (1) in, is the time step The prediction accuracy is calculated, and finally the score of the first candidate trajectory sequence at each time step is normalized and weighted.
[0053] In the embodiment of the present application, the influence of the noise itself and the time step on the noise error is taken into consideration, and the first noise error is calculated in a weighted manner, thereby improving the accuracy of the first noise error.
[0054] Based on the above, in a preferred embodiment, in step S410, multiple known noises are added to each first candidate trajectory sequence, thereby obtaining multiple second noise errors for each first candidate trajectory sequence. Furthermore, a first noise error for each first candidate trajectory sequence is calculated based on the second noise errors corresponding to all known noises in the first candidate trajectory sequence. For example, the first noise error is calculated using mean squared error (MSE).
[0055] Specifically, if the sampling Group known noise and add it to the candidate trajectory sequence In the A sequence of candidate trajectories with noisy , then the candidate trajectory sequence with noise is used as the input of the diffusion model, and the noise is predicted under the condition of the current observation state. Finally, the mean square error is calculated based on all the predicted noises and the corresponding known noises corresponding to each first candidate trajectory sequence to obtain the first noise error of the first candidate trajectory sequence (please refer to Figure 7 The left part of ). Among them, a first candidate trajectory sequence with the minimum first noise error Best matches the current observation state.
[0056] In the embodiment of the present application, a plurality of known noises are added to each first candidate trajectory sequence, and the first noise error is calculated using the plurality of second noise errors, thereby improving the accuracy of the noise error.
[0057] Existing navigation methods based on diffusion strategies typically directly input constraint information (such as geometric cost, semantic cost, map, target point, global path, etc.) into the diffusion model as guiding conditions. Although geometric or semantic information is generally feature-invariant outside the training dataset, which can improve the generalization of the model when deployed in unknown environments, training datasets often come from human demonstration data, which is generally safe and standardized. However, in actual deployment, the observed state distribution may shift, causing the diffusion model to be unable to adapt to the new environment, resulting in system failure. The diffusion model may also generate erroneous trajectory sequences that do not match the current observed state, leading to the robot entering potentially dangerous situations. If the trajectory sequence output by the diffusion model exceeds the allowed range, the robot may face unpredictable risks (such as collisions).
[0058] Furthermore, in diverse semantic scenarios, guidance conditions such as maps, target points, and global paths may change over time, necessitating retraining of the diffusion strategy model once new data is acquired. While geometric or semantic information is typically feature-invariant outside of the training dataset, improving the generalization of the model when deployed in unknown environments, relying solely on these invariant features in diverse semantic scenarios will limit the diffusion strategy model's ability to learn more navigation skills, thereby weakening the model's ability to transfer across different application scenarios. For example, complex navigation tasks such as navigating around puddles of water, traversing tall grass, or prioritizing flat and comfortable paved roads cannot be effectively learned and adapted based solely on invariant features.
[0059] Based on the above considerations, in a possible embodiment, Figure 6 As shown, the navigation method of the mobile robot provided by this application includes: S610: Taking the current observation state as a guide condition, obtaining first matching degrees between a plurality of first candidate trajectory sequences and the current observation state based on a diffusion model.
[0060] Step S610 is the same as step S110, please refer to the above description. This stage is recorded as the prediction and sorting stage.
[0061] S620: Based on the auxiliary constraints corresponding to the current observation state, at least one second candidate trajectory sequence is selected from the multiple first candidate trajectory sequences, and a second matching degree corresponding to the second candidate trajectory sequence is obtained. The second candidate trajectory sequence satisfies the auxiliary constraints, and the second matching degree is related to the first matching degree and the matching value of the second candidate trajectory sequence with the auxiliary constraints. This stage is referred to as the re-ranking stage and is independent of the diffusion model.
[0062] Among them, combined Figure 7Auxiliary constraints are obtained through data analysis of the current observation state, such as geometric maps such as local occupancy maps or elevation maps, semantic maps, combined cost maps, semantic cost maps, or hybrid cost maps. Specifically, a depth map is obtained through a sensor (such as a depth camera or lidar), or a depth map is obtained using monocular depth estimation. This depth map is then used to construct a geometric map such as a local occupancy map or elevation map. For example, an expansion algorithm is used to expand the edges of occupied space to ensure the mobile robot's passability in unoccupied space. Semantic segmentation is performed on the current observation image to obtain a passability score for each pixel, which is then used to construct a semantic map.
[0063] In an embodiment of the present application, auxiliary constraints are generated using the current observation state, making the auxiliary constraints real-time, thereby identifying constraints in a real-time unknown environment. Compared with the prior art that uses invariant features as guiding conditions, the embodiment of the present application can adapt to diverse environments, output a higher degree of matching, and ensure the correctness of the navigation trajectory.
[0064] In step S620 , at least one of the multiple auxiliary constraint conditions may be selected as needed to screen the multiple first candidate trajectory sequences, and the first candidate trajectory sequences that meet the auxiliary constraint conditions are retained as the second candidate trajectory sequences.
[0065] The purpose of screening is further explained below using the local occupancy map as an example. The local occupancy map represents the occupied spaces in the mobile robot's environment. During the screening process, the first candidate trajectory sequence is determined to determine whether it interferes with the occupied space (for example, if it passes through the occupied space). If so, the first candidate trajectory sequence is eliminated, retaining the first candidate trajectory sequence located in the unoccupied space as the second candidate trajectory sequence. Other auxiliary constraints have different constraints.
[0066] During the screening process, each second candidate trajectory sequence has a matching value with each auxiliary constraint condition. .in, Indicates the i The first candidate trajectory sequence and the j The matching value of auxiliary constraints, Represents the total number of all auxiliary constraints.
[0067] In a possible embodiment, the second matching degree of the second candidate trajectory sequence is its first matching degree Matching values for all auxiliary restrictions The arithmetic sum or weighted sum of .
[0068] Thus, all second candidate trajectory sequences are reordered according to the second matching degree to obtain a reordered list.
[0069] S630: Using the second candidate trajectory sequence with the second highest matching degree as the navigation trajectory of the mobile robot.
[0070] Specifically, the mobile robot's navigation device uses the current observation state as a guide condition and, based on a diffusion model, obtains a first degree of matching between multiple first candidate trajectory sequences and the current observation state. Then, based on the auxiliary constraints corresponding to the current observation state, a second candidate trajectory sequence that meets the auxiliary constraints and a second degree of matching corresponding to the second candidate trajectory sequence are selected from the multiple first candidate trajectory sequences. The second candidate trajectory sequence with the highest second degree of matching is then used as the mobile robot's navigation trajectory.
[0071] The embodiment of the present application uses auxiliary constraints outside the diffusion model to screen candidate trajectory sequences, eliminating dangerous candidate trajectory sequences, so that the final navigation trajectory is within a safe range. Through a hybrid strategy, the embodiment of the present application can generalize to unknown environments without retraining the diffusion model due to unknown environments. While retaining and inheriting the diverse navigation skills of the diffusion model, it reduces the security risks faced by deployment in unknown environments, thereby improving the navigation success rate deployed in unknown environments, and has demonstrated a high success rate and a longer exploration distance in multiple challenging test scenarios.
[0072] After obtaining the navigation trajectory, the navigation device of the mobile robot inputs the navigation trajectory into the controller of the mobile robot, outputs control signals such as the linear velocity and angular velocity of the mobile robot, and drives the chassis of the mobile robot to complete the navigation or exploration task.
[0073] In one possible embodiment, a proportional plus derivative controller (PPDC), a pure pursuit controller (PPC), or a model predictive controller (MPC) is selected as the robot controller, and the controller determines the sequence of actions to be executed corresponding to the navigation trajectory. , calculate the linear velocity of the driving robot chassis and angular velocity Wait for the control signal to control the robot to track the fifth point in the action sequence.
[0074] Figure 9This is an example of the navigational performance of a mobile robot based on the present invention in a single unknown scene. In the local occupancy map in the upper right corner of each frame, the black area represents the occupied grid, the white area represents the freely navigable area, the red lines represent rejected candidate trajectory sequences, the green lines represent accepted candidate trajectory sequences (i.e., the second candidate trajectory sequence), and the blue lines represent the candidate trajectory sequence finally selected for execution (i.e., the navigation trajectory). The Ours-DI on the left represents the depth map obtained using a sensor, while the Ours-DE on the right represents the depth map obtained using monocular depth estimation. Although their depth map acquisition methods differ, the robot can accurately obtain the local occupancy map even in the presence of small obstacles on the road. In most cases without obstacles, after two-stage sorting, the top-ranked trajectory sequences are similar to those generated by the diffusion model, demonstrating that the present invention inherits the navigation capabilities of the original diffusion model. However, the trajectory sequences selected by the diffusion model sometimes approach obstacles or make errors, while the trajectory sequences selected by the present invention avoid obstacles.
[0075] Figure 10 This is an example of the mobile robot navigation method provided by this application being able to pass through multiple unknown scenarios. Faced with these challenging scenarios in the figure, the implementation method of this application can successfully pass them, which also verifies the practical value of this application.
[0076] Based on the above, the present application further provides a navigation device for a mobile robot. The navigation device for a mobile robot and the above-mentioned navigation method for a mobile robot can be referred to in correspondence with each other.
[0077] As a possible embodiment, Figure 11 As shown, the navigation device of the mobile robot includes a matching module 1110 and a navigation trajectory determination module 1120 .
[0078] The matching module 1110 is configured to obtain a first matching degree between a plurality of first candidate trajectory sequences and the current observation state based on a diffusion model, taking the current observation state as a guide condition.
[0079] The navigation trajectory determining module 1120 is configured to use the first candidate trajectory sequence with the highest first matching degree as the navigation trajectory of the mobile robot.
[0080] The present embodiment selects the first candidate trajectory sequence with the highest degree of match with the current observation state from multiple first candidate trajectory sequences and uses it as the navigation trajectory for the mobile robot. Compared to existing navigation trajectory determination schemes that lack candidate trajectory sequences, this significantly improves the degree of match between the navigation trajectory and the current observation state and significantly reduces the navigation failure rate. Furthermore, inputting a representative first candidate trajectory sequence into a diffusion model reduces the time and space complexity of model inference compared to calculating the degree of match between the entire training dataset and the current observation state.
[0081] In one possible embodiment, Figure 12 As shown, the navigation device of the mobile robot includes a matching module 1210 , a screening module 1220 and a navigation trajectory determination module 1230 .
[0082] The matching module 1210 is configured to obtain a first matching degree between a plurality of first candidate trajectory sequences and the current observation state based on a diffusion model, taking the current observation state as a guide condition.
[0083] The screening module 1220 is configured to screen out at least one second candidate trajectory sequence from the plurality of first candidate trajectory sequences based on the auxiliary constraint corresponding to the current observation state, and obtain a second matching degree corresponding to the second candidate trajectory sequence; wherein the second candidate trajectory sequence meets the auxiliary constraint, and the second matching degree is related to the first matching degree and the matching value of the second candidate trajectory sequence with the auxiliary constraint. The navigation trajectory determining module 1230 is configured to use the second candidate trajectory sequence with the second highest matching degree as the navigation trajectory of the mobile robot.
[0084] The embodiment of the present application uses auxiliary constraints outside the diffusion model to screen candidate trajectory sequences, eliminate dangerous candidate trajectory sequences, so that the final navigation trajectory is within a safe range. It can be generalized to unknown environments without retraining the diffusion model due to unknown environments. Through a hybrid strategy, while retaining and inheriting the diverse navigation skills of the diffusion model, it reduces the safety risks faced by deployment in unknown environments, thereby improving the navigation success rate deployed in unknown environments, and showing a higher success rate and a longer exploration distance in multiple challenging test scenarios.
[0085] Based on the above, in a possible embodiment, the matching module 1110 and / or the matching module 1210 is further configured to: Adding known noise to the first candidate trajectory sequence to obtain a noisy candidate trajectory sequence; Inputting the noisy candidate trajectory sequence, the current observation state, and the time step into the diffusion model to obtain the predicted noise corresponding to the first candidate trajectory sequence; A first noise error corresponding to the first candidate trajectory sequence is calculated based on the predicted noise corresponding to the first candidate trajectory sequence and the corresponding known noise, and the first noise error is used as a first matching degree between the first candidate trajectory sequence and the current observation state.
[0086] The embodiment of the present application transforms the generation capability of the diffusion model into the discrimination capability, which not only retains and inherits the diverse navigation skills of the diffusion model based on the current observation state, but also can obtain the matching degree between each candidate trajectory sequence and the current observation state.
[0087] Figure 13 This is a schematic diagram of the structure of the electronic device provided by this application. Figure 13 As shown, the electronic device may include: a processor 1310, a communications interface 1320, a memory 1330, and a communications bus 1340. The processor 1310, communications interface 1320, and memory 1330 communicate with each other via the communications bus 1340. The processor 1310 may invoke logic instructions in the memory 1330 to execute the aforementioned mobile robot navigation method. In one possible embodiment, the method includes: using the current observation state as a guide condition, obtaining a first degree of match between multiple first candidate trajectory sequences and the current observation state based on a diffusion model; and selecting the first candidate trajectory sequence with the highest first degree of match as the navigation trajectory of the mobile robot.
[0088] In addition, the logical instructions in the above-mentioned memory 1330 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present application, or the part that contributes to the existing technology, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program code.
[0089] On the other hand, the present application also provides a computer program product, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions. When the program instructions are executed by a computer, the computer is capable of executing the mobile robot navigation method provided in each of the above embodiments. In one possible embodiment, the method includes: using the current observation state as a guide condition, obtaining a first matching degree between multiple first candidate trajectory sequences and the current observation state based on a diffusion model; and using the first candidate trajectory sequence with the highest first matching degree as the navigation trajectory of the mobile robot.
[0090] In another aspect, the present application further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the mobile robot navigation method provided in each of the above embodiments. In one possible embodiment, the method includes: using the current observation state as a guide condition, obtaining a first degree of match between multiple first candidate trajectory sequences and the current observation state based on a diffusion model; and selecting the first candidate trajectory sequence with the highest first degree of match as the navigation trajectory of the mobile robot.
[0091] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.
[0092] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.
[0093] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as a read-only memory (ROM) / random access memory (RAM), a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.
[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A navigation method for a mobile robot, characterized in that: include: Taking the current observation state as a guide condition, obtaining a first matching degree between a plurality of first candidate trajectory sequences and the current observation state based on a diffusion model; Using the first candidate trajectory sequence with the highest first matching degree as the navigation trajectory of the mobile robot; The diffusion model is used to obtain a prediction result corresponding to each first candidate trajectory sequence, and the prediction result is used to calculate the first matching degree.
2. The mobile robot navigation method according to claim 1, characterized in that: The navigation method of the mobile robot further includes: Based on the auxiliary constraint corresponding to the current observation state, at least one second candidate trajectory sequence is selected from the plurality of first candidate trajectory sequences, and a second matching degree corresponding to the second candidate trajectory sequence is obtained; wherein the second candidate trajectory sequence satisfies the auxiliary constraint, and the second matching degree is correlated with the first matching degree and a matching value of the second candidate trajectory sequence with the auxiliary constraint; The second candidate trajectory sequence with the second highest matching degree is used as the navigation trajectory of the mobile robot.
3. The mobile robot navigation method according to claim 1, characterized in that: For each first candidate trajectory sequence, obtaining a first matching degree between the first candidate trajectory sequence and the current observation state specifically includes: adding known noise to the first candidate trajectory sequence to obtain a noisy candidate trajectory sequence; Inputting the noisy candidate trajectory sequence, the current observation state, and the time step into a diffusion model to obtain predicted noise corresponding to the first candidate trajectory sequence; A first noise error corresponding to the first candidate trajectory sequence is calculated based on the predicted noise and the corresponding known noise, and the first noise error is used as a first matching degree between the first candidate trajectory sequence and the current observation state.
4. The mobile robot navigation method according to claim 3, characterized in that: The known noise includes random sampling noise and random sampling time step, and the first noise error is obtained by weighted calculation of the prediction error of the random sampling noise and the random sampling time step.
5. The mobile robot navigation method according to any one of claims 3 to 4, characterized in that: A plurality of known noises are added to each first candidate trajectory sequence, and a first noise error corresponding to each first candidate trajectory sequence is obtained based on second noise errors of all known noises.
6. The mobile robot navigation method according to claim 2, characterized in that: The auxiliary constraint condition is obtained by performing data analysis on the current observation state.
7. The mobile robot navigation method according to claim 1 or 2, characterized in that: The plurality of first candidate trajectory sequences are obtained by preprocessing all trajectory sequences in the training data set, where the preprocessing includes normalization and clustering.
8. A navigation device for a mobile robot, characterized in that: It includes a matching module and a navigation trajectory determination module; The matching module is configured to use the current observation state as a guide condition and obtain a first matching degree between a plurality of first candidate trajectory sequences and the current observation state based on a diffusion model. The diffusion model is configured to obtain a prediction result corresponding to each first candidate trajectory sequence, and the prediction result is used to calculate the first matching degree. The navigation trajectory determination module is configured to use the first candidate trajectory sequence with the highest first matching degree as the navigation trajectory of the mobile robot.
9. The navigation device for a mobile robot according to claim 8, wherein: The navigation device further includes a screening module, the screening module being configured to screen at least one second candidate trajectory sequence from the plurality of first candidate trajectory sequences based on an auxiliary constraint condition corresponding to a current observation state, and obtain a second matching degree corresponding to the second candidate trajectory sequence; wherein the second candidate trajectory sequence satisfies the auxiliary constraint condition, and the second matching degree is correlated with the first matching degree and a matching value of the second candidate trajectory sequence with the auxiliary constraint condition; The navigation trajectory determination module is further configured to use the second candidate trajectory sequence with the second highest matching degree as the navigation trajectory of the mobile robot.
10. The navigation device for a mobile robot according to claim 8, wherein: The matching module is further configured to: Adding known noise to the first candidate trajectory sequence to obtain a noisy candidate trajectory sequence; Inputting the noisy candidate trajectory sequence, the current observation state, and the time step into a diffusion model to obtain predicted noise corresponding to the first candidate trajectory sequence; A first noise error corresponding to the first candidate trajectory sequence is calculated based on predicted noise and corresponding known noise corresponding to the first candidate trajectory sequence, and the first noise error is used as a first matching degree between the first candidate trajectory sequence and the current observation state.
11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the navigation method of the mobile robot according to any one of claims 1 to 7 is implemented.
12. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the navigation method of the mobile robot according to any one of claims 1 to 7 is implemented.
13. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the navigation method of the mobile robot according to any one of claims 1 to 7 is implemented.