Robot-assisted navigation device and method based on artificial intelligence

Abstract

The invention discloses a robot aided navigation device and method based on artificial intelligence, and relates to the technical field of aided navigation, and the method comprises the following steps: collecting navigation environment data in real time through a multi-mode sensing equipment group; establishing a preliminary environment three-dimensional map through a simultaneous positioning and mapping technology; reversely deducing an unknown area by using the preliminary environment three-dimensional map, deducing based on physical and geometric constraints of the environment, and restoring and perfecting the environment three-dimensional map through a reverse simulation model; compensating and optimizing the environment three-dimensional map by using deep learning, and continuously updating and maintaining the environment three-dimensional map by fusing the dynamic data of the robot; an optimal navigation path is searched through a global path planning algorithm, the optimal path is adjusted in real time to respond to dynamic environment changes, and the model is used for predicting future navigation environment changes. According to the method, environment data are collected in real time through the multi-mode sensing equipment, an unknown area is reversely deduced by utilizing a preliminary three-dimensional map, optimization and updating are carried out, and dynamic navigation path planning is realized.

Description

Technical Field

[0001] The present invention relates to the field of assisted navigation technology, and in particular to an artificial intelligence-based robot assisted navigation device and method. Background Art

[0002] With the widespread application of robots in logistics, manufacturing, services and other fields, the demand for their autonomous navigation capabilities continues to increase. Currently, common navigation technologies mainly rely on sensors such as lidar and cameras to collect environmental information, combined with simultaneous positioning and mapping technology to achieve environmental modeling and path planning.

[0003] The existing technology has the following deficiencies: The existing navigation methods still have some deficiencies in complex and changing environments, including:

[0004] 1. Insufficient adaptability to dynamically changing environments, making it difficult to update environmental maps in real time to cope with obstacles and environmental changes;

[0005] 2. Error accumulation during the construction of 3D maps leads to positioning deviation, affecting navigation accuracy;

[0006] 3. Difficulty in achieving multi-scale and multi-level environmental understanding and analysis;

[0007] 4. Lack of inference and supplementation capabilities for unknown or unmapped areas limits autonomous navigation performance in unknown environments;

[0008] 5. Traditional path planning algorithms respond slowly to dynamic environments, making it difficult to achieve real-time optimization and predict future scene changes.

[0009] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0010] The purpose of the present invention is to provide a robot-assisted navigation device and method based on artificial intelligence to solve the problems in the above-mentioned background technology.

[0011] In order to achieve the above objectives, the present invention provides the following technical solutions: a robot-assisted navigation method based on artificial intelligence, specifically comprising:

[0012] Collect navigation environment data in real time through a multimodal sensing device group;

[0013] Pre-process the navigation environment data and build a preliminary 3D map of the environment through simultaneous positioning and mapping technology;

[0014] Use the preliminary 3D map of the environment to infer the unknown area, and make inferences based on the physical and geometric constraints of the environment, and restore and improve the 3D map of the environment through the reverse simulation model;

[0015] Use deep learning to compensate and optimize the 3D map of the environment, and integrate the robot's dynamic data to continuously update and maintain the 3D map of the environment;

[0016] Find the optimal navigation path through a global path planning algorithm, adjust the optimal path in real time to respond to dynamic environmental changes, and use the model to predict future changes in the navigation environment;

[0017] As a preferred solution of the intelligent agent-based power distribution system health monitoring and prediction method described in the present invention, the multimodal sensing device group and navigation environment data specifically include:

[0018] LiDAR for collecting spatial structure data of the navigation environment;

[0019] A camera for collecting dense depth image data;

[0020] Ultrasonic sensor for collecting distance information data;

[0021] Inertial measurement unit for collecting dynamic data of the robot;

[0022] The multimodal perception device group is installed at different positions of the robot, and the navigation environment coverage is increased through perspective complementarity. The robot's own master clock source is selected as the time reference of the multimodal perception device group, and the acquisition rate is adjusted according to the dynamic requirements of the environment.

[0023] As a preferred solution of the agent-based power distribution system health monitoring and prediction method described in the present invention, wherein:

[0024] The pretreatment specifically includes:

[0025] Select a reference point to initialize the robot position as the origin of the preliminary 3D map of the environment;

[0026] Filter, calibrate and register navigation environment data, extract high-dimensional features, and perform feature matching;

[0027] The establishment of a preliminary three-dimensional map of the environment specifically includes:

[0028] Estimate the change in the robot's current position relative to its previous position based on the feature matching results;

[0029] Optimize the current robot's pose estimation through extended Kalman filtering;

[0030] Integrate the current navigation environment data into the existing map and gradually build a preliminary 3D environment model using point cloud stitching;

[0031] The closed loop detection system checks whether the robot scans the known map area again to reduce the cumulative error. When a closed loop is detected, global fusion and scanning optimization are performed.

[0032] As a preferred solution of the agent-based power distribution system health monitoring and prediction method described in the present invention, wherein:

[0033] The method of using the preliminary three-dimensional map of the environment to infer the unknown area specifically includes:

[0034] Through deep learning, the error states of the multimodal perception device group in different navigation environments are learned and updated in real time;

[0035] Use Bayesian analysis to determine the uncertainty boundary of the error state and output the confidence interval of the error;

[0036] Propagate the error as a dynamic variable in the navigation environment to identify the deviation of unknown areas of the preliminary 3D map of the environment;

[0037] Calculate the error boundary of each unknown region based on error propagation, specifically including:

[0038] The expected value and standard deviation of the initialization error are defined as the error bound. Each unknown area is affected by the errors of neighboring areas. A neighborhood influence model is established by accumulating influence weights to propagate the error, and the expectation and covariance are updated. When any upper bound of the error exceeds the confidence interval of the error, the unknown area of ​​the preliminary 3D map of the environment is marked.

[0039] The restoration and improvement of the three-dimensional map of the environment through the reverse simulation model specifically includes:

[0040] Encode the geometric relationships and scene layout of the preliminary 3D map of the environment;

[0041] Input the high-dimensional features of the existing map and train the autoencoder to learn the latent space representation of the preliminary 3D map of the environment;

[0042] The decoder is used to reconstruct a preliminary 3D map of the environment, and the missing structures in its unknown areas are inferred through inverse reconstruction.

[0043] The identification of the unknown area deviation of the preliminary three-dimensional environment map specifically includes:

[0044] By introducing error impact transmission, the error is "amplified" or "suppressed" between adjacent areas. The impact strength is dynamically adjusted according to the uncertainty of the local error. When the errors in the neighborhood are significantly different, the impact is amplified and potential error mutations or missing areas are marked in time. The specific formula is:

[0045]

[0046] in, represents the increment of error in unknown region i, represents the neighborhood of the unknown region i, represents the adjacency weight, 、 Represent the error vectors of unknown region i and unknown region j respectively, () represents the adaptive enhancement function.

[0047] As a preferred solution of the agent-based power distribution system health monitoring and prediction method described in the present invention, wherein:

[0048] The fusion robot dynamic data continuously updates and maintains the three-dimensional map of the environment, specifically including:

[0049] Unified point cloud coordinate system based on the origin of the 3D map of the environment;

[0050] Based on the robot's pose points in the previous time period as the initial estimate, the nearest point of the target pose point is matched in the point cloud coordinate system to obtain the rigid transformation with the minimum distance between the two points;

[0051] Update the robot's current relative position in the three-dimensional map of the environment based on the rotation matrix and translation vector through the deep network;

[0052] The update quality is judged based on the matching error, and the currently updated relative pose point is used as the reference for the next fusion.

[0053] As a preferred solution of the agent-based power distribution system health monitoring and prediction method described in the present invention, wherein:

[0054] The searching for the optimal navigation path specifically includes:

[0055] Establish a perception network to divide the three-dimensional map of the environment into multi-scale situations, including:

[0056] Macroscale: The global scene of the 3D map of the environment;

[0057] Mesoscale: Demarcates the global scene into intersections, obstacle boundaries, and road widths, and outputs route details and potential obstacle locations based on deep learning models.

[0058] Microscale: Outputs the detailed obstacle locations and types of the global scene based on the missing structure of the unknown area;

[0059] Extract multi-scale situational features and use deep neural networks to predict the "path cost map" in the three-dimensional map of the environment. Define the navigation path length and obstacle risk level corresponding to each grid in the cost map.

[0060] Based on the acquisition time period of the multimodal sensing device group, a time series model is used to predict dynamic obstacles that may appear during navigation, adjust the current cost map, and dynamically increase the cost of potential danger areas;

[0061] By introducing a dynamic cost map, the A* algorithm is improved to select the starting point to initiate navigation search, generate candidate paths of corresponding scales, and select the optimal navigation path based on the three-dimensional map of the environment.

[0062] On the other hand, the present invention provides an artificial intelligence-based robot-assisted navigation device, which specifically includes a multimodal perception device group, a data processing unit, an analysis and decision-making unit, and a communication interface unit. The data processing unit specifically includes an embedded processor for filtering, calibrating, and aligning navigation environment data, and a memory for storing environmental three-dimensional maps and high-order features; the analysis and decision-making unit specifically includes a processing chip for real-time learning and compensation of environmental error states, and a path planning module for generating and adjusting navigation paths based on environmental three-dimensional maps; the communication interface unit is used to realize a wireless module for data transmission with other units.

[0063] On the other hand, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, the steps of an artificial intelligence-based robot-assisted navigation method as described above in the present invention are implemented.

[0064] On the other hand, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, the steps of an artificial intelligence-based robot-assisted navigation method as described above in the present invention are implemented.

[0065] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0066] Through the collaborative work of a multimodal perception device group, comprehensive, multi-perspective navigation environment information is obtained. Multi-scale situation division makes the environment understanding more delicate and comprehensive, and improves the accuracy and robustness of navigation. By using closed-loop detection, global fusion, and scanning optimization, the cumulative error is significantly reduced. Through deep learning and Bayesian analysis, deviations in unknown areas are detected in real time, potential environmental changes or deviations are identified in a timely manner, and the accuracy and stability of the map are effectively improved. With the help of the error propagation model, the neighborhood influence relationship is used to dynamically identify unknown areas with large deviations. Combined with inverse simulation and autoencoder technology, the inference and supplement capabilities of unknown or unmapped areas are enhanced, and the path cost map is adjusted according to the prediction to avoid potential dangerous areas in advance and achieve "active" obstacle avoidance. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0068] Figure 1 The present invention is a flowchart of a robot-assisted navigation method based on artificial intelligence. DETAILED DESCRIPTION

[0069] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0070] Example 1, with reference to Figure 1 , which is the first embodiment of the present invention, provides a robot-assisted navigation device and method based on artificial intelligence, specifically including:

[0071] S1. Real-time collection of navigation environment data through a multimodal sensing device group;

[0072] The multimodal sensing device group and navigation environment data specifically include:

[0073] LiDAR for collecting spatial structure data of the navigation environment;

[0074] A camera for collecting dense depth image data;

[0075] Ultrasonic sensor for collecting distance information data;

[0076] Inertial measurement unit for collecting dynamic data of the robot;

[0077] The multimodal perception device group is installed at different positions of the robot, and the navigation environment coverage is increased through perspective complementarity. The robot's own master clock source is selected as the time reference of the multimodal perception device group, and the acquisition rate is adjusted according to the dynamic requirements of the environment.

[0078] It should be noted that the complementary perspectives of different sensing devices can reduce shadow occlusion, avoid blind spots, and enhance the perception of depth and spatial structure;

[0079] It should be noted that the robot's own master clock source is selected as the synchronization reference for multiple sensors to ensure the time consistency of the collected data, thereby reducing the registration error caused by time difference;

[0080] It should also be noted that the acquisition rate is adjusted in real time according to environmental changes (such as speed, complexity, and the appearance of dynamic obstacles) to balance energy consumption and perception effect.

[0081] S2. Preprocess the navigation environment data and build a preliminary three-dimensional map of the environment through simultaneous positioning and mapping technology;

[0082] The pretreatment specifically includes:

[0083] Select a reference point to initialize the robot position as the origin of the preliminary 3D map of the environment;

[0084] Filter, calibrate and register navigation environment data, extract high-dimensional features, and perform feature matching;

[0085] It should be noted that different filtering algorithms (such as Kalman filtering, median filtering, mean filtering or outlier filtering) are used for different sensing devices to optimize the sensing effect;

[0086] It should be noted that internal and external calibrations are performed on different sensing devices to eliminate the deviations between the sensing devices;

[0087] It should be noted that feature matching achieves accurate motion estimation and environment alignment by matching continuous feature points through feature descriptors.

[0088] The establishment of a preliminary three-dimensional map of the environment specifically includes:

[0089] Estimate the change in the robot's current position relative to its previous position based on the feature matching results;

[0090] Optimize the current robot's pose estimation through extended Kalman filtering;

[0091] Integrate the current navigation environment data into the existing map and gradually build a preliminary 3D environment model using point cloud stitching;

[0092] The closed loop detection system checks whether the robot scans the known map area again to reduce the cumulative error. When a closed loop is detected, global fusion and scanning optimization are performed.

[0093] It should be noted that the extended Kalman filter optimizes the current robot's pose estimation, including the initialization, prediction and correction of the filter, to ensure accurate estimation of the current position in a dynamic environment;

[0094] It should also be noted that the closed-loop detection presets unique landmark features (such as special structures, color labels or persistent geometric features) in the navigation environment, constructs a hash index based on key features in the map area (such as local feature descriptor hash), and quickly retrieves potential matching areas. Once a closed loop is detected, a closed-loop constraint is established, and the current node is connected to the identified corresponding map node to form a closed-loop edge; the closed-loop constraint is used as a global optimization goal, and panoramic optimization is performed. Through multiple iterations, error convergence is achieved, thereby improving the accuracy and credibility of the environment map.

[0095] S3. Use the preliminary 3D map of the environment to infer the unknown area, and make inferences based on the physical and geometric constraints of the environment, and restore and improve the 3D map of the environment through the reverse simulation model;

[0096] The method of using the preliminary three-dimensional map of the environment to infer the unknown area specifically includes:

[0097] Through deep learning, the error states of the multimodal perception device group in different navigation environments are learned and updated in real time;

[0098] Use Bayesian analysis to determine the uncertainty boundary of the error state and output the confidence interval of the error;

[0099] Propagate the error as a dynamic variable in the navigation environment to identify the deviation of unknown areas of the preliminary 3D map of the environment;

[0100] Calculate the error bounds of each unknown area based on error propagation. When any upper bound of the error exceeds the confidence interval of the error, mark the unknown area of ​​the preliminary 3D map of the environment.

[0101] It should be noted that the error propagation mechanism in the environment specifically includes treating the error as a dynamic random variable, performing Bayesian propagation based on environmental characteristics (such as the error state, path, and scene layout of the neighboring area), dynamically adjusting the error transmission intensity through the neighborhood influence model, diffusing the error within the neighborhood, simulating the process of error accumulation and contagion, and updating the error estimate of each area in real time based on global information to form a spatially distributed error field;

[0102] It should also be noted that by continuously monitoring the evolution of the error boundary, combining the outputs of deep learning and Bayesian inference, the error model parameters can be dynamically adjusted to improve the accuracy of error prediction.

[0103] The restoration and improvement of the three-dimensional map of the environment through the reverse simulation model specifically includes:

[0104] Encode the geometric relationships and scene layout of the preliminary 3D map of the environment;

[0105] Input the high-dimensional features of the existing map and train the autoencoder to learn the latent space representation of the preliminary 3D map of the environment;

[0106] The decoder is used to reconstruct a preliminary 3D map of the environment, and the missing structures in its unknown areas are inferred through inverse reconstruction.

[0107] The identification of the unknown area deviation of the preliminary three-dimensional environment map specifically includes:

[0108] By introducing error impact transmission, the error is "amplified" or "suppressed" between adjacent areas. The impact strength is dynamically adjusted according to the uncertainty of the local error. When the errors in the neighborhood are significantly different, the impact is amplified and potential error mutations or missing areas are marked in time. The specific formula is:

[0109]

[0110] in, represents the increment of error in unknown region i, represents the neighborhood of the unknown region i, represents the adjacency weight, 、 Represent the error vectors of unknown region i and unknown region j respectively, () represents the adaptive enhancement function.

[0111] S4. Use deep learning to compensate and optimize the 3D map of the environment, and integrate the robot's dynamic data to continuously update and maintain the 3D map of the environment;

[0112] The fusion robot dynamic data continuously updates and maintains the three-dimensional map of the environment, specifically including:

[0113] Unified point cloud coordinate system based on the origin of the 3D map of the environment;

[0114] Based on the robot's pose points in the previous time period as the initial estimate, the nearest point of the target pose point is matched in the point cloud coordinate system to obtain the rigid transformation with the minimum distance between the two points;

[0115] Update the robot's current relative position in the three-dimensional map of the environment based on the rotation matrix and translation vector through the deep network;

[0116] The update quality is judged based on the matching error, and the currently updated relative pose point is used as the reference for the next fusion.

[0117] S5. Find the optimal navigation path through a global path planning algorithm, adjust the optimal path in real time to respond to dynamic environmental changes, and use the model to predict future changes in the navigation environment;

[0118] The searching for the optimal navigation path specifically includes:

[0119] Establish a perception network to divide the three-dimensional map of the environment into multi-scale situations, including:

[0120] Macroscale: The global scene of the 3D map of the environment;

[0121] Mesoscale: Demarcates the global scene into intersections, obstacle boundaries, and road widths, and outputs route details and potential obstacle locations based on deep learning models.

[0122] Microscale: Outputs the detailed obstacle locations and types of the global scene based on the missing structure of the unknown area;

[0123] Extract multi-scale situational features and use deep neural networks to predict the "path cost map" in the three-dimensional map of the environment. Define the navigation path length and obstacle risk level corresponding to each grid in the cost map.

[0124] Based on the acquisition time period of the multimodal sensing device group, a time series model is used to predict dynamic obstacles that may appear during navigation, adjust the current cost map, and dynamically increase the cost of potential danger areas;

[0125] By introducing a dynamic cost map, the A* algorithm is improved to select the starting point to initiate navigation search, generate candidate paths of corresponding scales, and select the optimal navigation path based on the three-dimensional map of the environment.

[0126] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0127] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of implementing a robot-assisted navigation method based on artificial intelligence as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0128] Example 2

[0129] The following is another embodiment of the present invention, which provides a robot-assisted navigation method based on artificial intelligence. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0130] In order to verify the effect of the present invention, a vehicle platform-based background is selected, and a multimodal perception device group consisting of lidar, camera, ultrasonic sensor, and inertial measurement unit is constructed to perform environmental scanning.

[0131] All sensing devices collect data once per second, totaling 16.7 hours of data samples, covering an area of ​​5,000 square meters. Within 50 meters of the vehicle, a higher resolution (10 cm / pixel) is used for detailed analysis.

[0132] In farther areas (such as 300 meters away), a lower resolution is used to capture the overall environmental characteristics. The navigation accuracy after using multi-scale situation division is improved by 14%, from the original error of 1.6 meters to 1.4 meters.

[0133] The closed-loop detection mechanism is used to compare the map with the global map. If a deviation is detected (for example, an error exceeding 5%), the error correction process is immediately initiated. The Bayesian analysis method is used to calculate the probability distribution of the error and determine the optimal correction strategy.

[0134] During the experiment, after 20 rounds of closed-loop detection and correction, the overall cumulative error was reduced from 7.8% to 3.2%, significantly improving the stability of navigation.

[0135] In complex environment scenarios, the deep learning model accurately identified 80% of unknown obstacles, significantly improving detection accuracy compared to 60% with traditional methods.

[0136] By analyzing the impact of errors between different regions using an error propagation model and combining it with neighborhood influence relationships, the system can identify areas associated with current deviations and update the path cost map in real time. For example, when a large error is detected on a road segment, the cost of that area is automatically adjusted based on the error propagation model, and the driving route is adjusted accordingly. This has increased the path optimization success rate by 25% and reduced unnecessary detours by 15%.

[0137] Through reverse simulation, the system infers unknown areas. When sudden obstacles appear, the possible obstacle trajectories are inferred through reverse simulation. Combined with autoencoder technology, the inference ability of unmapped areas is enhanced. Active obstacle avoidance successfully avoids 90% of dangerous incidents, which is 25% safer than the 65% of traditional methods.

[0138] Through the collaborative work of a multimodal perception device group, comprehensive, multi-perspective navigation environment information is obtained. Multi-scale situation division makes the environment understanding more delicate and comprehensive, and improves the accuracy and robustness of navigation. By using closed-loop detection, global fusion, and scanning optimization, the cumulative error is significantly reduced. Through deep learning and Bayesian analysis, deviations in unknown areas are detected in real time, potential environmental changes or deviations are identified in a timely manner, and the accuracy and stability of the map are effectively improved. With the help of the error propagation model, the neighborhood influence relationship is used to dynamically identify unknown areas with large deviations. Combined with inverse simulation and autoencoder technology, the inference and supplement capabilities of unknown or unmapped areas are enhanced, and the path cost map is adjusted according to the prediction to avoid potential dangerous areas in advance and achieve "active" obstacle avoidance.

[0139] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. A robot-assisted navigation method based on artificial intelligence, characterized in that: Specifically include: Collect navigation environment data in real time through a multimodal sensing device group; Pre-process the navigation environment data and build a preliminary 3D map of the environment through simultaneous positioning and mapping technology; Use the preliminary 3D map of the environment to infer the unknown area, and make inferences based on the physical and geometric constraints of the environment, and restore and improve the 3D map of the environment through the reverse simulation model; Use deep learning to compensate and optimize the 3D map of the environment, and integrate the robot's dynamic data to continuously update and maintain the 3D map of the environment; The optimal navigation path is found through a global path planning algorithm, and the optimal path is adjusted in real time to respond to dynamic environmental changes, and the model is used to predict future changes in the navigation environment.

2. The artificial intelligence-based robot-assisted navigation method according to claim 1, characterized in that: The multimodal sensing device group and navigation environment data specifically include: LiDAR for collecting spatial structure data of the navigation environment; A camera for collecting dense depth image data; Ultrasonic sensor for collecting distance information data; Inertial measurement unit used to collect dynamic data of the robot.

3. The robot-assisted navigation method based on artificial intelligence according to claim 1, characterized in that: The pretreatment specifically includes: Select a reference point to initialize the robot position as the origin of the preliminary 3D map of the environment; Filter, calibrate and register navigation environment data, extract high-dimensional features, and perform feature matching; The establishment of a preliminary three-dimensional map of the environment specifically includes: Estimate the change in the robot's current position relative to its previous position based on the feature matching results; Optimize the current robot's pose estimation through extended Kalman filtering; Integrate the current navigation environment data into the existing map and gradually build a preliminary 3D environment model using point cloud stitching; The closed loop detects whether the robot scans the known map area again.

4. The artificial intelligence-based robot-assisted navigation method according to claim 1, characterized in that: The method of using the preliminary three-dimensional map of the environment to infer the unknown area specifically includes: Through deep learning, the error states of the multimodal perception device group in different navigation environments are learned and updated in real time; Use Bayesian analysis to determine the uncertainty boundary of the error state and output the confidence interval of the error; Propagate the error as a dynamic variable in the navigation environment to identify the deviation of unknown areas of the preliminary 3D map of the environment; Calculate the error bounds of each unknown area based on error propagation. If any upper bound of the error exceeds the confidence interval of the error, mark the unknown area of ​​the preliminary 3D map of the environment. The restoration and improvement of the three-dimensional map of the environment through the reverse simulation model specifically includes: Encode the geometric relationships and scene layout of the preliminary 3D map of the environment; Input the high-dimensional features of an existing map and train the autoencoder to learn its latent space representation; Use the decoder to reconstruct a preliminary 3D map of the environment; Reverse reconstruction to infer the missing structure in the unknown area.

5. The robot-assisted navigation method based on artificial intelligence according to claim 4, characterized in that: The identification of the unknown area deviation of the preliminary three-dimensional environment map specifically includes: By introducing error impact transfer, the error is "amplified" or "suppressed" between adjacent areas. The impact strength is dynamically adjusted according to the uncertainty of the local error. When the errors in the neighborhood are significantly different, the impact is amplified and potential error mutations or missing areas are marked in time. The specific formula is: in, represents the increment of error in unknown region i, represents the neighborhood of the unknown region i, represents the adjacency weight, 、 Represent the error vectors of unknown region i and unknown region j respectively, () represents the adaptive enhancement function.

6. The robot-assisted navigation method based on artificial intelligence according to claim 1, characterized in that: The fusion robot dynamic data continuously updates and maintains the three-dimensional map of the environment, specifically including: Unified point cloud coordinate system based on the origin of the 3D map of the environment; Based on the robot's pose points in the previous time period as the initial estimate, the nearest point of the target pose point is matched in the point cloud coordinate system to obtain the rigid transformation with the minimum distance between the two points; Update the robot's current relative position in the three-dimensional map of the environment based on the rotation matrix and translation vector through the deep network; The update quality is judged based on the matching error, and the currently updated relative pose point is used as the reference for the next fusion.

7. The artificial intelligence-based robot-assisted navigation method according to claim 1, characterized in that: The searching for the optimal navigation path specifically includes: Establish a perception network to divide the three-dimensional map of the environment into multi-scale situations; Extract multi-scale situational features and use deep neural networks to predict the "path cost map" in the three-dimensional map of the environment. Define the navigation path length and obstacle risk level corresponding to each grid in the cost map. Based on the acquisition time period of the multimodal sensing device group, a time series model is used to predict dynamic obstacles that may appear during navigation, adjust the current cost map, and dynamically increase the cost of potential danger areas; By introducing a dynamic cost map, the A* algorithm is improved to select the starting point to initiate navigation search, generate candidate paths of corresponding scales, and select the optimal navigation path based on the three-dimensional map of the environment.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the artificial intelligence-based robot-assisted navigation method according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the artificial intelligence-based robot-assisted navigation method according to any one of claims 1 to 7 are implemented.

Images