Path planning method and device based on reconstruction uncertainty

Abstract

Embodiments of the present application provide a path planning method and device based on reconstruction uncertainty, the method comprising: continuously collecting environment data through an environment perception sensor mounted on a mobile robot, extracting corresponding environment data key frames and pose estimation of each frame, constructing an initial voxel volume, and introducing an explicit uncertainty voxel volume; sampling path candidate points within the movable range of the mobile robot, eliminating unreachable candidate points in the candidate point set based on the initial voxel volume, performing uncertainty voxel screening and visibility domain judgment on the candidate point set after elimination through the explicit uncertainty voxel volume, obtaining observable voxels for each candidate point after eliminating non-visible voxels, constructing a planning objective function based on the observable voxels, pose estimation information, and path cost of each candidate point, obtaining a next observation pose, and realizing path planning. The present application can improve the mapping efficiency and path planning quality of a three-dimensional map.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to a path planning method and apparatus based on reconstruction uncertainty. Background Technology

[0002] With the rapid development of robotics technology, mobile robots have been widely used in various fields such as warehousing and logistics, intelligent manufacturing, and service robots. In these application scenarios, mobile robots need to achieve autonomous navigation, environmental mapping, and precise localization in complex and dynamically changing indoor environments to ensure that they can complete tasks stably and efficiently. Simultaneous Localization and Mapping (SLAM) technology, as one of the core technologies for autonomous navigation of mobile robots, aims to solve the problem of robot localization and mapping in unknown or semi-unknown environments.

[0003] Traditional SLAM technology primarily relies on sensor data such as LiDAR and cameras, using filtering or optimization methods to achieve robot localization and environmental mapping. However, with the increasing complexity of application scenarios, traditional methods face numerous challenges when dealing with large-scale, highly dynamic, or feature-sparse environments. In recent years, 3D reconstruction methods based on neural implicit fields or hybrid neural representations have been gradually introduced into the SLAM field, providing new insights for solving these problems.

[0004] Neural implicit field methods, by providing continuous representations of scene geometry and texture, enable more refined modeling of environmental information. These methods also introduce uncertainty estimation mechanisms to measure reconstruction quality; that is, by assessing the uncertainty of the reconstruction results, subsequent observations and motion planning are guided, thereby improving the completeness and accuracy of scene reconstruction. However, in practical applications, considering only the existing reconstruction uncertainties is often insufficient to achieve optimal path planning and decision-making.

[0005] Therefore, there is an urgent need for a path planning method based on reconstruction uncertainty, which can improve the mapping efficiency and path planning quality of 3D maps. Summary of the Invention

[0006] To address the problems in the prior art, this application provides a path planning method and apparatus based on reconstruction uncertainty, which can improve the mapping efficiency of 3D maps and the quality of path planning.

[0007] To solve at least one of the above problems, this application provides the following technical solution:

[0008] Firstly, this application provides a path planning method based on reconstruction uncertainty, comprising:

[0009] The mobile robot continuously collects environmental observation data using environmental perception sensors and extracts key frame sequences of the corresponding environmental observation data. The key frame sequences are then calculated using a sensor fusion algorithm to determine the pose estimate for each key frame.

[0010] Based on the keyframe sequence and the pose estimation, an initial voxel volume of the known 3D scene is constructed in a neural implicit manner, and an explicit uncertainty voxel volume is constructed at the same time, wherein the explicit uncertainty voxel volume is used to represent the uncertainty of the unknown scene.

[0011] A hierarchical target space is constructed within the mobile robot's operational range, and grid points are sampled on each layer plane to determine the corresponding candidate point set. The candidate point set is then used to eliminate candidate points in obstacle areas based on the initial voxel volume. The candidate point set after the elimination of candidate points is then used to screen uncertain voxels and determine the visible field based on the explicit uncertainty voxel volume to determine the observable voxel set corresponding to each candidate point. The reconstruction uncertainty aggregation amount of each candidate point is calculated based on the observable voxel set, and the localization uncertainty improvement amount of each candidate point is calculated based on the pose estimation.

[0012] The path cost of the mobile robot is calculated according to the path planning algorithm. Based on the path cost, the amount of reconstruction uncertainty aggregation, and the amount of localization uncertainty improvement, a corresponding unified planning objective function is determined. The next observation pose of the mobile robot is determined according to the unified planning objective function, and the path planning is realized by executing the observation pose.

[0013] Furthermore, the continuous acquisition of environmental observation data by the environmental perception sensors mounted on the mobile robot, extraction of keyframe sequences from the corresponding environmental observation data, and calculation of the keyframe sequences using a sensor fusion algorithm to determine the pose estimation for each keyframe include:

[0014] The mobile robot continuously collects RGB image data, depth image data, and inertial measurement data during operation using RGB cameras, LiDAR, and IMU sensors. The collected data is packaged into a set of environmental observation data. After time synchronization and preprocessing of the environmental observation data, the corresponding keyframe sequence is extracted.

[0015] The environmental observation data of adjacent frames in the keyframe sequence are matched using a sensor fusion-based SLAM algorithm to determine the initial pose estimate and the corresponding pose covariance matrix for each keyframe.

[0016] Further, the step of constructing an initial voxel volume of a known 3D scene based on the keyframe sequence and the pose estimation, and simultaneously constructing an explicit uncertain voxel volume, includes:

[0017] The spatial coordinates of the 3D scene are hybrid-encoded according to the neural implicit method to determine the corresponding SDF network and color network. The SDF network is subjected to observation constraints by the keyframe sequence and the pose estimation. Based on the constrained SDF network and color network, the corresponding initial voxel volume is determined. The initial voxel volume is used to store the geometric symbol distance values ​​on the discrete 3D mesh.

[0018] In the aforementioned neural implicit approach, an explicit uncertainty voxel volume is introduced to represent the uncertainty of an unknown scenario, wherein:

[0019] During the initialization phase, the explicit uncertainty voxel volume is uniformly assigned a high value, indicating high uncertainty during the initialization phase.

[0020] As the keyframe sequence is observed, the volume of the explicit uncertainty voxel gradually decreases in the fully observed region, indicating that its uncertainty decreases as observation data is acquired.

[0021] Furthermore, the process of constructing a hierarchical target space within the mobile robot's operational range and sampling grid points on each layer to determine the corresponding candidate point set includes:

[0022] Based on the current initial voxel volume and the current explicit uncertainty voxel volume of the mobile robot, determine the three-dimensional bounding box of the corresponding robot's movable area;

[0023] The three-dimensional bounding box is discretized into multiple height layers along the height direction according to a preset fixed step size, and each height layer corresponds to an observation plane.

[0024] According to the preset spatial step size, each observation plane is sampled by a regular grid to determine the grid point set of each corresponding plane layer;

[0025] Multiple candidate orientation angles are generated for each grid point according to a preset angle step size, and the candidate orientation angles are merged into the grid point set to determine the corresponding candidate point set.

[0026] Further, the step of removing obstacle region candidate points from the candidate point set based on the initial voxel volume includes:

[0027] Unreachable candidate targets are identified based on the initial voxel volume;

[0028] Candidate points in the candidate point set are eliminated based on the unreachable candidate target.

[0029] Further, the step of performing uncertainty voxel screening and visibility determination on the candidate point set after the candidate point elimination based on the explicit uncertainty voxel volume, to determine the observable voxel set corresponding to each candidate point, includes:

[0030] For the candidate point set after the candidate point elimination, all voxels in the explicit uncertainty voxel volume are sorted in descending order according to their stored uncertainty values, and voxel screening is performed according to a preset voxel quantity threshold to determine the corresponding high uncertainty voxel set.

[0031] For each candidate point, a visible field of view is determined based on the set of high-uncertainty voxels to identify the observable voxels corresponding to each candidate point. The criteria for this visible field of view determination include:

[0032] Distance determination is used to determine whether each high uncertainty voxel is within the effective observation range of the sensor. If it is, the high uncertainty voxel is added to the set of observable voxels.

[0033] The occlusion determination is used to determine whether each high uncertainty voxel is occluded by an obstacle. If not, the high uncertainty voxel is added to the observable voxel set.

[0034] Further, the step of calculating the aggregated reconstruction uncertainty of each candidate point based on the observable voxel set, and calculating the improvement in localization uncertainty of each candidate point based on the pose estimation, includes:

[0035] Based on the set of observable voxels corresponding to each candidate point, the uncertainty value of each voxel in the set is summed in a weighted manner. The weight coefficient of the summation is dynamically adjusted according to at least one of the following factors: the distance between the voxel and the candidate point, the observation angle, and the scene structure features to which the voxel belongs.

[0036] Predict the pose covariance at the next moment based on the pose covariance matrix at the current moment, simulate new predictions at each candidate point, determine the updated pose covariance based on the Jacobian matrix of the observation model and the observation noise covariance matrix, and determine the corresponding improvement in positioning uncertainty based on the expected decrease between the pose covariance at the next moment and the updated pose covariance.

[0037] Secondly, this application provides a path planning device based on reconstruction uncertainty, comprising:

[0038] The front-end perception module is used to continuously collect environmental observation data through the environmental perception sensors mounted on the mobile robot, extract the key frame sequence of the corresponding environmental observation data, and calculate the pose estimate of each key frame according to the sensor fusion algorithm.

[0039] A 3D map construction module is used to construct an initial voxel volume of a known 3D scene based on the keyframe sequence and the pose estimation in a neural implicit manner, and at the same time construct an explicit uncertainty voxel volume, wherein the explicit uncertainty voxel volume is used to represent the uncertainty of the unknown scene.

[0040] The candidate point and high uncertainty voxel determination module is used to construct a hierarchical target space within the mobile robot's operational range, sample grid points on each layer plane, determine the corresponding candidate point set, remove candidate points in obstacle areas from the candidate point set based on the initial voxel volume, perform uncertainty voxel screening and visibility judgment on the candidate point set after the candidate point removal based on the explicit uncertainty voxel volume, determine the observable voxel set corresponding to each candidate point, calculate the reconstruction uncertainty aggregation amount of each candidate point based on the observable voxel set, and calculate the localization uncertainty improvement amount of each candidate point based on the pose estimation.

[0041] The path planning module is used to calculate the path cost of the mobile robot according to the path planning algorithm, and determine the corresponding unified planning objective function according to the path cost, the amount of reconstruction uncertainty aggregation, and the amount of localization uncertainty improvement. Based on the unified planning objective function, the module determines the next observation pose of the mobile robot and executes the observation pose to realize path planning.

[0042] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the path planning method based on reconstruction uncertainty.

[0043] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the path planning method based on reconstruction uncertainty.

[0044] Fifthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the described path planning method based on reconstruction uncertainty.

[0045] As can be seen from the above technical solution, this application provides a path planning method and apparatus based on reconstruction uncertainty. It continuously collects environmental data through environmental perception sensors mounted on a mobile robot, extracts corresponding environmental data keyframes and pose estimations for each frame, constructs an initial voxel volume, and introduces an explicit uncertainty voxel volume. Path candidate points are sampled within the mobile robot's operational range, and unreachable candidate points are eliminated from the candidate point set based on the initial voxel volume. For the eliminated candidate point set, uncertain voxels are screened and the visible field is determined using the explicit uncertainty voxel volume. After eliminating unvisible voxels, observable voxels for each candidate point are obtained. A planning objective function is constructed based on the observable voxels, pose estimation information, and the path cost of each candidate point to obtain the next observed pose, thus achieving path planning. This improves the efficiency of 3D map construction and the quality of path planning. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is one of the flowcharts illustrating the path planning method based on reconstruction uncertainty in the embodiments of this application;

[0048] Figure 2 This is a structural diagram of the path planning device based on reconstruction uncertainty in the embodiments of this application;

[0049] Figure 3 This is a schematic diagram of the structure of the electronic device in the embodiments of this application.

[0050] Figure label:

[0051] Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver storage unit 9144, antenna 9111, speaker 9131, microphone 9132. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] The acquisition, storage, use, and processing of data in this application all comply with relevant laws and regulations.

[0054] Given the rapid development of robotics technology, mobile robots have been widely used in various fields. However, the problem of optimal path planning and decision-making for mobile robots in unknown scenarios remains. This application provides a path planning method and apparatus based on reconstruction uncertainty. The method involves continuously collecting environmental data through environmental perception sensors mounted on the mobile robot, extracting corresponding keyframes and pose estimates for each frame, constructing an initial voxel volume, and introducing an explicit uncertainty voxel volume. Path candidate points are sampled within the mobile robot's operational range, and unreachable candidate points are eliminated based on the initial voxel volume. For the eliminated candidate point set, uncertainty voxels are screened and the visible area is determined using the explicit uncertainty voxel volume. After eliminating unvisible voxels, observable voxels for each candidate point are obtained. A planning objective function is constructed based on the observable voxels, pose estimation information, and path costs for each candidate point to obtain the next observed pose, thus achieving path planning. This improves the efficiency of 3D map construction and the quality of path planning.

[0055] To improve the efficiency of 3D map creation and the quality of path planning, this application provides an embodiment of a path planning method based on reconstruction uncertainty, see [link to embodiment]. Figure 1 The path planning method based on reconstruction uncertainty specifically includes the following:

[0056] In SLAM scenarios for mobile robots, how can we automatically select the next observation pose while building a map in real time, thereby minimizing map reconstruction uncertainty, obtaining a high-quality 3D map, and simultaneously reducing robot localization uncertainty, ensuring navigation and planning stability, and balancing path cost and collision safety while meeting the motion constraints of the ground mobile platform? This embodiment addresses these problems by implementing the following steps:

[0057] Step S101: Continuously collect environmental observation data through the environmental perception sensor on the mobile robot, extract the key frame sequence of the corresponding environmental observation data, and calculate the pose estimate of each key frame according to the sensor fusion algorithm.

[0058] Optionally, in this embodiment, front-end sensing and odometer information acquisition are performed first.

[0059] The mobile robot continuously collects environmental observation data during operation using sensors such as RGB-D cameras, LiDAR, and IMU. The front-end perception and odometry module performs time synchronization and preprocessing of the sensor data to obtain keyframe sequences and preliminary pose estimations. Specifically, this includes:

[0060] Keyframes are selected from the sensor data to obtain the keyframe observation set {It, Dt, zt}, where:

[0061] It represents the RGB image of frame t;

[0062] Dt represents the depth map;

[0063] zt represents the inertial observations such as the IMU corresponding to this frame;

[0064] A coarse pose estimate for each keyframe is calculated using laser odometry, visual odometry, or multi-sensor fusion algorithms. And the corresponding pose covariance matrix Σt, which serves as the input for the backend SLAM and active planning.

[0065] For the purposes of this embodiment:

[0066] Through the keyframe extraction mechanism, massive, continuous raw sensor data streams are condensed into a discrete keyframe sequence with high information density and low redundancy.

[0067] By fusing visual / depth and inertial data, sensor fusion algorithms can overcome the limitations of single sensors. For example, when texture loss or rapid motion causes image blurring, IMU data can provide reliable short-term motion predictions; when the IMU accumulates errors due to drift, visual observations can provide absolute position correction. This complementarity makes the output keyframe pose estimation both smooth and accurate. Simultaneously, the covariance matrix defines the reliability of the mobile robot's localization, providing a quantitative input for the subsequent "localization uncertainty improvement," which is a prerequisite for realizing localization uncertainty-driven planning.

[0068] This step, through data acquisition, filtering, and fusion calculation processes, outputs reliable initial state and observation data containing uncertainty information for the entire active planning system for neural SLAM.

[0069] Step S102: Based on the keyframe sequence and the pose estimation, construct the initial voxel volume of the known 3D scene in a neural implicit manner, and at the same time construct the explicit uncertainty voxel volume, wherein the explicit uncertainty voxel volume is used to represent the uncertainty of the unknown scene.

[0070] Optionally, in this embodiment, a neural 3D map is constructed with an uncertain voxel volume.

[0071] Based on keyframe observations and preliminary pose, the neural 3D mapping and uncertainty estimation module utilizes neural implicit representations to construct a 3D scene map and explicit uncertainty voxel volumes. Specifically, this includes:

[0072] Initial voxel volume V s :

[0073] First, a hybrid neural representation (multi-resolution hashing) is used to establish a continuous implicit model of the 3D scene. Specifically, a hybrid encoding method is used to encode the spatial 3D coordinates to construct an SDF network f. s and color network f c The SDF voxel volume V is obtained on a discrete three-dimensional mesh. s Preferred, the hybrid encoding method is multi-resolution hash grid encoding.

[0074] Explicit uncertainty voxel volume V σ :

[0075] In parallel with geometric reconstruction, an uncertain voxel volume V with the same resolution and spatial extent as the SDF voxel grid is explicitly constructed and maintained by introducing an uncertain network into the neural network or estimating the prediction residuals during training. σ .

[0076] Uncertain voxel volume V σ As an independent data structure, each voxel unit stores a scalar value used to quantify the uncertainty (or variance) of the geometric information (SDF value) at that location.

[0077] Among them, the volume of the uncertain voxel V σ In the initialization phase, a relatively high value is taken overall. As keyframe observations are continuously added and neural SLAM optimization is performed, the uncertainty gradually decreases only in areas that are fully observed and well-fitted, while unobserved or poorly fitted areas maintain a high level of uncertainty.

[0078] Ultimately, the SDF voxel volume (V s ) and uncertain voxel volume (V σ Together, they form a cognitive map of the scene. That is, the SDF value corresponding to each voxel center x is d(x) = V. s(x)The reconstruction uncertainty is σ²(x) = V σ(x)

[0079] The former uses d(x) to indicate whether a point x in space is "free" (positive value), "surface" (zero value), or "occupied" (negative value); the latter uses σ²(x) to quantify the confidence level of this judgment. Both are linked through the same spatial index, enabling the system to perform compound queries such as "obtaining the geometric information and reliability of a point" or "finding all free space voxels with high uncertainty." This explicit, structured representation provides a directly and efficiently computed foundational data field for the subsequent active planning module.

[0080] This embodiment introduces explicit uncertainty, through V σ This answers the question of where to go next, namely, the higher the uncertainty of a region, the greater its information value. The planning algorithm no longer needs to make decisions based on a coarse "known-unknown" boundary (frontier), but can precisely target those geometrically explored areas (potentially free space in Vs) but cognitively unreliable areas (in V...). σ The "grey area" (with a high median) makes the robot's observation behavior more purposeful and information-efficient, enabling it to directly strengthen weak points in the map.

[0081] In regions where dynamic objects appear only briefly or where low texture makes feature tracking difficult, neural SLAM optimization is often unstable, particularly in areas where V... σ The values ​​will remain high due to weak observation constraints. This is equivalent to the system automatically marking these unreliable areas. The active planning module can guide the robot to re-observe from other angles to eliminate ambiguity, or in actual navigation, path planning can actively avoid these highly uncertain areas to improve safety, even if they are geometrically marked as "free".

[0082] In summary, the explicit uncertainty voxel volume constructed in this step is a key bridge for deeply coupling the perception capabilities of neural SLAM with the robot's proactive decision-making capabilities.

[0083] Step S103: Construct a hierarchical target space within the mobile robot's operational range, sample grid points on each layer plane, determine the corresponding candidate point set, eliminate candidate points in obstacle areas from the candidate point set based on the initial voxel volume, perform uncertainty voxel screening and visual field judgment on the candidate point set after the candidate point elimination based on the explicit uncertainty voxel volume, determine the observable voxel set corresponding to each candidate point, calculate the reconstruction uncertainty aggregation amount of each candidate point based on the observable voxel set, and calculate the localization uncertainty improvement amount of each candidate point based on the pose estimation;

[0084] Optionally, in this embodiment, target construction and candidate point target generation are performed. This step generates a set of candidate points for the next moving target of the mobile robot, based on the voxel construction described above.

[0085] To adapt to the motion model of mobile robots, which combines "two-dimensional planar motion + viewpoint rotation," and to fully utilize the structural information of the three-dimensional scene, this embodiment employs a hierarchical target space approach to construct a set of candidate observation poses, specifically including:

[0086] Based on the current SDF voxel volume V s and uncertain voxel volume V σ Within the robot's operational range, a three-dimensional bounding box region is defined, and along the height direction, it is discretized into L height layers with a fixed step size Δh=0.5. Each height layer corresponds to an observation plane.

[0087] On the plane of each height level, the plane position is regularly sampled with a preset spatial step size Δp = 1m to obtain the vertex set on the regular mesh. For the l-th height level and the u-th and v-th mesh point, the corresponding target pose is denoted as:

[0088]

[0089] in For ground coordinates, Let be the robot's orientation angle, which can be discretized in the range [0, 2π) with a fixed angular step size to obtain several candidate orientations;

[0090] All height layers, all grid points, and their respective candidate orientations are uniformly denoted as a multi-layered target space:

[0091]

[0092] in, This represents the set of candidate points consisting of all candidate observation poses; Indicates the i-th candidate observation pose; This represents the total number of candidate observation poses.

[0093] This hierarchical target space forms a regular three-dimensional grid structure, with grid vertices serving as candidate points, ensuring a spatial interval of approximately 1m between each candidate target point, thereby covering the entire task space with a limited number of candidate points;

[0094] Next, by querying the SDF voxel volume V s Candidate targets located inside obstacles or in obviously inaccessible areas are eliminated, and only candidate targets located in free space that meet the robot's kinematics and occupation constraints are retained as evaluation objects for subsequent active planning.

[0095] It is understood that the above fixed step size, spatial step size, and angular step size are all examples for ease of understanding in this embodiment, and replacing them with other data will not affect the implementation of this embodiment.

[0096] Optionally, in this embodiment, high-uncertainty voxel screening and visible area determination are performed. After obstacle removal, the remaining set of candidate points is located in free space, and the potential of each candidate point to reduce reconstruction uncertainty needs to be evaluated.

[0097] To reduce computational load and highlight high-value areas, this embodiment aggregates and evaluates voxels with high uncertainty, and combines sensor field-of-view limitations to screen observable voxels. Specifically, this includes:

[0098] In the uncertain voxel volume V σ In the middle, according to the uncertainty σ²(x) sorted from largest to smallest, the top k=200 voxels are selected to form a set of voxels with high uncertainty, as shown below:

[0099]

[0100] Next, the set of voxels with high uncertainty will be screened:

[0101] Distance constraint: For each candidate target ∈ S g Determine each high-uncertainty voxel x j Does x satisfy? j arrive The distance falls within the sensor's effective observation range [r] min , r max [Inside (e.g., r)] min =0.5 meters, r max =5 meters).

[0102] Unobstructed constraint: along Point to x j In the ray direction, the SDF value d is always greater than the robot's safe radius r, meaning there are no obstacles obstructing the view.

[0103] Voxels that meet the above conditions are added to the observable voxel set corresponding to the candidate target. The purpose is to measure the potential gain of the candidate target in terms of reconstruction uncertainty. This step combines sensor physical characteristics with scene geometry information to ensure that the evaluated observational gains are consistent with actual sensing capabilities and to avoid ineffective planning for invisible or inaccessible areas.

[0104] Optionally, in this embodiment, the amount of reconstruction uncertainty aggregation and the amount of positioning uncertainty improvement are calculated.

[0105] Based on the judgment of the visible field, for each candidate target Calculate the corresponding aggregation of reconstruction uncertainty and improvement of positioning uncertainty to characterize the potential benefits of the target in map reconstruction and positioning.

[0106] Reconstructing the uncertainty aggregation volume is the core revenue metric driving the robot to navigate to unmapped areas. For candidate targets... Define the uncertainty aggregation quantity U of its observable high uncertainty voxels. map for:

[0107]

[0108] Where w j The voxel weights can be a constant of 1, or weighted according to factors such as the distance between the voxel and the target point, and the viewing angle, to emphasize high uncertainty regions that are close to the target or have a greater geometric contribution.

[0109] The improvement in positioning uncertainty should consider not only reducing map uncertainty, but also the improvement of the robot's own positioning accuracy by observation.

[0110] The pose covariance matrix maintained by the SLAM backend at the current moment is Σ. t Without new observations, its propagation is as follows:

[0111]

[0112] Among them, F t Let Q be the state transition Jacobian matrix. t Let be the process noise covariance matrix.

[0113] If the robot is in the candidate target If a new observation zt+1(i) is obtained at point i, then the updated covariance under the information filtering framework can be approximately expressed as:

[0114]

[0115] Among them, H i For the observation model Jacobian matrix, R i To observe the noise covariance matrix.

[0116] Define the improvement amount U for positioning uncertainty loc for:

[0117]

[0118] That is, in candidate targets The expected decrease in pose covariance trace after observation at the target location. In practical implementation, an approximate information content index can be constructed using information such as the number of observable feature points at the candidate target location, disparity distribution, and observation baseline, to estimate the pose covariance trace. .

[0119] This reflects the contribution of the observation point to improving positioning stability, which is especially important in scenarios where loop closure is lacking or motion accumulation errors are large.

[0120] Understandably, this step ensures the safety and accessibility of candidate points through stratified sampling and obstacle removal; it focuses on the areas with the greatest reconstruction value through high-uncertainty voxel screening and visual field judgment; and it achieves a unified consideration of map quality and positioning accuracy through the quantification of reconstruction and positioning uncertainties. This step enables the mobile robot to not only actively explore unknown areas in each decision-making step but also to take into account the stability of the positioning system, thereby achieving more efficient and robust autonomous mapping and navigation in complex environments.

[0121] Step S104: Calculate the path cost of the mobile robot according to the path planning algorithm, and determine the corresponding unified planning objective function according to the path cost, the amount of reconstruction uncertainty aggregation, and the amount of localization uncertainty improvement. Determine the next observation pose of the mobile robot according to the unified planning objective function, and execute the observation pose to realize path planning.

[0122] Optionally, in this embodiment, path cost calculation and E-RRT path planning are performed.

[0123] Based on the aforementioned revenue metrics, the path cost from the current pose to the candidate target is constructed by further combining path length and turning cost, and the improved E-RRT algorithm is used for path planning on the SDF map.

[0124] Specifically, in the three-dimensional SDF voxel volume V s In the environment described, starting from the current robot state st, and with candidate targets... If the destination is the target, the E-RRT algorithm is used for sampled path planning; if the planning fails, the candidate target is considered unreachable and is removed from the candidate set.

[0125] For a successfully planned candidate objective, its corresponding path is denoted as {s}. j {(i)} Define path cost. for:

[0126]

[0127] Among them, L trans L is the total length of the path translation. rot λ represents the cumulative rotation angle or maximum steering angle along the path. rot This is a turning cost weight used to reflect the increased cost of mobile robots when turning in narrow passages or at large angles.

[0128] The improvements of E-RRT over traditional RRT include: prioritizing the attempt of straight-line connections from tree nodes to the target point in each iteration, introducing multi-point expansion between sampling points and the nearest tree node to accelerate tree growth, and using parallel computing to accelerate nearest neighbor search, thereby improving planning efficiency in high-dimensional target spaces.

[0129] Optionally, in this embodiment, the construction of the objective function and the selection of the optimal objective are unified.

[0130] In summary, the aggregation amount U of the above reconstruction uncertainty map Improvement in positioning uncertainty U loc and path cost C path In this embodiment, a unified multi-objective programming function is constructed to comprehensively evaluate candidate objectives and select the optimal objective.

[0131] Unified Programming Objective Function Defined as:

[0132]

[0133] α, β, and γ are positive weighting coefficients, which are used to adjust the degree of attention paid to the reduction of reconstruction uncertainty, the reduction of positioning uncertainty, and the path cost, respectively.

[0134] From the candidate target set S g The target pose with the largest objective function value is selected as the next observation pose:

[0135]

[0136] And adopt the corresponding planned path {s j This serves as the robot's current motion path.

[0137] After completing the optimal target selection and path planning, the robot performs motion according to the planned path and continuously updates the map and the volume of uncertain voxels during the execution process.

[0138] 1) The robot follows the path {s} j The system executes control commands segment by segment, collecting new sensor data in real time during the process, and completing obstacle detection and local obstacle avoidance.

[0139] 2) When the preset keyframe triggering conditions are met, new observations are added to the observation database. The neural SLAM backend is used to jointly optimize the scene implicit representation and pose graph, and the SDF voxel volume V is updated. s Explicit uncertainty voxel volume V σ and pose covariance Σt ;

[0140] 3) When it is detected that the robot has reached the vicinity of the target, the planned path has failed, or there is a risk of collision, the active planning process is re-triggered, returning to the step "Target Construction and Candidate Point Target Generation", and the above process is executed iteratively until the high-precision 3D mapping and stable positioning task of the predetermined area is completed.

[0141] In the aforementioned embodiments, the triggering conditions for active planning are mainly as follows: when the robot reaches the destination or detects that the current planned path has failed, the steps from "target construction and candidate point target generation" to "optimal target selection" are re-executed to perform target search and path planning. The reason for triggering at two times is to simplify the system control logic and reduce the computational burden, so that the active planning module can work stably in the basic implementation.

[0142] In another application scenario of this embodiment, to address the issue of large reconstruction errors in paths during long-term operation and dynamic environments, the active planning triggering conditions in the "optimal target selection" step can be further optimized. For example, a more flexible triggering and local replanning mechanism can be introduced. During the robot's movement along the current path, the changes in the local SDF map and the volume of uncertain voxels can be monitored in real time. When significant changes are detected, such as a large area of ​​high uncertainty voxels appearing in the local region ahead, the appearance of new obstacles, or the occupation of the original free space, local replanning can be triggered in advance before reaching the original target point: the uncertainty aggregation and path cost are recalculated only within the affected local space window to generate a new collision-free local path to replace the remaining old path. This allows for timely bypassing of new obstacles or priority observation of newly emerging high uncertainty areas, improving the system's adaptability to dynamic scenarios.

[0143] This example demonstrates how this embodiment achieves real-time mapping while automatically selecting the next observation pose by explicitly constructing a rapidly queryable uncertain voxel volume in the neural implicit map and aggregating and evaluating the top-k high-uncertainty voxels in the candidate target space. This reduces map reconstruction uncertainty as quickly as possible, resulting in a high-quality 3D map. Simultaneously, the robot's own positioning uncertainty is reduced, ensuring the stability of navigation and planning, and balancing path cost and collision safety while satisfying the motion constraints of the ground mobile platform.

[0144] As described above, the path planning method based on reconstruction uncertainty provided in this application can continuously collect environmental data through environmental perception sensors mounted on a mobile robot, extract corresponding environmental data keyframes and pose estimations for each frame, construct an initial voxel volume, and introduce the construction of an explicit uncertainty voxel volume. Path candidate points are sampled within the mobile robot's active range, and unreachable candidate points in the candidate point set are eliminated based on the initial voxel volume. For the candidate point set after elimination, uncertain voxels are screened and the visible field is determined using the explicit uncertainty voxel volume. After eliminating unvisible voxels, observable voxels for each candidate point are obtained. Based on the observable voxels, pose estimation information, and the path cost of each candidate point, a planning objective function is constructed to obtain the next observed pose, thus achieving path planning. This improves the mapping efficiency of 3D maps and the quality of path planning.

[0145] In one embodiment of the path planning method based on reconstruction uncertainty in this application, it may further include the following:

[0146] Step S201: During operation, the RGB camera, LiDAR and IMU sensor on the mobile robot continuously collect RGB image data, depth image data and inertial measurement data, and package the collected data into a set of environmental observation data. After time synchronization and preprocessing of the environmental observation data, the corresponding key frame sequence is extracted.

[0147] Step S202: Match the environmental observation data of adjacent frames in the key frame sequence according to the sensor fusion-based SLAM algorithm to determine the initial pose estimate and the corresponding pose covariance matrix of each key frame.

[0148] Optionally, in this embodiment, the mobile robot perceives the environment through a heterogeneous sensor system. An RGB camera provides texture and color information, a LiDAR directly acquires high-precision 3D point cloud data, and an IMU outputs high-frequency angular velocity and acceleration information. The system performs hardware timestamp alignment and software-level interpolation on these three sensors to achieve millisecond-level synchronization. Preprocessing operations include distortion correction and feature enhancement of the RGB image, downsampling and outlier filtering of the LiDAR point cloud, and zero-bias correction of the IMU data. Based on this, the system extracts representative and informative keyframes from the continuous data stream according to a preset strategy (such as parallax changes exceeding a threshold, a significant decrease in the number of feature points, or the time interval reaching an upper limit). Each keyframe contains a complete set of time-aligned RGB images, a depth map, and an inertial measurement data packet.

[0149] Optionally, in this embodiment, a tightly coupled sensor fusion SLAM front-end algorithm is employed. This algorithm uses visual feature points, laser feature points (such as corner points and planar points), and IMU pre-integral quantities as unified constraints to construct a joint objective function based on graph optimization. By matching visual descriptors and laser features between adjacent keyframes and fusing inter-frame motion priors provided by the IMU, the algorithm iteratively solves for the optimal estimate of the robot pose. During the optimization process, the algorithm simultaneously calculates the curvature of the objective function at the Hessian matrix or Fisher information matrix level, thereby deriving the covariance matrix of the pose estimation for each keyframe. This matrix quantifies the uncertainty of the pose in each degree of freedom.

[0150] Through step S202, this embodiment provides a precise spatiotemporal reference with uncertainty measurement for all subsequent modules.

[0151] In one embodiment of the path planning method based on reconstruction uncertainty in this application, it may further include the following:

[0152] Step S301: Hybrid encoding of the spatial coordinates of the 3D scene according to the neural implicit method, determine the corresponding SDF network and color network, apply observation constraints to the SDF network through the keyframe sequence and the pose estimation, and determine the corresponding initial voxel volume based on the constrained SDF network and color network, wherein the initial voxel volume is used to store the geometric symbol distance values ​​on the discrete 3D mesh.

[0153] Step S302: In the neural implicit approach, an explicit uncertainty voxel volume is introduced to represent the uncertainty of the unknown scenario, wherein:

[0154] During the initialization phase, the explicit uncertainty voxel volume is uniformly assigned a high value, indicating high uncertainty during the initialization phase.

[0155] As the keyframe sequence is observed, the volume of the explicit uncertainty voxel gradually decreases in the fully observed region, indicating that its uncertainty decreases as observation data is acquired.

[0156] Optionally, in this embodiment, during the construction of the neural 3D map and the uncertain voxel volume, the spatial coordinates of the 3D scene are first hybrid-encoded, typically using a multi-resolution hash lattice encoding method to achieve an efficient and continuous representation of spatial geometry and texture. Using this encoding method, a symbolic distance function network and a color network are constructed respectively, where the symbolic distance function network is used to model the continuous symbolic distance field of the scene geometry, and the color network is used to model the scene appearance.

[0157] By utilizing keyframe sequences acquired from sensors and their corresponding pose estimates, observation constraints are imposed on the symbolic distance function network. Joint optimization is then performed through a neural SLAM backend to align the network's output geometry with the actual observation data, resulting in an optimized symbolic distance function network and color network. Based on the optimized network, an initial voxel volume is generated by sampling on a discrete 3D mesh. Each voxel in this volume stores its 3D geometric symbolic distance value, forming an explicit geometric map that can be used for collision detection and path planning.

[0158] Optionally, in this embodiment, to quantify the uncertainty in the reconstruction process, an explicit uncertainty voxel volume with the same resolution as the geometric voxel volume is introduced into the above-mentioned neural implicit representation, which is specifically used to represent the reconstruction confidence of different regions in the scene.

[0159] During the system initialization phase, all voxels of this uncertain voxel volume are assigned a uniform high uncertainty value to reflect the high uncertainty of the perception of the entire scene at the initial moment due to the lack of observation.

[0160] As the robot moves through the environment and continuously acquires keyframe observations, this data is fed into the neural SLAM backend for optimization, simultaneously driving the update of uncertain voxel volumes:

[0161] In regions that are well observed and have good geometric fit, the uncertainty value of voxels gradually decreases during the optimization process, indicating that the reconstruction quality and confidence level of the region are improved.

[0162] In regions that are not observed or have poor fitting, the uncertainty values ​​of voxels remain high or only slightly decrease, continuously characterizing the uncertainty state in these regions.

[0163] Through step S302, this embodiment successfully introduces explicit uncertainty voxel volume, providing a queryable and quantifiable uncertainty measurement basis for subsequent active planning, enabling the robot to identify and prioritize the exploration of high uncertainty areas, thereby systematically improving the integrity of mapping and the stability of localization.

[0164] In one embodiment of the path planning method based on reconstruction uncertainty in this application, it may further include the following:

[0165] Step S401: Based on the current initial voxel volume and the current explicit uncertainty voxel volume of the mobile robot, determine the three-dimensional bounding box of the corresponding robot's movable area;

[0166] Step S402: Discretize the three-dimensional bounding box into multiple height layers along the height direction according to a preset fixed step size, and each height layer corresponds to an observation plane;

[0167] Step S403: Perform regular grid sampling on each observation plane according to the preset spatial step size to determine the grid point set of each corresponding layer plane;

[0168] Step S404: Generate multiple candidate orientation angles for each grid point according to a preset angle step size, and merge the candidate orientation angles into the grid point set to determine the corresponding candidate point set.

[0169] Optionally, in this embodiment, the SDF voxel volume (Vs) and the explicit uncertainty voxel volume (V) constructed based on the current neural SLAM system are... σ The spatial boundaries of the robot's movable area are extracted. Specifically, by statistically analyzing the spatial extent of all voxels in Vs that belong to free space (SDF value greater than the safety threshold), the minimum and maximum coordinates in the X, Y, and Z directions are calculated to form an axis-aligned 3D bounding box (AABB). This bounding box represents the continuous spatial extent that the robot can safely enter in the currently known map, providing a spatial basis for subsequent structured sampling.

[0170] Optionally, in this embodiment, after obtaining the 3D bounding box, it is uniformly discretized along its height direction by a preset fixed step size Δh, dividing the entire bounding box into L height layers. For example, if Δh = 0.5 meters and the bounding box height is 2 meters, it can be divided into 4 height layers. Each height layer corresponds to a horizontal observation plane, and its height value is the reference plane height plus the layer index multiplied by the step size. This layered structure adapts to the motion characteristics of ground robots (mainly moving along the horizontal plane), while achieving the ability to observe the vertical scene structure through multi-layer coverage, making it particularly suitable for environments with multi-layered structures or undulating terrain.

[0171] Optionally, in this embodiment, for the observation plane corresponding to each height layer, a two-dimensional regular grid is sampled within the plane with a preset spatial step size Δp (e.g., 1 meter). Starting from the lower left corner of the plane, grid vertices are generated in the X and Y directions according to the step size, forming a set of grid points for that layer of the plane. Each grid point is represented by its two-dimensional coordinates (x_g, y_g), representing a potential ground target location. Regular sampling ensures the uniformity and coverage of the sampling points, avoiding the regional omissions or uneven sampling problems that may be caused by random sampling, and also provides a structured search space for subsequent path planning.

[0172] Optionally, in this embodiment, for each grid point, its azimuth degree of freedom is further discretized, generating multiple candidate orientation angles θ_g within the range [0, 2π) with a preset angle step size Δθ (e.g., π / 4 radians). Each "position-orientation" combination constitutes a complete candidate target pose s_g = (x_g, y_g, θ_g). All height layers, all grid points, and their corresponding candidate orientations are merged to form the final hierarchical target space candidate point set S_g. By introducing the orientation dimension, the system not only evaluates where to go but also where to look, thereby more precisely quantifying the differences in the benefits of different observation directions for reconstruction and localization.

[0173] Through step S404, this embodiment successfully constructs a hierarchical sampling space, providing a high-quality search decision space and laying a solid spatial foundation for the complete extraction of sampling points.

[0174] In one embodiment of the path planning method based on reconstruction uncertainty in this application, it may further include the following:

[0175] Step S501: Identify unreachable candidate targets based on the initial voxel volume;

[0176] Step S502: Eliminate candidate points in the candidate point set according to the unreachable candidate target.

[0177] Optionally, in this embodiment, unreachable candidate targets are identified based on the initial voxel volume.

[0178] Example 1: Unreachability Identification Based on SDF Threshold Judgment

[0179] After generating the candidate target set, the system queries the SDF voxel volume Vs to obtain the SDF value of each candidate point's location. If the SDF value is less than the preset safety radius (e.g., 0.3 meters), the point is determined to be located inside an obstacle or too close to an obstacle, and is therefore considered an unreachable candidate target, and is removed from the candidate set.

[0180] Example 2: Multi-point sampling judgment based on the occupancy state of neighboring voxels

[0181] This method not only checks the SDF value of the candidate point itself, but also performs SDF queries on multiple sampling points within a certain neighborhood (e.g., a cube centered on the candidate point with a side length of 0.5 meters). If the SDF value of any sampling point in the neighborhood is lower than a safety threshold, or if more than a certain proportion (e.g., 30%) of the sampling points are in an "occupied" state (SDF ≤ 0), then the candidate point is determined to be unreachable. This method is suitable for situations where the robot has a certain volume and can avoid misjudgments caused by SDF interpolation errors or local map noise.

[0182] Example 3: Accessibility Analysis Based on SDF Gradient and Local Curvature

[0183] Analyze the SDF gradient and local curvature information near the candidate point. If the SDF gradient magnitude in the area where the candidate point is located is too large (indicating proximity to obstacle edges), or the local curvature is high (indicating complex or narrow terrain), even if the SDF value is slightly higher than the safety threshold, it may still be judged as "high risk" or "unreachable". This embodiment is suitable for scenarios with higher requirements for navigation smoothness and safety, such as narrow corridors and environments with dense obstacles.

[0184] Through step S502, this embodiment uses a continuous geometric representation built based on neural SLAM to accurately reflect the environmental occupancy, ensuring that subsequent planning only considers safe and reachable candidate points, providing reliable input for path search and benefit evaluation, and avoiding guiding the robot into collision risk areas.

[0185] In one embodiment of the path planning method based on reconstruction uncertainty in this application, it may further include the following:

[0186] Step S601: For the candidate point set after the candidate point elimination, sort all voxels in the explicit uncertainty voxel volume in descending order according to their stored uncertainty values, and perform voxel screening according to the preset voxel quantity threshold to determine the corresponding high uncertainty voxel set.

[0187] Step S602: For each candidate point, perform a field-of-view determination based on the set of high-uncertainty voxels to determine the set of observable voxels corresponding to each candidate point. The criteria for the field-of-view determination include:

[0188] Distance determination is used to determine whether each high uncertainty voxel is within the effective observation range of the sensor. If it is, the high uncertainty voxel is added to the set of observable voxels.

[0189] The occlusion determination is used to determine whether each high uncertainty voxel is occluded by an obstacle. If not, the high uncertainty voxel is added to the observable voxel set.

[0190] Optionally, in this embodiment, the system performs data filtering based on the explicit uncertainty voxel volume. First, it iterates through all voxels in Vσ, reading the uncertainty value σ stored in each voxel. 2 (x) represents the uncertainty measure for reconstructing the scene geometry or texture at that voxel location. Subsequently, all voxels are sorted according to σ. 2 The values ​​of (x) are sorted in descending order to obtain a voxel sequence ordered from highest to lowest uncertainty. Based on a preset voxel count threshold (e.g., k=200), the system selects the first k voxels from the beginning of the sorted sequence to form a high-uncertainty voxel set X. σThis filtering operation directly focuses on the areas with the worst reconstruction quality and the least certainty in the current map, reflecting the core principle of proactive planning: "prioritizing the exploration of the most unknown parts." By imposing a hard limit on the number of voxels, it effectively controls the scale of subsequent calculations and ensures the real-time nature of the method.

[0191] Optionally, in this embodiment, for each candidate target point after reachability filtering... The assessment evaluates which high-uncertainty voxels can be effectively observed from this point. This evaluation is accomplished through a visibility judgment, which comprises two sequentially executed sub-conditions:

[0192] Distance constraint: For each candidate target ∈Sg, determine each highly uncertain voxel x j Does x satisfy? j arrive The distance falls within the sensor's effective observation range [r] min , r max [Inside (e.g., r)] min =0.5 meters, r max =5 meters).

[0193] Unobstructed constraint: along Point to x j In the ray direction, the SDF value d is always greater than the robot's safe radius r, meaning there are no obstacles obstructing the view.

[0194] Only when a voxel with high uncertainty passes both the distance and unobstructedness criteria mentioned above will it be included in the set of observable voxels corresponding to the candidate point.

[0195] Through step S602, this embodiment successfully combines theoretically high-value voxels (high uncertainty) with actual perception constraints (sensor range and scene occlusion), truly reflecting the information gains that can be effectively obtained by observing at the candidate point, and avoiding assigning observation tasks to the robot that cannot be actually executed.

[0196] In one embodiment of the path planning method based on reconstruction uncertainty in this application, it may further include the following:

[0197] Step S701: Based on the set of observable voxels corresponding to each candidate point, perform a weighted summation of the uncertainty values ​​of each voxel in the set, wherein the weight coefficients of the weighted summation are dynamically adjusted according to at least one of the following factors: the distance between the voxel and the candidate point, the observation angle, and the scene structural features to which the voxel belongs.

[0198] Step S702: Predict the pose covariance at the next moment based on the pose covariance matrix at the current moment, simulate new predictions at each candidate point, determine the updated pose covariance based on the observation model Jacobian matrix and the observation noise covariance matrix, and determine the corresponding improvement amount of positioning uncertainty based on the expected decrease between the pose covariance at the next moment and the updated pose covariance.

[0199] Optionally, in this embodiment, after determining the observable voxel set for each candidate point, the system calculates the reconstruction uncertainty aggregation amount corresponding to the candidate point by weighted summation.

[0200] Specifically, each voxel in the candidate point observable set is dynamically adjusted based on one or more of the following factors:

[0201] Distance decay weight: The closer the Euclidean distance between the voxel and the candidate point, the higher the observation quality and the greater the contribution to map update. Therefore, the weight can be set as a distance decay function.

[0202] Observation angle weighting: The angle between a voxel and the sensor's optical axis affects the geometric accuracy and texture integrity of the observation. The weight can be set based on the cosine of the angle or the projected area, emphasizing the contribution of frontal observation.

[0203] Scene structure weights: If a voxel is located at an edge, corner, or weakly textured region, its reduced uncertainty is more crucial for understanding the overall scene. Its weight can be adaptively increased using scene structure features (such as SDF gradients and curvature).

[0204] The weighted summation formula is:

[0205]

[0206] The dynamic weighting mechanism enables the system to not only focus on the absolute value of uncertainty, but also to conduct a refined benefit assessment based on the geometric quality of the observation and the semantic importance of the scene, thereby guiding the robot to prioritize poses that can both observe high uncertainty areas and obtain high-quality observations.

[0207] Optionally, the improvement in localization uncertainty is used to measure the effect of observations at candidate points on the robot's own localization accuracy. The calculation process is based on an information filtering framework:

[0208] The pose covariance matrix maintained by the SLAM backend at the current moment is Σ. t Without new observations, its propagation is as follows:

[0209]

[0210] Among them, F t Let Q be the state transition Jacobian matrix.t Let be the process noise covariance matrix.

[0211] If the robot is in the candidate target If a new observation zt+1(i) is obtained at point i, then the updated covariance under the information filtering framework can be approximately expressed as:

[0212]

[0213] Among them, H i For the observation model Jacobian matrix, R i To observe the noise covariance matrix.

[0214] Define the improvement amount U for positioning uncertainty loc for:

[0215]

[0216] That is, in the candidate target The expected decrease in the pose covariance trace after observation at the target location. In practical implementation, an approximate information content index can be constructed using information such as the number of observable feature points at the candidate target location, disparity distribution, and observation baseline, which can be used to estimate the pose covariance trace. .

[0217] This reflects the contribution of the observation point to improving positioning stability, which is especially important in scenarios where loop closure is lacking or motion accumulation errors are large.

[0218] Through step S702, this embodiment successfully reduces the uncertainty of positioning by quantifying the degree to which different candidate points reduce positioning uncertainty, enabling active planning to pursue map integrity while taking into account the stability of the positioning system.

[0219] To improve the efficiency of 3D map creation and the quality of path planning, this application provides an embodiment of a path planning device based on reconstruction uncertainty for implementing all or part of the aforementioned path planning method based on reconstruction uncertainty. See [link to embodiment]. Figure 2 The path planning device based on reconstruction uncertainty specifically includes the following components:

[0220] The front-end perception module 10 is used to continuously collect environmental observation data through the environmental perception sensor mounted on the mobile robot, extract the key frame sequence of the corresponding environmental observation data, and calculate the pose estimate of each key frame according to the sensor fusion algorithm.

[0221] The 3D map construction module 20 is used to construct an initial voxel volume of a known 3D scene based on the keyframe sequence and the pose estimation in a neural implicit manner, and at the same time construct an explicit uncertainty voxel volume, wherein the explicit uncertainty voxel volume is used to represent the uncertainty of the unknown scene.

[0222] The candidate point and high uncertainty voxel determination module 30 is used to construct a hierarchical target space within the mobile robot's active range, sample grid points on each layer plane, determine the corresponding candidate point set, remove candidate points in obstacle areas from the candidate point set based on the initial voxel volume, perform uncertainty voxel screening and field of view judgment on the candidate point set after the candidate point removal based on the explicit uncertainty voxel volume, determine the observable voxel set corresponding to each candidate point, calculate the reconstruction uncertainty aggregation amount of each candidate point based on the observable voxel set, and calculate the localization uncertainty improvement amount of each candidate point based on the pose estimation.

[0223] The path planning module 40 is used to calculate the path cost of the mobile robot according to the path planning algorithm, and determine the corresponding unified planning objective function according to the path cost, the reconstruction uncertainty aggregation amount and the localization uncertainty improvement amount, determine the next observation pose of the mobile robot according to the unified planning objective function, and execute the observation pose to realize path planning.

[0224] As described above, the path planning device based on reconstruction uncertainty provided in this application embodiment can continuously collect environmental data through environmental perception sensors mounted on a mobile robot, extract corresponding environmental data keyframes and pose estimations for each frame, construct an initial voxel volume, and introduce an explicit uncertainty voxel volume; sample path candidate points within the mobile robot's activity range, and eliminate unreachable candidate points from the candidate point set based on the initial voxel volume; for the candidate point set after elimination, perform uncertainty voxel screening and visibility judgment through the explicit uncertainty voxel volume; after eliminating unvisible voxels, obtain the observable voxel of each candidate point; construct a planning objective function based on the observable voxel, pose estimation information, and path cost of each candidate point to obtain the next observed pose and realize path planning, thereby improving the mapping efficiency of 3D maps and the quality of path planning.

[0225] From a hardware perspective, in order to improve the efficiency of 3D map creation and the quality of path planning, this application provides an embodiment of an electronic device for implementing all or part of the path planning method based on reconstruction uncertainty. The electronic device specifically includes the following components:

[0226] The system comprises a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to transmit information between the path planning method based on reconstruction uncertainty and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the path planning method based on reconstruction uncertainty in the previous embodiments, and the contents of these embodiments are incorporated herein, with repeated details not repeated.

[0227] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0228] In practical applications, the path planning method based on reconstruction uncertainty can be partially executed on the electronic device side as described above, or all operations can be completed in the client device. The choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed in the client device, the client device may further include a processor.

[0229] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0230] Figure 3 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 3 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 3 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0231] In one embodiment, the path planning method based on reconstruction uncertainty can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control:

[0232] Step S101: Continuously collect environmental observation data through the environmental perception sensor on the mobile robot, extract the key frame sequence of the corresponding environmental observation data, and calculate the pose estimate of each key frame according to the sensor fusion algorithm.

[0233] Step S102: Based on the keyframe sequence and the pose estimation, construct the initial voxel volume of the known 3D scene in a neural implicit manner, and at the same time construct the explicit uncertainty voxel volume, wherein the explicit uncertainty voxel volume is used to represent the uncertainty of the unknown scene.

[0234] Step S103: Construct a hierarchical target space within the mobile robot's operational range, sample grid points on each layer plane, determine the corresponding candidate point set, eliminate candidate points in obstacle areas from the candidate point set based on the initial voxel volume, perform uncertainty voxel screening and visual field judgment on the candidate point set after the candidate point elimination based on the explicit uncertainty voxel volume, determine the observable voxel set corresponding to each candidate point, calculate the reconstruction uncertainty aggregation amount of each candidate point based on the observable voxel set, and calculate the localization uncertainty improvement amount of each candidate point based on the pose estimation;

[0235] Step S104: Calculate the path cost of the mobile robot according to the path planning algorithm, and determine the corresponding unified planning objective function according to the path cost, the amount of reconstruction uncertainty aggregation, and the amount of localization uncertainty improvement. Determine the next observation pose of the mobile robot according to the unified planning objective function, and execute the observation pose to realize path planning.

[0236] As described above, the electronic device provided in this application embodiment continuously collects environmental data through an environmental perception sensor mounted on a mobile robot, extracts corresponding environmental data keyframes and pose estimations for each frame, constructs an initial voxel volume, and introduces an explicit uncertainty voxel volume. Path candidate points are sampled within the mobile robot's active range, and unreachable candidate points are eliminated from the candidate point set based on the initial voxel volume. For the eliminated candidate point set, uncertain voxels are screened and the visible field is determined using the explicit uncertainty voxel volume. After eliminating unvisible voxels, observable voxels for each candidate point are obtained. Based on the observable voxels, pose estimation information, and path costs for each candidate point, a planning objective function is constructed to obtain the next observed pose, thus achieving path planning. This improves the efficiency of 3D map construction and the quality of path planning.

[0237] In another implementation, the path planning method based on reconstruction uncertainty can be configured separately from the central processing unit 9100. For example, the path planning method based on reconstruction uncertainty can be configured as a chip connected to the central processing unit 9100, and the function of the path planning method based on reconstruction uncertainty can be implemented through the control of the central processing unit.

[0238] like Figure 3 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 3 All components shown; in addition, the electronic device 9600 may also include Figure 3 For components not shown, please refer to existing technologies.

[0239] like Figure 3 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0240] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0241] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0242] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0243] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device for communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0244] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 is coupled to the central processing unit 9100 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0245] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Furthermore, the audio processor 9130 is also coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored sound via the speaker 9131.

[0246] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the path planning method based on reconstruction uncertainty, where the execution subject is a server or client, as described in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the path planning method based on reconstruction uncertainty, where the execution subject is a server or client, as described in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0247] Step S101: Continuously collect environmental observation data through the environmental perception sensor on the mobile robot, extract the key frame sequence of the corresponding environmental observation data, and calculate the pose estimate of each key frame according to the sensor fusion algorithm.

[0248] Step S102: Based on the keyframe sequence and the pose estimation, construct the initial voxel volume of the known 3D scene in a neural implicit manner, and at the same time construct the explicit uncertainty voxel volume, wherein the explicit uncertainty voxel volume is used to represent the uncertainty of the unknown scene.

[0249] Step S103: Construct a hierarchical target space within the mobile robot's operational range, sample grid points on each layer plane, determine the corresponding candidate point set, eliminate candidate points in obstacle areas from the candidate point set based on the initial voxel volume, perform uncertainty voxel screening and visual field judgment on the candidate point set after the candidate point elimination based on the explicit uncertainty voxel volume, determine the observable voxel set corresponding to each candidate point, calculate the reconstruction uncertainty aggregation amount of each candidate point based on the observable voxel set, and calculate the localization uncertainty improvement amount of each candidate point based on the pose estimation;

[0250] Step S104: Calculate the path cost of the mobile robot according to the path planning algorithm, and determine the corresponding unified planning objective function according to the path cost, the amount of reconstruction uncertainty aggregation, and the amount of localization uncertainty improvement. Determine the next observation pose of the mobile robot according to the unified planning objective function, and execute the observation pose to realize path planning.

[0251] As described above, the computer-readable storage medium provided in this application embodiment continuously collects environmental data through an environmental perception sensor mounted on a mobile robot, extracts corresponding environmental data keyframes and pose estimations for each frame, constructs an initial voxel volume, and introduces an explicit uncertainty voxel volume. Path candidate points are sampled within the mobile robot's operational range, and unreachable candidate points are eliminated from the candidate point set based on the initial voxel volume. For the eliminated candidate point set, uncertain voxels are screened and the visible area is determined using the explicit uncertainty voxel volume. After eliminating unvisible voxels, observable voxels for each candidate point are obtained. Based on the observable voxels, pose estimation information, and path costs for each candidate point, a planning objective function is constructed to obtain the next observed pose, thus achieving path planning. This improves the mapping efficiency and path planning quality of 3D maps.

[0252] Embodiments of this application also provide a computer program product capable of implementing all steps of the path planning method based on reconstruction uncertainty, where the execution subject is a server or client, as described in the above embodiments. When executed by a processor, this computer program / instruction implements the steps of the path planning method based on reconstruction uncertainty. For example, the computer program / instruction implements the following steps:

[0253] Step S101: Continuously collect environmental observation data through the environmental perception sensor on the mobile robot, extract the key frame sequence of the corresponding environmental observation data, and calculate the pose estimate of each key frame according to the sensor fusion algorithm.

[0254] Step S102: Based on the keyframe sequence and the pose estimation, construct the initial voxel volume of the known 3D scene in a neural implicit manner, and at the same time construct the explicit uncertainty voxel volume, wherein the explicit uncertainty voxel volume is used to represent the uncertainty of the unknown scene.

[0255] Step S103: Construct a hierarchical target space within the mobile robot's operational range, sample grid points on each layer plane, determine the corresponding candidate point set, eliminate candidate points in obstacle areas from the candidate point set based on the initial voxel volume, perform uncertainty voxel screening and visual field judgment on the candidate point set after the candidate point elimination based on the explicit uncertainty voxel volume, determine the observable voxel set corresponding to each candidate point, calculate the reconstruction uncertainty aggregation amount of each candidate point based on the observable voxel set, and calculate the localization uncertainty improvement amount of each candidate point based on the pose estimation;

[0256] Step S104: Calculate the path cost of the mobile robot according to the path planning algorithm, and determine the corresponding unified planning objective function according to the path cost, the amount of reconstruction uncertainty aggregation, and the amount of localization uncertainty improvement. Determine the next observation pose of the mobile robot according to the unified planning objective function, and execute the observation pose to realize path planning.

[0257] As described above, the computer program product provided in this application continuously collects environmental data through an environmental perception sensor mounted on a mobile robot, extracts corresponding environmental data keyframes and pose estimations for each frame, constructs an initial voxel volume, and introduces the construction of an explicit uncertainty voxel volume. Path candidate points are sampled within the mobile robot's operational range, and unreachable candidate points in the candidate point set are eliminated based on the initial voxel volume. For the candidate point set after elimination, uncertain voxels are screened and the visible field is determined using the explicit uncertainty voxel volume. After eliminating unvisible voxels, observable voxels for each candidate point are obtained. Based on the observable voxels, pose estimation information, and the path cost of each candidate point, a planning objective function is constructed to obtain the next observed pose, thus achieving path planning. This improves the mapping efficiency and path planning quality of 3D maps.

[0258] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0259] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0260] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0261] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0262] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A path planning method based on reconstruction uncertainty, characterized in that, The method includes: The mobile robot continuously collects environmental observation data using environmental perception sensors and extracts key frame sequences of the corresponding environmental observation data. The key frame sequences are then calculated using a sensor fusion algorithm to determine the pose estimate for each key frame. Based on the keyframe sequence and the pose estimation, an initial voxel volume of the known 3D scene is constructed in a neural implicit manner, and an explicit uncertainty voxel volume is constructed at the same time, wherein the explicit uncertainty voxel volume is used to represent the uncertainty of the unknown scene. A hierarchical target space is constructed within the mobile robot's operational range, and grid points are sampled on each layer plane to determine the corresponding candidate point set. The candidate point set is then used to eliminate candidate points in obstacle areas based on the initial voxel volume. The candidate point set after the elimination of candidate points is then used to screen uncertain voxels and determine the visible field based on the explicit uncertainty voxel volume to determine the observable voxel set corresponding to each candidate point. The reconstruction uncertainty aggregation amount of each candidate point is calculated based on the observable voxel set, and the localization uncertainty improvement amount of each candidate point is calculated based on the pose estimation. The path cost of the mobile robot is calculated according to the path planning algorithm. Based on the path cost, the amount of reconstruction uncertainty aggregation, and the amount of localization uncertainty improvement, a corresponding unified planning objective function is determined. The next observation pose of the mobile robot is determined according to the unified planning objective function, and the path planning is realized by executing the observation pose.

2. The path planning method based on reconstruction uncertainty according to claim 1, characterized in that, The process involves continuously collecting environmental observation data using environmental perception sensors mounted on the mobile robot, extracting keyframe sequences from the corresponding environmental observation data, and calculating the pose estimate for each keyframe using a sensor fusion algorithm. This includes: The mobile robot continuously collects RGB image data, depth image data, and inertial measurement data during operation using RGB cameras, LiDAR, and IMU sensors. The collected data is packaged into a set of environmental observation data. After time synchronization and preprocessing of the environmental observation data, the corresponding keyframe sequence is extracted. The environmental observation data of adjacent frames in the keyframe sequence are matched using a sensor fusion-based SLAM algorithm to determine the initial pose estimate and the corresponding pose covariance matrix for each keyframe.

3. The path planning method based on reconstruction uncertainty according to claim 1, characterized in that, The process of constructing an initial voxel volume for a known 3D scene based on the keyframe sequence and pose estimation, using a neural implicit approach, and simultaneously constructing an explicit uncertain voxel volume, includes: The spatial coordinates of the 3D scene are hybrid-encoded according to the neural implicit method to determine the corresponding SDF network and color network. The SDF network is subjected to observation constraints by the keyframe sequence and the pose estimation. Based on the constrained SDF network and color network, the corresponding initial voxel volume is determined. The initial voxel volume is used to store the geometric symbol distance values ​​on the discrete 3D mesh. In the aforementioned neural implicit approach, an explicit uncertainty voxel volume is introduced to represent the uncertainty of an unknown scenario, wherein: During the initialization phase, the explicit uncertainty voxel volume is uniformly assigned a high value, indicating high uncertainty during the initialization phase. As the keyframe sequence is observed, the volume of the explicit uncertainty voxel gradually decreases in the fully observed region, indicating that its uncertainty decreases as observation data is acquired.

4. The path planning method based on reconstruction uncertainty according to claim 1, characterized in that, The step of constructing a hierarchical target space within the mobile robot's operational range and sampling grid points on each layer to determine the corresponding candidate point set includes: Based on the current initial voxel volume and the current explicit uncertainty voxel volume of the mobile robot, determine the three-dimensional bounding box of the corresponding robot's movable area; The three-dimensional bounding box is discretized into multiple height layers along the height direction according to a preset fixed step size, and each height layer corresponds to an observation plane. According to the preset spatial step size, each observation plane is sampled by a regular grid to determine the grid point set of each corresponding plane layer; Multiple candidate orientation angles are generated for each grid point according to a preset angle step size, and the candidate orientation angles are merged into the grid point set to determine the corresponding candidate point set.

5. The path planning method based on reconstruction uncertainty according to claim 1, characterized in that, The step of eliminating candidate points for obstacle regions from the candidate point set based on the initial voxel volume includes: Unreachable candidate targets are identified based on the initial voxel volume; Candidate points in the candidate point set are eliminated based on the unreachable candidate target.

6. The path planning method based on reconstruction uncertainty according to claim 1, characterized in that, The step of performing uncertainty voxel screening and visibility determination on the candidate point set after the candidate point elimination based on the explicit uncertainty voxel volume, to determine the observable voxel set corresponding to each candidate point, includes: For the candidate point set after the candidate point elimination, all voxels in the explicit uncertainty voxel volume are sorted in descending order according to their stored uncertainty values, and voxel screening is performed according to a preset voxel quantity threshold to determine the corresponding high uncertainty voxel set. For each candidate point, a visible field of view is determined based on the set of high-uncertainty voxels to identify the observable voxels corresponding to each candidate point. The criteria for this visible field of view determination include: Distance determination is used to determine whether each high uncertainty voxel is within the effective observation range of the sensor. If it is, the high uncertainty voxel is added to the set of observable voxels. The occlusion determination is used to determine whether each high uncertainty voxel is occluded by an obstacle. If not, the high uncertainty voxel is added to the observable voxel set.

7. The path planning method based on reconstruction uncertainty according to claim 1, characterized in that, The calculation of the reconstruction uncertainty aggregation amount for each candidate point based on the observable voxel set, and the calculation of the localization uncertainty improvement amount for each candidate point based on the pose estimation, include: Based on the set of observable voxels corresponding to each candidate point, the uncertainty value of each voxel in the set is summed in a weighted manner. The weight coefficient of the summation is dynamically adjusted according to at least one of the following factors: the distance between the voxel and the candidate point, the observation angle, and the scene structure features to which the voxel belongs. Predict the pose covariance at the next moment based on the pose covariance matrix at the current moment, simulate new predictions at each candidate point, determine the updated pose covariance based on the Jacobian matrix of the observation model and the observation noise covariance matrix, and determine the corresponding improvement in positioning uncertainty based on the expected decrease between the pose covariance at the next moment and the updated pose covariance.

8. A path planning device based on reconstruction uncertainty, characterized in that, The device includes: The front-end perception module is used to continuously collect environmental observation data through the environmental perception sensors mounted on the mobile robot, extract the key frame sequence of the corresponding environmental observation data, and calculate the pose estimate of each key frame according to the sensor fusion algorithm. A 3D map construction module is used to construct an initial voxel volume of a known 3D scene based on the keyframe sequence and the pose estimation in a neural implicit manner, and at the same time construct an explicit uncertainty voxel volume, wherein the explicit uncertainty voxel volume is used to represent the uncertainty of the unknown scene. The candidate point and high uncertainty voxel determination module is used to construct a hierarchical target space within the mobile robot's operational range, sample grid points on each layer plane, determine the corresponding candidate point set, remove candidate points in obstacle areas from the candidate point set based on the initial voxel volume, perform uncertainty voxel screening and visibility judgment on the candidate point set after the candidate point removal based on the explicit uncertainty voxel volume, determine the observable voxel set corresponding to each candidate point, calculate the reconstruction uncertainty aggregation amount of each candidate point based on the observable voxel set, and calculate the localization uncertainty improvement amount of each candidate point based on the pose estimation. The path planning module is used to calculate the path cost of the mobile robot according to the path planning algorithm, and determine the corresponding unified planning objective function according to the path cost, the amount of reconstruction uncertainty aggregation, and the amount of localization uncertainty improvement. Based on the unified planning objective function, the module determines the next observation pose of the mobile robot and executes the observation pose to realize path planning.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the path planning method based on reconstruction uncertainty as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the path planning method based on reconstruction uncertainty as described in any one of claims 1 to 7.

Images