Outdoor cleaning robot and positioning state evaluation and control triggering method thereof

Abstract

The application discloses an outdoor cleaning robot and a positioning state evaluation and control triggering method thereof, and the outdoor cleaning robot comprises the following steps: collecting original data of multiple sensors of the outdoor cleaning robot in real time; calculating four types of features, i.e., coverage, geometric quality, multi-source consistency residual and motion consistency, based on the original data of the multiple sensors and a prior global high-precision point cloud map, and constructing a comprehensive feature vector; inputting the comprehensive feature vector into a pre-trained ordered multi-classification gradient boosting decision tree positioning state evaluation model to output a positioning state grade; and finally generating a control triggering code for controlling behaviors of the outdoor cleaning robot according to the grade and related feature values and a preset rule. The application realizes accurate evaluation of the positioning state and automatically generates a hierarchical control triggering code, thereby effectively improving the operation safety and running reliability of the outdoor cleaning robot in a complex outdoor environment.

Description

Technical Field

[0001] This invention belongs to the field of environmental perception and status monitoring technology for unmanned robots, specifically relating to an outdoor cleaning robot and its positioning status assessment and control triggering method. Background Technology

[0002] In recent years, with the rapid development of autonomous driving and robotics technologies, outdoor cleaning robots have been widely used in large outdoor scenarios such as urban roads, industrial parks, and residential areas. To achieve precise navigation and operation, these robots typically rely on prior high-precision global point cloud maps for localization. However, the outdoor environment is dynamic and complex, presenting various challenges such as satellite signal obstruction, interference from dynamic obstacles, and sensor data fluctuations. These challenges can lead to varying degrees of performance degradation or even complete failure of the robot's localization system. If the localization is inaccurate and the system fails to detect it in time, it can easily cause serious operational safety accidents such as the robot deviating from its path, colliding, or being unable to return autonomously.

[0003] Currently, the industry generally uses simple rules based on threshold judgments from a single sensor to monitor robot positioning status. For example, when the signal-to-noise ratio of the Global Navigation Satellite System (GNSS) signal falls below a certain fixed value, or when the inlier rate of LiDAR point cloud matching suddenly drops, it is considered a positioning anomaly. However, this type of method has significant limitations: First, it relies on human experience to set thresholds, which is poorly adaptable to complex and changing outdoor scenarios and prone to false alarms or missed alarms; second, judgment based on a single indicator cannot comprehensively and accurately reflect the overall health status of the positioning system. For example, in scenarios where the GNSS signal is temporarily lost but LiDAR positioning remains reliable, simple rules may make overly conservative misjudgments, leading to unnecessary operational interruptions.

[0004] Furthermore, some studies have attempted to use traditional machine learning or deep learning models to process sensor data, but these solutions often suffer from problems such as dependence on large-scale labeled data, poor model interpretability, and low inference efficiency on computationally limited vehicle platforms. More importantly, most existing methods focus on binary judgments of "whether it is abnormal" and lack fine-grained distinctions between ordered levels of positioning status such as "poor," "edge," and "normal," thus failing to support the generation of differentiated control trigger codes that match the status level. Summary of the Invention

[0005] To address the shortcomings of existing technologies and achieve refined and robust assessment of the positioning status of outdoor cleaning robots by integrating information from multiple sensor sources, and to automatically generate control trigger codes to trigger corresponding safety strategies, thereby improving the operational safety and reliability of outdoor cleaning robots in complex environments, this invention adopts the following technical solution:

[0006] A method for robot localization state assessment and control triggering includes the following steps:

[0007] Step S1: Real-time acquisition of multi-source heterogeneous raw data, including real-time acquisition of robot motion information, positioning information, and point cloud frames acquired by the robot in real time;

[0008] Step S2: Using the original data and the global point cloud map, calculate the coverage features of the point cloud frame in the global point cloud map, the geometric quality features of the point cloud frame matching the global point cloud map, the multi-source consistency residual features of the robot motion information and localization information, and the motion consistency features of the robot motion information and localization information, and construct comprehensive features based on the calculated features.

[0009] Step S3: Based on comprehensive features, a pre-trained localization status evaluation model is used to generate the robot's current localization status level;

[0010] Step S4: Based on the positioning status level and the specific values ​​of specific features in the comprehensive features, and according to the preset mapping rules, trigger the control of robot behavior.

[0011] Furthermore, in step S2, the coverage features include coverage and inter-frame overlap rate. Coverage is the degree to which the point cloud map is covered by the robot's effective detection range; inter-frame overlap rate is the crossover ratio of the point cloud map areas covered by adjacent point cloud frames. Coverage and inter-frame overlap rate can effectively reflect the completeness of the LiDAR's perception of the environmental structure. High coverage and high overlap rate usually mean strong positioning stability and are suitable for positioning quality assessment in dynamic environments.

[0012] Further, in step S1, the positioning information includes position coordinates and positioning covariance matrix; in step S2, the multi-source consistency residual features include consistency residual features based on the robot itself and consistency residual features based on the robot's positioning information; the consistency residual features based on the robot itself are residual features obtained by integrating the robot's estimated pose at adjacent time points and the robot's motion information within adjacent time intervals; the consistency residual features based on the robot's positioning information are residual features obtained by integrating the translation vector of the current estimated pose, the measured value of the positioning information at the current time, and the positioning covariance matrix at the current time. The multi-source consistency residual features, by comparing the pose differences between different sensor sources, enable the robot to detect whether there is sensor drift or anomaly, thereby improving the ability to judge positioning consistency.

[0013] Furthermore, in step S1, the raw data also includes the current estimated pose, which is obtained by matching the point cloud frame with the pre-constructed global point cloud map; in step S2, the geometric quality features include the point cloud matching score of the localization robot, the matching inlier rate, the quantile of the matching residual, and the trace of the current pose estimation covariance matrix. The geometric quality features directly reflect the accuracy and confidence of the point cloud matching and are an important basis for judging whether the localization is reliable.

[0014] Furthermore, in step S3, the positioning status assessment model adopts an ordered multi-classification gradient boosting decision tree model. During training, a weighted ordinal loss function is constructed based on multi-level binary classification of samples, and monotonicity constraints are applied to the comprehensive features of the input to ensure the logical consistency between the model output and the features. The constraint features include coverage features, geometric quality features, multi-source consistency residual features, and motion consistency features. Among them, the quantiles and the trace of the current pose estimation covariance matrix in the motion consistency features, multi-source consistency residual features, and geometric quality features are monotonically positively correlated with the output risk of the positioning status assessment model. The matching inlier rate in the coverage features and geometric quality features are monotonically negatively correlated with the output risk of the positioning status assessment model. The ordered multi-classification model can better handle the orderliness of the positioning status and avoid unreasonable jumps in the level judgment of ordinary classification models, while the monotonicity constraints enhance the interpretability and reliability of the model.

[0015] Further, in step S1, the motion information includes the robot's three-axis acceleration and three-axis angular velocity, linear velocity, and angular velocity. In step S2, the motion consistency features include linear velocity inconsistency and angular velocity inconsistency. Linear velocity inconsistency is the inconsistency between the robot's measured linear velocity and the linear velocity between adjacent poses obtained from the positioning. Angular velocity inconsistency is the inconsistency between the robot's measured angular velocity and the angular velocity between adjacent poses obtained from the positioning. The motion consistency features are used to detect whether there is a deviation between the robot's actual motion and the positioning estimate, and are suitable for judging whether abnormal motion states such as slippage or collision have occurred.

[0016] Furthermore, in step S1, the positioning information also includes the solution state; in step S2, based on the consistency residual characteristics of the robot positioning information, gating is performed through the solution state, and the consistency residual characteristics based on the robot positioning information are used only when the solution state is a fixed solution or a floating-point solution.

[0017] Furthermore, in step S4, four positioning status levels and six control trigger codes are set: normal, slow, paused, repositioning, safe stop, and remote assistance. Based on the positioning status level and related characteristics output by the positioning status evaluation model, the generation rules for the control trigger codes are as follows:

[0018] If the positioning status is at the first level, a normal control trigger code is generated to indicate that the robot's positioning status is normal.

[0019] If the positioning status is level 2 and the coverage feature is less than the coverage decision threshold, a slow control trigger code is generated to indicate that the robot should decelerate; if the positioning status is level 2 and the coverage feature is greater than or equal to the coverage decision threshold, a normal control trigger code is generated to indicate that the robot's positioning status is normal.

[0020] If the positioning status is level 3 and the solution status is a fixed solution, and the consistency residual characteristic based on the robot positioning information is greater than the residual characteristic decision threshold, a pause control trigger code is generated to indicate that the robot should pause and wait; if the positioning status is level 3 and the solution status is not a fixed solution, or the consistency residual characteristic based on the robot positioning information is less than or equal to the residual characteristic decision threshold, a repositioning control trigger code is generated to indicate that the robot should perform repositioning.

[0021] If the positioning status is at level four and the robot's measured linear velocity is greater than the safe speed, a control trigger code for safe stop is generated to indicate that the robot should brake urgently.

[0022] If the positioning status is at level four and the robot's measured linear velocity is less than or equal to the safe speed, a remote assistance control trigger code is generated to indicate that the robot requests remote intervention.

[0023] By using the positioning status level output by the positioning status evaluation model and the specific values ​​of related features, the robot's positioning status is bound to the control trigger code, which defines the robot's specific behavior and realizes closed-loop control from perception to decision-making, thereby improving the robot's perception ability, safety, and intelligence level in complex environments.

[0024] A method for assessing the positioning status and triggering control of an outdoor cleaning robot is provided, which is then applied to an outdoor cleaning robot.

[0025] An outdoor cleaning robot includes a robot body and a lidar, an inertial measurement unit, a wheel speedometer, and a global navigation satellite system receiver mounted on the robot body. The lidar acquires point cloud frames, the inertial measurement unit and wheel speedometer acquire the robot's motion information, and the global navigation satellite system receiver acquires positioning information. According to the robot positioning state assessment and control triggering method, the robot sequentially performs real-time acquisition of multi-source heterogeneous raw data, construction of comprehensive features, generation of positioning state level, and triggering of outdoor cleaning robot behavior control.

[0026] The advantages and beneficial effects of this invention are as follows:

[0027] This invention is applicable to the positioning tasks of outdoor cleaning robots in complex outdoor environments. Based on raw data from multiple sensors and a priori high-precision global point cloud maps, it constructs a comprehensive feature vector by calculating four types of features: coverage, geometric quality, multi-source consistency residuals, and motion consistency. It employs an efficient, reliable, and monotonic ordered multi-class gradient boosting decision tree model to achieve multi-level accurate and robust evaluation of the positioning status, replacing fixed rules that rely on human experience and overcoming the shortcomings of the single threshold judgment method. A hierarchical control trigger code is generated based on the positioning status level and feature values, realizing a closed loop from state perception to safety control, significantly improving the autonomy and safety of the unmanned system. The sensors used in this invention are all conventionally configured, and the model training and deployment costs are low, making it highly practical, economical, and portable, suitable for large-scale application. Attached Figure Description

[0028] Figure 1 This is a flowchart of a method according to an embodiment of the present invention.

[0029] Figure 2 This is a flowchart illustrating the training process of the positioning status evaluation model in this embodiment of the invention.

[0030] Figure 3 This is a flowchart of the method for applying the positioning status evaluation model and the method for generating control trigger codes in an embodiment of the present invention. Detailed Implementation

[0031] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0032] like Figure 1 As shown, this embodiment proposes a method for evaluating the positioning status and triggering control of an outdoor cleaning robot. The outdoor cleaning robot performs its tasks in a priori global high-precision point cloud map and is equipped with at least a lidar, an inertial measurement unit, a wheel speedometer, and a global navigation satellite system receiver. It is also able to read the current pose output by the positioning node. The method specifically includes the following steps:

[0033] Step S1: Collect raw data from multiple heterogeneous sources in real time from lidar, inertial measurement unit, wheel speedometer, global navigation satellite system receiver and positioning node.

[0034] The raw data includes the current time. Point cloud frames obtained from lidar scanning Triaxial acceleration from the inertial measurement unit With triaxial angular velocity Linear velocity from wheel speed gauge With angular velocity Position coordinates from a Global Navigation Satellite System receiver Solution status and the location covariance matrix Point cloud frames obtained from scanning using lidar based on the positioning node With pre-built global high-precision point cloud map The current estimated pose output after matching .

[0035] By collecting raw data, outdoor cleaning robots can capture motion and positioning information from different dimensions, avoiding positioning loss caused by the failure of a single sensor, enhancing their ability to perceive environmental changes, and improving the robustness of the system.

[0036] Step S2: Based on the original data and the global high-precision point cloud map, calculate the coverage features, geometric quality features, multi-source consistency residual features, and motion consistency features, and construct a comprehensive feature vector.

[0037] Specifically, the calculation of coverage features includes coverage With inter-frame overlap rate .

[0038] Coverage The calculation method is as follows:

[0039]

[0040] in, This indicates the effective detection radius of the lidar. Within the effective forward field of view, a pre-built global high-precision point cloud map Total number of grid cells in the grid; It is an indicator function, when the first Point cloud frames obtained by scanning a grid with LiDAR The value is 1 if at least one point in the array is hit, and 0 otherwise.

[0041] Inter-frame overlap rate The calculation method is as follows:

[0042]

[0043] in, and Each represents the current time. Compared to the previous moment The raster occupancy state is determined by the point cloud frame. and Calculated separately; and These represent the intersection and union of the grid, respectively, when the denominator is 0. Take 0.

[0044] Coverage With inter-frame overlap rate It can effectively reflect the completeness of LiDAR's perception of environmental structures. High coverage and high overlap rate usually mean strong positioning stability, making it suitable for positioning quality assessment in dynamic environments.

[0045] Specifically, the geometric quality features are output by the localization nodes, including: point cloud matching scores. Matching in-point rate 95th percentile of the matched residuals and the current pose estimation covariance matrix traces .

[0046] Geometric quality features directly reflect the accuracy and confidence of point cloud matching and are an important basis for judging whether the localization is reliable.

[0047] Specifically, multi-source consistency residual characteristics include odometer consistency residuals. Consistency residual with Global Navigation Satellite System .

[0048] Odometer Consistency Residual The calculation method is as follows:

[0049]

[0050] in, It is a moment The estimated pose output by the localization node. The linear velocity is measured by the wheel speed gauge. angular velocity Triaxial acceleration measured by the inertial measurement unit Triaxial angular velocity In time interval Odometry pose obtained by internal integration For Li Qun To its corresponding Lie algebra The mapping, It is a 2-norm.

[0051] Global Navigation Satellite System Consistency Residual The calculation method is as follows:

[0052]

[0053] in, yes The translation vector, It is a global navigation satellite system receiver at a certain time The measured value, It is a global navigation satellite system receiver at a certain time The provided positioning covariance matrix.

[0054] Among them, the consistency residual of the global navigation satellite system The effectiveness of the features is determined by the solution status of the Global Navigation Satellite System receiver. Gating is performed only when When adopted Otherwise, Set to a predefined invalid flag value. In this embodiment, the predefined invalid flag value is taken as... ;in, This indicates that the solution state of the Global Navigation Satellite System receiver is a fixed solution. This indicates that the solution state of the Global Navigation Satellite System receiver is a floating-point solution.

[0055] Multi-source consistency residual features enable outdoor cleaning robots to detect sensor drift or anomalies by comparing pose differences between different sensor sources, thus improving their ability to judge positioning consistency.

[0056] Specifically, the calculation of motion consistency characteristics includes linear velocity inconsistency. Inconsistent with angular velocity .

[0057] Inconsistent linear velocity The calculation method is as follows:

[0058]

[0059] in, It is the linear velocity measured by the wheel speed gauge; The adjacent poses are output by the localization node. and The calculated linear velocity.

[0060] Inconsistent angular velocity The calculation method is as follows:

[0061]

[0062] in, It is the angular velocity measured by the inertial measurement unit; The adjacent poses are output by the localization node. and The calculated angular velocity.

[0063] Motion consistency features are used to detect whether there is a deviation between the actual motion of an outdoor cleaning robot and its positioning estimate. They are suitable for determining whether abnormal motion states such as slippage or collision have occurred.

[0064] After calculating the coverage feature, geometric quality feature, multi-source consistency residual feature, and motion consistency feature respectively, these feature values ​​from different dimensions and with different physical meanings are concatenated into a unified comprehensive feature vector in a predetermined order.

[0065] The comprehensive feature vector can be represented as:

[0066]

[0067] The integrated feature vector completes information fusion and unified modeling, breaking down the barriers between different sensors and data sources. It integrates information reflecting environmental structure, the confidence level within the positioning algorithm, and the cross-validation results between sensors, providing a global perspective for the positioning status assessment model, thereby achieving accurate assessment of the positioning status.

[0068] Step S3: Input the comprehensive feature vector into the pre-trained positioning status evaluation model. The positioning status evaluation model is a gradient boosting decision tree model that adopts an ordered multi-classification strategy. The model outputs the current positioning status level of the outdoor cleaning robot.

[0069] The training process of the localization state assessment model is as follows: Figure 2 As shown, by minimizing the following weighted ordinal loss function To complete:

[0070]

[0071] in, It is the total number of training samples; It is the number of levels in an ordered classification; It is an indicator function, when the sample The true label is greater than the level The value is 1 if the condition is met, otherwise it is 0. It is the model on the sample In the The original output scores on a binary task; It is the Sigmoid activation function; It is the first The weight coefficients of each binary task.

[0072] Specifically, during the training process of the localization state evaluation model, monotonicity constraints are imposed on the input features to ensure logical consistency between the model output and the features, thus constraining the features. , , , , , The output risk of the model is monotonically positively correlated; constraint features , , It is monotonically negatively correlated with the output risk of the model.

[0073] Ordered multi-class classification models can better handle the ordered nature of location states, avoiding unreasonable jumps in ranking judgments that are common in ordinary classification models. Monotonicity constraints enhance the interpretability and reliability of the model.

[0074] Step S4: Based on the positioning status level and the specific values ​​of specific features in the comprehensive feature vector, and according to the preset mapping rules, generate control trigger codes for controlling the behavior of the outdoor cleaning robot.

[0075] In embodiments of the present invention, such as Figure 3 As shown, the positioning status level The control trigger code set is The positioning status level output by the positioning status assessment model Based on the relevant characteristics, the rules for generating the trigger code are as follows:

[0076] like Generate control trigger code This indicates that the outdoor cleaning robot's positioning status is normal;

[0077] like and Generate control trigger code This indicates that the outdoor cleaning robot should slow down; otherwise... At that time, generate control trigger codes. This indicates that the outdoor cleaning robot's positioning status is normal;

[0078] like and Generate control trigger code This indicates that the outdoor cleaning robot should pause and wait; otherwise, when At that time, generate control trigger codes. This indicates that the outdoor cleaning robot should perform repositioning.

[0079] like and Generate control trigger code This indicates that the outdoor cleaning robot should apply an emergency stop.

[0080] like and Generate control trigger code This indicates that the outdoor cleaning robot is requesting remote intervention;

[0081] in, , , , All of these are preset decision thresholds.

[0082] The positioning status level output by the positioning status assessment model The system binds the positioning status of the outdoor cleaning robot to the control trigger code, along with the specific values ​​of related features, thus defining the robot's specific behavior. This achieves closed-loop control from perception to decision-making, enhancing the robot's perception capabilities, safety, and intelligence in complex environments.

[0083] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for robot localization state assessment and control triggering, characterized in that... Includes the following steps: Step S1: Real-time acquisition of multi-source heterogeneous raw data, including real-time acquisition of robot motion information, positioning information, and point cloud frames acquired by the robot in real time; motion information includes the robot's three-axis acceleration and three-axis angular velocity, linear velocity and angular velocity; positioning information includes position coordinates, positioning covariance matrix and solution status; Step S2: Using the original data and the point cloud map, calculate the coverage features of the point cloud frame in the point cloud map, the geometric quality features of the matching between the point cloud frame and the point cloud map, the multi-source consistency residual features of the robot motion information and the localization information, and the motion consistency features of the robot motion information and the localization information, and construct a comprehensive feature based on the calculated features. Multi-source consistent residual features include consistent residual features based on the robot itself and consistent residual features based on robot positioning information; The residual feature based on the robot's own consistency is the residual feature obtained by integrating the robot's pose estimation at adjacent time points and the robot's motion information within adjacent time intervals. The consistent residual feature based on robot localization information is a residual feature obtained based on the translation vector of the current estimated pose, the measured value of the localization information at the current time, and the localization covariance matrix at the current time. It is gating through the solution state, and the consistent residual feature based on robot localization information is only used when the solution state is a fixed solution or a floating-point solution. Motion consistency features include linear velocity inconsistency and angular velocity inconsistency; linear velocity inconsistency is the inconsistency between the linear velocity measured by the robot itself and the linear velocity between adjacent poses obtained from localization; angular velocity inconsistency is the inconsistency between the angular velocity measured by the robot itself and the angular velocity between adjacent poses obtained from localization. Step S3: Based on comprehensive features, use the positioning status evaluation model to generate the robot's current positioning status level; Step S4: Based on the positioning status level and comprehensive characteristics, and according to the preset mapping rules, trigger the control of robot behavior; set four positioning status levels and six control trigger codes: normal, slow, pause, repositioning, safe stop, and remote assistance. The generation rules for the control trigger codes are as follows, based on the positioning status level and related characteristics output by the positioning status evaluation model: If the positioning status is at the first level, a normal control trigger code is generated to indicate that the robot's positioning status is normal. If the positioning status is level 2 and the coverage feature is less than the coverage decision threshold, a slow control trigger code is generated to indicate that the robot should decelerate; if the positioning status is level 2 and the coverage feature is greater than or equal to the coverage decision threshold, a normal control trigger code is generated to indicate that the robot's positioning status is normal. If the positioning status is level 3 and the solution status is a fixed solution, and the consistency residual characteristic based on the robot positioning information is greater than the residual characteristic decision threshold, a pause control trigger code is generated to indicate that the robot should pause and wait; if the positioning status is level 3 and the solution status is not a fixed solution, or the consistency residual characteristic based on the robot positioning information is less than or equal to the residual characteristic decision threshold, a repositioning control trigger code is generated to indicate that the robot should perform repositioning. If the positioning status is at level four and the robot's measured linear velocity is greater than the safe speed, a control trigger code for safe stop is generated to indicate that the robot should brake urgently. If the positioning status is at level four and the robot's measured linear velocity is less than or equal to the safe speed, a remote assistance control trigger code is generated to indicate that the robot requests remote intervention.

2. The robot localization state evaluation and control triggering method according to claim 1, characterized in that: In step S2, the coverage features include coverage and inter-frame overlap rate. Coverage is the extent to which the point cloud map is covered by the robot's effective detection range. The inter-frame overlap rate is the crossover ratio of the point cloud map regions covered by adjacent point cloud frames.

3. The robot positioning state evaluation and control triggering method according to claim 1, characterized in that: In step S1, the raw data also includes the current estimated pose, which is obtained by matching point cloud frames with point cloud maps; in step S2, the geometric quality features include the point cloud matching score of the localization robot, the matching inlier rate, the quantile of the matching residual, and the trace of the current pose estimation covariance matrix.

4. The robot positioning state evaluation and control triggering method according to claim 3, characterized in that: In step S3, the localization status evaluation model adopts an ordered multi-classification strategy gradient boosting decision tree model. During the training process, a weighted ordinal loss function is constructed based on the multi-level binary classification of samples, and a monotonicity constraint is applied to the comprehensive features of the input. The constraint features include coverage features, geometric quality features, multi-source consistency residual features, and motion consistency features. Among these, the quantiles of the motion consistency features, multi-source consistency residual features, and geometric quality features, as well as the trace of the current pose estimation covariance matrix, are monotonically positively correlated with the output risk of the positioning state assessment model. The matching interior point rate in the constraint feature coverage feature and geometric quality feature is monotonically negatively correlated with the output risk of the positioning status assessment model.

5. A method for assessing the positioning status and triggering control of an outdoor cleaning robot, characterized in that: The robot positioning status assessment and control triggering method according to any one of claims 1 to 4 is used in an outdoor cleaning robot.

6. An outdoor cleaning robot, comprising a robot body and a lidar, an inertial measurement unit, a wheel speedometer, and a global navigation satellite system receiver mounted on the robot body, characterized in that, The robot acquires point cloud frames through the lidar, motion information through the inertial measurement unit and wheel speed meter, and positioning information through the global navigation satellite system receiver. According to any one of claims 1 to 4, the robot positioning status assessment and control triggering method sequentially performs real-time acquisition of multi-source heterogeneous raw data, construction of comprehensive features, generation of positioning status level, and triggering of outdoor cleaning robot behavior control.

Images