CN-121979214-A - Self-adaptive navigation system and method for outdoor patrol robot

CN121979214ACN 121979214 ACN121979214 ACN 121979214ACN-121979214-A

Abstract

The invention discloses an adaptive navigation system and method for an outdoor patrol robot, and belongs to the technical field of robot navigation. The method aims at solving the problems of poor path adaptability, insufficient navigation stability of complex scenes and low intelligent degree of the existing patrol robot. The technical scheme includes that an original path is optimized through an angle constraint iterative smoothing algorithm, included angles of adjacent line segments are calculated and corrected, a path point is adjusted according to the maximum allowable rotation angle of a robot chassis, the path is smooth and matched with kinematic constraints through a double iteration termination condition, the path is segmented according to lane attribute consistency, the path segments are divided based on preset attribute vectors and trigger conditions, and finally the robot adjusts a navigation strategy in a self-adaptive mode according to each segment attribute. The invention improves the path tracking precision and the driving stability, enhances the adaptability of the robot to different outdoor scenes, does not need complex manual calibration, reduces the operation and maintenance cost, and can be widely applied to unmanned patrol in multiple scenes such as parks, factories and the like.

Inventors

ZHONG WEIMING
LIN XIBIAO
ZHAI JINGJING
HU XIANDA
HE HANHUI

Assignees

广州申迪智能科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260130

Claims (10)

1. An outdoor patrol robot adaptive navigation system, comprising: The lane attribute identification module is used for acquiring lane attribute information of each point on the global lane map where the robot is currently located in real time; The path segment dividing module is used for generating a global navigation path according to the global lane map and the start and stop points, and dividing the global navigation path into a plurality of continuous path segments, wherein lane attributes in each path segment keep consistent; the control strategy configuration module is used for dynamically configuring a preset control strategy and a corresponding control parameter set according to the matching of each path segment; the scene decision module is used for fusing lane attribute information, real-time environment perception information and robot state information of the current path segment and deciding a behavior mode which is required to be executed by the robot at present through multi-factor cost evaluation; And the execution control module is used for calling the corresponding motion planning controller to generate a motion control instruction of the robot chassis according to the decided behavior mode and the configured control strategy and parameters, and driving the execution.
2. The outdoor patrol robot-adaptive navigation system of claim 1, wherein the lane attribute-identifying module comprises: The map loading sub-module is used for reading lane points and attribute information thereof from an XML-format global lane map file and constructing a lane network map, wherein the lane point attribute information at least comprises lane line numbers, speed limit values, a sensor visible range, a narrow area level, a road surface type and a driving mode; the position matching sub-module is used for searching a lane point with the closest distance in the lane network diagram through a matching algorithm according to the real-time positioning information of the robot and taking the lane point as map attribute information of a current lane; The data fusion sub-module is used for sensing the real-time state of the current lane by combining the real-time sensor data and supplementing or correcting the map attribute information; and the attribute output sub-module is used for outputting final lane attribute information after the map attribute information and the real-time state of the lane are fused in a structured mode, wherein the final lane attribute comprises, but is not limited to, lane line numbers, speed limit values, laser radar visible ranges, depth camera visible ranges, narrow area grades, road surface types, driving modes and special scene attributes.
3. The outdoor patrol robot adaptive navigation system of claim 1, wherein the path segment dividing module specifically comprises: The global path generation sub-module is used for generating a global lane map and a starting point through A The algorithm or BFS algorithm generates a global navigation path; The path smoothing sub-module is used for carrying out smoothing optimization on the global path, and ensuring that the path corner does not exceed a preset threshold value through an iterative smoothing algorithm of angle constraint; and the path dividing sub-module traverses the smoothed global path and divides the path into a plurality of segments according to attribute consistency, and each segment corresponds to a driving mode and a control strategy.
4. The adaptive navigation system of the outdoor patrol robot according to claim 1, wherein a policy matching table is preset in the control policy configuration module, the policy matching table defines a mapping relationship from lane attribute combinations to control policies and parameters, and when the control policy configuration module works, the control policy configuration module queries attribute combinations corresponding to lane attribute information of a current path segment, retrieves matched control policy identifications and basic parameters from the policy matching table, and calculates final real-time control parameters in combination with a preset dynamic adjustment formula.
5. The self-adaptive navigation method of the outdoor patrol robot is characterized by comprising the following steps of: S1, acquiring lane attribute information of each point on a global lane map where a robot is currently located in real time through a lane attribute identification module, wherein the lane attribute information comprises path geometric features, environment state features and manual pre-configuration attributes; S2, generating a global navigation path according to a global lane map and a start and stop point through a path segment dividing module, and dividing the global navigation path into a plurality of continuous path segments, wherein lane attributes in each path segment keep consistent; s3, matching and dynamically configuring a preset control strategy and a corresponding control parameter set for each path segment through a control strategy configuration module; S4, fusing lane attribute information, real-time environment perception information and robot state information of the current path segment through a scene decision module, and deciding a behavior mode which is required to be executed by the robot at present through multi-factor cost evaluation; S5, calling a corresponding motion planning controller to generate a motion control instruction of the robot chassis through an execution control module according to the decided behavior mode and the configured control strategy and parameters, and driving the execution.
6. The method according to claim 5, wherein the step S1 is to find the closest lane point in the lane network map by a matching algorithm and use the closest lane point as the map attribute information of the current lane, specifically: obtaining real-time positioning information of robot ) Wherein Is the horizontal axis of the robot and is defined by the horizontal axis, In the form of the ordinate of the robot, Is the driving direction angle of the robot; traversing all lane points in a lane network graph Wherein Is the abscissa of the i-th lane point, Is the ordinate of the i-th lane point, The driving direction angle corresponding to the ith lane point; Calculating the distance between the robot and each lane point through an Euclidean distance formula: ; verifying the direction matching degree through a direction consistency formula: ; Wherein the method comprises the steps of Is the direction angle difference; Screening out The method comprises the steps that the lane points smaller than a preset direction threshold value are determined, attribute information corresponding to the lane point with the smallest d is used as map attribute information of a current lane, and a matching algorithm expression is as follows: ; Is a robot and the first The driving direction angle difference value of each lane point; A preset direction matching threshold value; Screening conditions The index corresponding to the minimum Euclidean distance in the lane points meeting the direction conditions is the matching result.
7. The method of claim 5, wherein the step S2 is based on a global lane map and start-stop points, by a The algorithm generates a global navigation path, specifically: abstracting global lane maps into graphs Wherein For the lane point set, each vertex E is a set of connected edges between lane points, and the edge weight is the actual passing distance between the lane points; Setting a navigation start point And endpoint Wherein Is a lane point corresponding to the initial position of the robot, Lane points corresponding to navigation target positions; Defining a heuristic function: ; Wherein% , ) The coordinates of the current lane point v are [ (] , ) Is the end point Coordinates of (c); A The path search formula of the algorithm is: wherein Representation A Algorithm function, path represents A Searching the obtained global path by an algorithm; By iteratively calculating a cost function for each node Wherein And screening out a path with the minimum cost as a global path for the actual cost from the starting point to the current node v.
8. The method according to claim 5, wherein in step S2 the global navigation path is divided into a plurality of consecutive path segments, wherein lane properties inside each path segment remain consistent, in particular: s21, defining a path point sequence, and setting an original global path point sequence generated preliminarily as Wherein The index of the path point is represented, Representing the total number of path points, Representing an i-th path point; s22, calculating the included angle between the adjacent path segments and the coordinate axis, and the included angle between the line segment from the ith-1 path point to the ith path point Wherein As a four-quadrant arctangent function, Is that And (3) with Is used for the difference value of the vertical coordinates, Is that And (3) with Is a horizontal coordinate difference value of (2); the line segment included angle from the ith path point to the (i+1) th path point is Wherein Is that And (3) with Is used for the difference value of the vertical coordinates, Is that And (3) with Is a horizontal coordinate difference value of (2); S23, calculating the angle difference value of the adjacent line segments and correcting the range: If (if) Pi is corrected to =2π- Ensuring that the angle difference is within the range of [0, pi ]; S24, if the angle is different > For the path point And (3) performing smooth adjustment: wherein For a preset maximum allowable rotation angle threshold of the robot chassis, Is that The average value position of two adjacent points is obtained by Adjusting to the position to reduce the path rotation angle; s25, continuously segmenting the path after smooth optimization according to the consistency of lane attributes, wherein the method is specifically realized as follows: let the smoothed sequence of path points be Each path point The corresponding lane attribute vector is The segmentation result is Each continuous path segment , ,..., The method comprises the following steps: ; Is the first A number of successive path segments of the path, For the path segment index to be used, To smooth the optimized jth path point, The starting waypoint index of the kth path segment, For the termination path point index of the kth path segment, Index of any two path points within the kth path segment, Index is Lane attribute vectors corresponding to the path points of (c), Index is The lane attribute vector corresponding to the path point, Lane attribute vectors representing any two points in the same path segment are completely consistent; s26, traversing the path point sequence, and triggering segmentation when any one of the following attribute changes, namely driving mode change, speed limit value change exceeding a preset threshold value, narrow area grade change and road surface type change, occurs.
9. The method according to claim 5, wherein in step S3, according to a preset control policy and a corresponding control parameter set that are matched and dynamically configured for each path segment, specifically: S31, constructing a path segment attribute feature vector. Extracting lane attribute information of the current path segment and quantizing the information into an n-dimensional feature vector Wherein each dimension represents an attribute, including path width level Gradient value Road adhesion coefficient Whether or not it is a gate region Whether or not it is a pedestrian mixing area ; S32, performing static matching based on the policy matching table. Querying a predefined policy matching table defining combinations of standard attribute features to a base control policy Basic parameter set Mapping relation of (c): by calculating the feature vector of the current path segment And each standard feature vector in the table Matching degree of (2) Selecting an item with highest matching degree, wherein the algorithm expression is as follows: ; Wherein, the Representing the index of the entry in the policy matching table, The degree of match score is calculated based on the degree of match score, For the best match index to be found, As the weight of the material to be weighed, For the similarity calculation function, The i-th dimension attribute value of the current path segment, The ith dimension attribute value of the kth standard vector in the strategy matching table is used as the ith dimension attribute value; and S33, carrying out dynamic parameter adjustment based on the real-time state. Generating a dynamic adjustment factor vector according to the real-time environment sensing information and the robot state information The basic parameter set is corrected by utilizing a preset dynamic adjustment formula to obtain a real-time control parameter set finally implemented The expression is: ; Wherein, the As a basis for the set of control parameters, Representing an element-by-element multiplication, For the adjustment factor mapping function, the m-dimensional adjustment factor α is mapped to the same adjustment coefficient vector as the parameter set P dimension.
10. The method according to claim 5, wherein the deciding the behavior pattern that the robot should currently execute in step S4 through multi-factor cost evaluation specifically comprises: Defining a multi-factor cost function: ; Where d is the candidate pattern of behavior, For the security penalty of the behavior pattern d, In order to be at the expense of time efficiency, In order for the rule to adhere to the cost, 、、 The weight of each cost is respectively; security cost: wherein For the actual distance of the robot from the nearest obstacle, For real-time safety distance, when < In the time-course of which the first and second contact surfaces, =1; Time efficiency cost: wherein For the expected time of normal travel to the target point, The estimated driving time after the behavior pattern d is adopted; Rule compliance costs If the behavior pattern d accords with the current path segment traffic rule =0, Violate rule =1。

Description

Self-adaptive navigation system and method for outdoor patrol robot Technical Field The invention relates to the technical field of intelligent mobile robot navigation and control, in particular to an outdoor patrol robot self-adaptive navigation system and method. Background With the rapid development of automation and artificial intelligence technologies, autonomous mobile robots (such as outdoor patrol robots, security inspection robots, cleaning robots, logistics distribution robots, etc.) are widely used in many closed or semi-closed outdoor scenes. These scenarios typically include industrial parks, scientific parks, large warehouse logistics centers, campuses, park roads, and the like. In these environments, one of the central capabilities of robots is to achieve stable, reliable, efficient autonomous navigation, i.e. planning a path from a starting point, perceiving the environment, avoiding obstacles and reaching a target point safely without relying on manual remote control. The existing robot navigation system is mostly based on the combination of global path planning and local obstacle avoidance, such as AAnd (3) carrying out path planning by using an algorithm and a Dijkstra algorithm, and carrying out real-time obstacle avoidance by combining a laser radar and a vision sensor. However, when dealing with complex and changeable outdoor scenes with rich structured information, there are still significant limitations: 1. The control strategy is single, and the scene adaptability is lacking, and most of the existing systems adopt a set of fixed control parameter sets (such as maximum speed, minimum turning radius, acceleration limit, safety distance threshold value and the like) and single decision logic in the whole task execution period. However, the path properties of the outdoor environments vary greatly. 2. The structured attributes of the paths are underutilized in that lanes or navigable paths in the real world carry rich prior attribute information that exceeds the simple binary division of "navigable" and "unviable". 3. Most existing navigation systems either implicit these attribute information in the map data or merely as an environmental context, fail to promote them as explicit "knowledge" that can be directly understood and utilized by the navigation decision core module. The robot cannot "know" that it is driving into a narrow area where deceleration is required, or a ramp where special control laws need to be activated, and its behavior lacks intelligence based on scene understanding. 4. The decision dimension is single, multi-factor fusion optimization is not realized, the local decision of the traditional system is often focused on instant obstacle avoidance, the cost function design is simpler, and the distance between the traditional system and an obstacle is mainly considered. In complex practical tasks, robots need to trade-off in multiple dimensions of safety (collision avoidance), efficiency (rapid arrival), rule compliance (compliance with traffic rules), energy economy, etc. The existing system lacks a decision mechanism capable of fusing multi-source information (path attribute, real-time perception and self state) and performing multi-objective cost evaluation, and is difficult to make optimal or near optimal behavior selection (such as parking waiting for pedestrians to pass through or bypassing at a low speed, precisely parking before stopping a line and slowly sliding to pass through) under a complex scene. Therefore, a new navigation scheme of the patrol robot needs to be designed to solve the above technical problems. Disclosure of Invention The invention aims to solve the technical problems of poor adaptability, low safety, low efficiency and the like caused by single control strategy and incapability of utilizing path attribute knowledge when an outdoor patrol robot navigates in a complex structural environment. In order to solve the technical problems, the embodiment of the invention provides the following technical scheme that the self-adaptive navigation system of the outdoor patrol robot comprises: the system comprises a lane attribute identification module, a lane attribute analysis module and a robot, wherein the lane attribute identification module is used for acquiring lane attribute information of each point on a global lane map where the robot is currently located in real time, and the lane attribute information comprises path geometric features, environment state features and manual pre-configuration attributes; The path segment dividing module is used for generating a global navigation path according to the global lane map and the start and stop points, and dividing the global navigation path into a plurality of continuous path segments, wherein lane attributes in each path segment keep consistent; The control strategy configuration module is used for dynamically configuring a preset control strategy and a corresponding control parameter set according