Environment fast autonomous exploration planning method and device suitable for dynamic restriction ground robot

Abstract

The application discloses a kind of suitable for the environment quick autonomous exploration planning method and device of dynamics limited ground robot, it is related to robot unknown environment autonomous exploration field.A kind of suitable for the environment quick autonomous exploration planning method and device of dynamics limited robot, the method includes: using rotatable mechanism loads depth camera, in the case without being limited by chassis direction, environment quick modeling is carried out in conjunction with robot chassis odometer information;On the basis of environment modeling, extract environment boundary, and establish the best viewpoint for each boundary;Establish the cost matrix of travel cost between each best viewpoint, solve global coverage path;Under the guidance of global coverage path, solve dynamics feasible path for robot control;Using the information gain of each viewpoint belonging to boundary establishes attractive force function, calculates and obtains the rotatable mechanism angle control quantity on robot smooth path, to complete the quick environment exploration not being limited by robot orientation.The application is in the autonomous exploration process of dynamics limited ground robot, using rotatable mechanism carries depth camera, decouples robot travel direction and sensor orientation, designs attractive force function of different boundary, completes environment quick modeling not being limited by robot orientation, can continuously and quickly complete autonomous exploration of complex environment.

Description

Technical Field

[0001] This invention relates to the field of autonomous exploration technology for ground robots, specifically to a method and apparatus for rapid autonomous exploration planning in environments with limited dynamics for ground robots. Background Technology

[0002] With the development of simultaneous localization and mapping (SMR) technology, mobile robots have acquired the ability to perceive and locate their environment in unfamiliar settings without relying on external sensors. This makes autonomous task execution possible, and autonomous exploration by robots is a hot research topic in the field of robotics. The problem that autonomous exploration addresses is completing the exploration of unknown environments with the least possible energy consumption and the shortest possible time. The autonomous exploration strategy is the most important factor influencing the efficiency of robot exploration.

[0003] Current autonomous exploration strategies for robots mainly fall into three categories: boundary-based exploration strategies, sampling-based strategies, and hybrid strategies.

[0004] Boundary-based exploration strategies refer to selecting the optimal boundary to guide the robot in exploring the unknown environment, considering factors such as the boundary size, the potential information gain the robot can obtain at that boundary, and the energy consumption of the robot traversing the boundary. Disadvantages of this approach include: calculating the boundary consumes significant computational resources, boundary updates are slow, and the method for calculating the information gain of the boundary is not intuitive.

[0005] Sampling-based exploration strategies involve uniformly or non-uniformly sampling the space to be explored, constructing a roadmap composed of viewpoints, calculating the potential information gain for each viewpoint, planning the path with the maximum information gain, and executing it. A drawback of this method is the potential for frequent robot switching between different paths, leading to decreased exploration efficiency.

[0006] The hybrid exploration strategy combines boundary-based and sampling-based strategies. The main steps are: extracting boundaries in the environment using a boundary-based approach; selecting a viewpoint for each boundary using a sampling-based strategy; calculating the information gain of each viewpoint; and choosing the optimal viewpoint for that boundary. Simultaneously, the optimal viewpoint is selected by considering the information gain of each viewpoint and the energy consumption of the robot traveling to each viewpoint, and a smooth path is planned for execution. This hybrid strategy is more efficient than the previous two approaches.

[0007] Improving exploration efficiency is a crucial research area in autonomous robot exploration. Currently, autonomous exploration using differential or Ackerman chassis faces challenges due to the coupling between the heading angle and the robot's direction of travel. This prevents the continuous and smooth execution of trajectories where the optimal observation direction differs from the robot's travel direction, leading to decreased exploration efficiency. Therefore, addressing these issues is a significant research focus. Summary of the Invention

[0008] This invention aims to improve the efficiency and continuous trajectory execution of dynamically constrained robots during autonomous environmental exploration. The purpose of this invention is to decouple the robot's direction of travel from the sensor orientation using a rotatable mechanism, enabling the robot to continuously execute the planned path regardless of the observation angle from different viewpoints. A boundary gravity action distance function and a boundary gravity calculation function are designed to plan the orientation of the rotation mechanism at various points along the robot's path, thereby improving the efficiency of autonomous exploration.

[0009] To achieve the above objectives, this invention proposes a method and apparatus for rapid autonomous exploration planning in environments suitable for dynamically constrained ground robots, comprising:

[0010] Step S1: Using the environmental depth map provided by the depth camera mounted on the rotatable mechanism, and the rotation angle of the depth camera relative to the robot chassis, combined with odometry information, an environmental occupancy grid map is established.

[0011] Step S2: Update the boundaries based on the established occupied grid map. The updated objects are only the boundaries in the current sensor field of view.

[0012] Step S3: For the updated boundary, acquire viewpoints and calculate the optimal observation angle for each viewpoint. Evaluate the potential information gain for each viewpoint to obtain the optimal viewpoint for each boundary;

[0013] Step S4: Update the connection cost matrix between the optimal viewpoints of each boundary and from the optimal viewpoints of each boundary to the robot's current position;

[0014] Step S5: Solve for the global coverage path that covers all optimal viewpoints within a certain range of the robot's current position;

[0015] Step S6: Solve for the dynamically feasible path connecting the robot's current position with the top N optimal viewpoints in the global coverage path, and perform trajectory optimization;

[0016] Step S7: Establish the attraction function based on the size of the boundary, and plan the orientation of the rotatable mechanism at each position on the dynamically feasible path;

[0017] Step S8: The chassis executes the planned dynamically feasible smooth trajectory, and the rotatable mechanism adjusts its orientation according to the unknown location of the chassis to obtain the maximum information gain;

[0018] Step S9: S1 will continue during the movement. When the target boundary is explored or the planned path collides with the newly established environment, the process will jump from S8 to S1. When there are no boundaries in the environment, the exploration is complete.

[0019] The rotatable mechanism is equipped with a depth camera to provide a depth map characterizing the environment from the current viewpoint. The rotatable mechanism is also equipped with a brushless motor to change the orientation of the depth camera. A potentiometer is mounted in the shaft of the rotatable mechanism to measure the relative angle between the depth camera and the robot chassis.

[0020] The method for updating the boundaries of the area within the current sensor's field of view specifically refers to detecting whether the bounding boxes of each boundary intersect with the current field of view. If there is an intersection, the boundary is updated.

[0021] The method for solving the cost matrix is ​​as follows: using A * The path search algorithm searches for feasible paths between various optimal viewpoints and connecting each optimal viewpoint to the current robot position in an idle, known environment, and sets the length of the feasible path as an element of the cost matrix.

[0022] The specific method for solving the global coverage path problem involves setting up a virtual warehouse, setting the connection cost between the virtual warehouse and the robot to 0, and then abstracting the global coverage path problem into a traveling salesman problem by combining the cost matrix.

[0023] The specific method for solving the dynamic feasible trajectory is as follows: guided by a globally covered path, a hybrid A... * The algorithm finds a dynamically feasible smooth path, then parameterizes the path as a B-spline curve, sets a loss function, and uses the gradient descent algorithm to optimize the trajectory, resulting in a smooth trajectory.

[0024] The boundary attraction distance function is specifically defined as D = k1N i Where D is the effective distance of the attraction, k1 is a parameter, and N i It represents the number of grid cells at the i-th boundary. Its specific function is to calculate the possible distance of action based on the size of the boundary. Within this distance range, the rotatable mechanism controls the orientation of the depth camera, ensuring it faces the centroid that always confronts the target boundary.

[0025] The boundary attraction function is specifically defined as follows: Where F is the calculated attractive force, k2 is the parameter, and N is the value of N.i is the number of grid cells on the i-th boundary, and r is the distance between a point on the trajectory and the centroid of the boundary. Its specific function is that if the areas of influence of two boundaries intersect, within that intersection range, this function calculates the gravitational force between the two boundaries. The directions of both gravitational forces are from the current position towards their respective centroids. Then, the parallelogram law is used to merge the two gravitational forces, and their direction is the orientation of the depth camera of the rotatable mechanism at that point.

[0026] The centroid of the boundary refers to the intersection point of the boundary grid along the optimal viewing direction at the optimal viewpoint of the boundary. The centroid location, combined with the attraction distance function and the attraction calculation function, ensures that when the ground robot travels to the optimal viewpoint of the boundary, the orientation of the depth camera is exactly at the optimal viewing angle of that viewpoint.

[0027] A rapid autonomous exploration system for dynamically constrained ground robots, using any of the methods and apparatus described above, includes the following modules:

[0028] Rotatable mechanism hardware module: A brushless motor is used to control the rotation of the depth camera to face the planned direction, and a potentiometer is used to record the rotation angle of the rotating mechanism relative to the chassis;

[0029] Robot chassis hardware module: Generally a differential robot or an Ackerman steering robot, it is the mobile platform of the entire system;

[0030] Occupation grid map creation software module: Based on odometer information, depth map, and the relative rotation angle described in the rotatable mechanism hardware module, an occupation grid map representing the entire environment is created;

[0031] Boundary Update and Viewpoint Acquisition Software Module: Based on the occupied grid map described in the occupied grid map creation software module, the boundary is updated, and the optimal viewpoint is selected for each boundary. When there are no boundaries in the environment, it indicates that the exploration task has been completed.

[0032] Global coverage path solving software module: Based on the optimal viewpoint and occupied grid map described in the boundary update and viewpoint acquisition software module, this module establishes the occupied grid map described in the software module. By solving the traveling salesman problem, this module obtains the robot's global coverage path.

[0033] The dynamic feasible path solution software module: Based on the global coverage path and occupied grid map described in the global coverage path solution module, this module establishes the occupied grid map described in the software module. This module uses a hybrid A... * By combining gradient optimization, a smooth, dynamically feasible trajectory is obtained by connecting the various optimal viewpoints.

[0034] Rotatable mechanism control quantity solution software module: Based on the boundary information described in the boundary update and viewpoint acquisition software module, and based on the boundary attraction distance function and boundary gravity calculation function, it plans the orientation of the rotatable mechanism at each point on the dynamic feasible trajectory described in the dynamic feasible path solution module, so as to complete the efficient exploration of the unknown environment.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] Firstly, this invention mounts a depth camera on a rotatable mechanism, which includes a rotation servo mechanism and an angle sensor. This decouples the sensor's orientation from the heading of the ground robot platform, allowing the platform to continuously execute the planned trajectory without interruption during exploration, without needing to rotate the chassis for easier exploration. This significantly improves the continuity and smoothness of the robot's trajectory execution, increases the average travel speed, and enhances the autonomous exploration efficiency of dynamically constrained ground robots.

[0037] Secondly, this invention designs a function to calculate the range of gravity and magnitude of gravity for boundary exploration, which is used to plan the orientation of the rotating mechanism that controls the orientation of the sensors during the operation of the robot platform, thereby improving the autonomous exploration efficiency of the dynamically constrained ground robot platform.

[0038] Third, this invention proposes a boundary attraction distance function and a boundary attraction calculation function. Using these two functions, the orientation of the depth camera on a dynamically feasible path can be planned, thereby improving the efficiency of autonomous exploration.

[0039] In summary, compared with existing technical solutions, the rapid autonomous exploration planning method and device for dynamically constrained ground robots designed in this invention greatly improves the efficiency of autonomous exploration. Attached Figure Description

[0040] The present invention will be further described below with reference to the accompanying drawings and examples.

[0041] Figure 1 Flowchart of the rapid environment exploration and planning method for dynamically constrained ground robots of the present invention;

[0042] Figure 2 This invention includes a differential ground robot equipped with a rotatable mechanism;

[0043] Figure 3 Three-view drawing of a differential ground robot with a rotatable mechanism added in this invention;

[0044] Figure 4 The difference between mapping using the differential robot with a rotatable mechanism in this invention and mapping using a regular differential robot;

[0045] Figure 5The sampling and evaluation method for the optimal viewpoint at the boundary in this invention;

[0046] Figure 6 A schematic diagram of the global coverage path in this invention;

[0047] Figure 7 Schematic diagram of the attraction distance function and attraction calculation function in this invention;

[0048] Figure 8 The present invention provides a structural block diagram of a rapid autonomous exploration system for dynamically constrained ground robots. Detailed Implementation

[0049] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0050] like Figure 1 As shown, this invention proposes a rapid autonomous exploration planning method suitable for dynamically constrained robot environments, comprising the following steps:

[0051] Step S1: Using the environmental depth information provided by the depth camera mounted on the rotatable mechanism, and the rotation angle of the depth camera relative to the robot chassis, combined with odometry information, an environmental occupancy grid map is established.

[0052] Differential robots equipped with this rotatable mechanism, such as Figure 2 As shown, its three views are as follows Figure 3 As shown, its specific components are as follows: Figure 2 As shown, ① is the rotating mechanism, ② is the depth camera, and ③ is the differential chassis.

[0053] The rotating mechanism has a potentiometer mounted on its shaft. The output voltage of the potentiometer changes with the rotation angle of the rotating mechanism. After processing by the embedded processor, the rotation angle of the depth camera relative to the robot chassis can be output in real time.

[0054] Occupancy grid maps divide a continuous state space into small cubes of a set size, which are called voxels in the occupancy grid map. If an obstacle exists in the space corresponding to a voxel in the occupancy grid map, the corresponding voxel will be set to an occupied state. If there is no obstacle in the space corresponding to the occupancy grid map, the corresponding voxel will be set to an idle state. Due to sensor noise, different measurement results may occur for the same voxel, indicating it may be occupied or idle. Therefore, the occupancy grid map mapping algorithm will fuse multiple measurement results to obtain a globally consistent global map.

[0055] The difference between this mapping method and ordinary ground robot mapping is as follows: Figure 4As shown, when a typical dynamically constrained ground robot autonomously explores its environment, it needs to stop and turn to the optimal viewpoint after reaching a certain point of view, resulting in discontinuous motion, such as... Figure 4 As shown above, the ground robot with a rotatable mechanism proposed in this invention decouples the robot's orientation from the depth camera's orientation, thereby enabling environmental mapping by changing the depth camera's orientation while continuously executing a trajectory. Figure 4 As shown below.

[0056] Step S2: Update the boundaries based on the established occupancy grid map. The update only applies to the boundaries within the current sensor field of view.

[0057] A boundary refers to the intersection of known free areas and unknown areas in an occupied grid map. If the robot moves towards the boundary, it can explore the unknown area. If no boundary exists in the space to be explored, the exploration is considered complete. The specific method for boundary updating is as follows: When each boundary is established, a bounding box that exactly contains it is created. During boundary updating, it first checks whether the bounding box of each boundary intersects with the current view. If there is an intersection, the boundary within that bounding box is considered to need updating. In addition, it searches for boundaries within the current view; if found, a new boundary is established.

[0058] Step S3: For the updated boundary, viewpoints are acquired, and the optimal observation angle is set for each viewpoint. The potential information gain of each viewpoint is evaluated to obtain the optimal viewpoint for each boundary.

[0059] Viewpoint acquisition such as Figure 5 As shown, the specific method is as follows: First, the average position of all grids in the boundary is calculated and set as the center of the boundary. Then, N equally spaced concentric circles are established with the boundary center as the center, where the radius of the largest concentric circle is R. Viewpoints are uniformly collected within the known area on each arc; these viewpoints are candidate viewpoints for the boundary. At each viewpoint, the optimal viewing angle and information gain are calculated using ray casting. The point with the highest information gain is selected as the optimal viewpoint for the boundary, and its optimal viewing angle is set as the optimal viewing angle for the boundary. When the ground robot reaches this viewpoint, the rotation mechanism will control the orientation of the depth camera to align with the optimal viewing angle.

[0060] Step S4: Update the connection cost matrix between the optimal viewpoints of each boundary and from the optimal viewpoint of each boundary to the robot's current position.

[0061] The cost matrix is ​​used to describe the travel costs between optimal viewpoints at each boundary and between each boundary and the robot. Specifically, it is constructed using A... *The algorithm searches for paths between optimal viewpoints and paths from each optimal viewpoint to the current robot position in the free known area, sets the length of these paths as the connection cost, and thus constructs a cost matrix.

[0062] Step S5: Solve for the global coverage path that covers all optimal viewpoints within a certain range of the robot's current position. Combining this with the cost matrix, the problem is abstracted into a traveling salesman problem for solution. The global coverage path is as follows: Figure 6 As shown.

[0063] The Traveling Salesman Problem (TSP) is a well-known combinatorial optimization problem that describes the shortest path problem for a traveling salesman who wants to visit multiple cities and return to his starting city. This problem can be formalized as follows: given a set of cities and the distances (or costs) between them, find a path that allows the salesman to visit each city only once, travel through all cities, and return to his starting city, while minimizing the total distance or total cost. In this problem, a virtual warehouse is introduced, and the connection cost between the current location and the virtual warehouse is set. The optimal viewpoints in this problem are equivalent to the cities in the TSP problem; solving the TSP problem yields the optimal globally covered path.

[0064] The Time-Limited Solution (TSP) is an NP-hard problem, meaning that the computation time required to solve it increases exponentially with the number of cities, making the computational complexity of finding an exact solution extremely high. Due to the complexity of the TSP, approximate and heuristic algorithms are currently the primary methods used to solve practical problems. For example, greedy algorithms can heuristically select the next city to visit; ant colony optimization, inspired by foraging behavior, simulates multiple ants searching for the optimal path among cities. For smaller-scale problems, exact algorithms such as backtracking or dynamic programming can be used to find the optimal solution. For larger-scale problems, approximate or heuristic algorithms are typically used to find better solutions.

[0065] Step S6, using hybrid A * The algorithm solves for a dynamically feasible path connecting the robot's current position to the top N optimal viewpoints in the global coverage path. This path is then parameterized as a B-spline curve, a suitable loss function is set, and gradient descent is used to optimize it. The resulting dynamically feasible trajectory is shown below. Figure 6 As shown;

[0066] Mixed A * Algorithm (Hybrid A) * ) is a heuristic search algorithm for path planning, particularly suitable for navigation problems of mobile robots or autonomous vehicles in continuous environments.

[0067] Mixed A * The algorithm combines the ideas of sampling and graph search. It divides a continuous plane into a discrete grid and performs a graph search on this discrete grid map. Specifically, it hybridizes A... * The algorithm divides the plane into a finite number of discrete grid cells, each representing a discrete state.

[0068] The algorithm steps are as follows:

[0069] Based on the vehicle's motion constraints, a discrete state grid is generated by uniformly sampling in a continuous space, and the start and end states are set.

[0070] Use a heuristic function to evaluate the cost of each state (usually estimated using Euclidean distance).

[0071] Construct a search graph that minimizes cost by adding the starting state to the Open list.

[0072] By gradually expanding the search graph, the cost and path of each state are continuously updated until the endpoint state is found or the search yields no solution.

[0073] By tracing back the path from the endpoint state, the final path planning result is obtained.

[0074] Mixed A * The algorithm's key advantage lies in its balance between continuous and discrete spaces. By uniformly sampling the continuous space, the computationally intensive continuous space search can be largely avoided. Simultaneously, by performing A on the discrete space... * For searching, an optimized graph search algorithm can be used to obtain better path planning results.

[0075] It should be noted that, in combination A * The algorithm's performance is affected by the grid resolution. A grid that is too large may prevent finding a suitable path, while a grid that is too small will increase computational complexity. Therefore, in practical applications, parameter selection and optimization need to be tailored to the specific circumstances.

[0076] The obtained path is parameterized as a B-spline curve, and then a loss function is designed for the B-spline curve for gradient descent.

[0077] B-spline curves are a type of curve representation widely used in computer graphics and computer-aided design (CAD). They are defined by a series of control points and node vectors.

[0078] The definition of a B-spline curve has two key parts: control points and node vectors. Control points: Control points are a set on the curve that determine its shape. Adjusting the positions of the control points changes the shape of the curve. Node vectors: Node vectors are non-decreasing sequences of real numbers used to define the parameterization of the curve. Parameterization refers to mapping parameter values ​​to points on the curve. The length of the node vector is usually greater than or equal to the number of control points. The node vectors determine the piecewise division and smoothness of the curve in the parameter space.

[0079] B-spline curves possess many useful properties, such as local controllability (adjusting control points locally can change the shape of the curve) and smoothness (the curve is smooth and continuous between control points). They can also represent complex curve shapes and can be finely tuned by adding control points and changing the node vectors.

[0080] Step S7: Establish the attraction function based on the size of the boundary, and plan the orientation of the rotatable mechanism at each position on the dynamically feasible path.

[0081] The attraction function consists of two parts. The first part is the attraction distance function D = k1N. i Where D is the effective distance of the attraction, k1 is a parameter, and N i The first part is the number of grid cells at the i-th boundary; the second part is the attraction function. Where F is the calculated attractive force, k2 is the parameter, and N is the value of N. i is the number of grid cells on the i-th boundary, and r is the distance between a point on the trajectory and the centroid of the boundary.

[0082] The control parameters of the rotatable mechanism are specifically planned as follows: the intersection of the viewpoint direction of the optimal viewpoint on the boundary and the boundary is set as the centroid of that boundary. The range of gravitational force on each boundary is calculated using the attraction force distance function. If the gravitational ranges of two consecutive boundaries do not intersect, the orientation of the depth camera rotatable mechanism within its respective gravitational range is the centroid of the boundary. If there is an intersection, the attraction force at the intersection is calculated using the attraction force function, and then synthesized using the parallelogram law to obtain the current orientation of the rotatable mechanism.

[0083] Step S8: The chassis executes the planned dynamically feasible smooth trajectory, and the rotatable mechanism adjusts its orientation according to the unknown location of the chassis to obtain the maximum information gain.

[0084] Step S9 continues during the movement. When the target boundary is fully explored or the planned path collides with the newly established environment, the process jumps from S8 to S1. When there are no boundaries in the environment, the exploration is complete.

[0085] like Figure 2The image shows a differential ground robot equipped with the rotatable mechanism proposed in this patent. ① is the rotatable mechanism proposed in this patent, ② is a depth camera, and ③ is the differential chassis.

[0086] Specifically, in implementation, the rotatable mechanism proposed in this invention is fixedly connected to the chassis of a ground robot via a base. A depth camera is mounted on the rotatable mechanism, providing a depth map representing the environment from the current viewpoint. The rotatable mechanism is equipped with a brushless motor, which drives the rotating part of the mechanism to rotate around an axis, thereby changing the orientation of the depth camera to meet the exploration viewpoint requirements. A potentiometer is mounted on the rotating shaft of the rotatable mechanism; the voltage output by this potentiometer changes with the angle of the rotatable mechanism. After processing by an embedded device, the relative rotation angle between the depth camera and the robot chassis can be output. The depth map output by the depth camera and its rotation angle relative to the chassis, combined with the chassis's odometry information, can complete the establishment of an environmental occupancy grid map.

[0087] like Figure 7 The diagram shown illustrates the attraction distance function and attraction calculation function in this invention.

[0088] This invention proposes a boundary attraction distance function and a boundary attraction function to plan the orientation of rotatable mechanisms on a dynamically feasible path, enabling rapid environmental exploration while the robot chassis moves continuously. The specific application of this method is as follows: First, the attraction distance function is used to calculate the attraction distance of each boundary. Within the range of influence, the rotatable mechanism controls the depth camera to face the centroid of the boundary. If the ranges of influence of two consecutive boundaries intersect, the attraction distance function is used to calculate the magnitude of the attraction between the two boundaries at that location within the intersection area. These forces are then synthesized using the parallelogram law to obtain the orientation of the rotatable mechanism at that location. This ensures that more exploration time is allocated to larger boundaries during movement, resulting in higher exploration gains and improved exploration efficiency. A schematic diagram of the above method is shown below. Figure 7 As shown.

Claims

1. A method and apparatus for rapid autonomous exploration and planning in environments with constrained dynamics for ground robots, characterized in that... The process includes the following steps: Utilizing environmental depth information provided by a depth camera mounted on a rotatable mechanism, the camera's rotation angle relative to the robot chassis, and odometry information, an environmental occupancy grid map is established; the boundaries are updated based on the established occupancy grid map; viewpoints are acquired for the updated boundaries, and the optimal viewing angle is calculated for each viewpoint; the information gain of each viewpoint is evaluated to obtain the optimal viewpoint and optimal viewing angle for each boundary; the connection cost matrix between optimal viewpoints on each boundary and from each optimal viewpoint to the robot's current position is updated; and the dynamically feasible paths connecting the robot's current position to the top N optimal viewpoints in the global coverage path are solved. The A* algorithm is used to search for feasible paths between optimal viewpoints at each boundary and from each optimal viewpoint to the robot's current position. A cost matrix is ​​set, and using the obtained cost matrix, the order in which the robot explores each boundary is abstracted into a traveling salesman problem. The global coverage path is obtained by solving this problem. A dynamically feasible path connecting the robot's current position to the top N optimal viewpoints on the global coverage path is planned, and trajectory optimization is performed. An attraction function is established based on the size of the boundary. The intersection of the viewpoint direction of the optimal viewpoint at the boundary and the boundary is set as the centroid of the boundary to plan the orientation of rotatable mechanisms at different positions on the path. The attraction function consists of two parts: the first part is the attraction distance function. ,in It is the distance at which attraction works. It is a parameter. It is the first The number of grid cells at each boundary; the second part is the attraction function. ,in It is the appeal of computation. It is a parameter. It is the first The number of grid cells at each boundary. It is the distance between a point on the trajectory and the centroid of the boundary. The range of gravity of each boundary is calculated by the attraction force distance function. If the ranges of gravity of the two boundaries do not intersect, then the orientation of the depth camera's rotatable mechanism is the centroid of the boundary within its respective range of gravity. If there are boundaries, the attraction of each boundary is calculated using an attraction function at the boundary, and then synthesized using the parallelogram law to obtain the current orientation of the rotatable mechanism. The orientation of the rotatable mechanism at each position on the planned dynamically feasible path is then determined. The robot chassis executes the planned dynamically feasible smooth trajectory, and the rotatable mechanism adjusts its orientation according to the position of the chassis. During the movement, the environmental occupancy grid map is continuously built. When the target boundary is explored or the planned path collides with the newly built environment, the global path search and dynamically feasible path generation are restarted. When there are no boundaries in the environment, the exploration is completed.

2. The rapid autonomous exploration planning method and apparatus for dynamically constrained ground robots according to claim 1, characterized in that: A rotatable mechanism carrying a depth camera is used. This mechanism is equipped with a brushless motor, which decouples the direction of the ground robot's movement from the orientation of the depth camera. This allows the robot to perform environmental perception without being restricted by its direction of movement during the execution of its trajectory. The rotatable mechanism is equipped with a potentiometer, which can output the current angle of the depth camera relative to the chassis. This angle information, combined with the depth information provided by the depth camera and the odometry information, completes the establishment of the occupation grid map.

3. The rapid autonomous exploration planning method and apparatus for dynamically constrained ground robots according to claim 1, characterized in that: The system updates the area and boundaries within the current sensor field of view, deletes explored boundaries, adds newly generated boundaries, uniformly samples viewpoints in a circular area around the boundary center, and calculates the optimal boundary observation angle for each viewpoint using ray casting. The viewpoint with the greatest information gain and its observation angle are selected as the viewpoint of that boundary.

Images