CN118533165A - An autonomous navigation method for mobile inspection robots based on narrow space perception - Google Patents

Abstract

The invention discloses an autonomous navigation method of a mobile inspection robot based on narrow space perception, which comprises the steps of S1, generating a global path, S2, introducing a novel space obstacle descriptor of an inspection robot view angle, S3, preprocessing the global path, S4, advancing in a non-narrow area, advancing in a narrow area, S5, and completing navigation. The invention reduces the burden of the inspection robot, ensures that the inspection robot accurately moves in a preset path, has higher track tracking precision, and safely passes through a narrow space, thereby smoothly realizing navigation tasks.

Description

An autonomous navigation method for mobile inspection robots based on narrow space perception

Technical Field

The present invention relates to the field of perception planning and autonomous navigation technology of mobile inspection robots, and in particular to an autonomous navigation method of a mobile inspection robot based on narrow space perception.

Background Art

Inspection robots are designed using modern inspection robot technology, artificial intelligence, data analysis and other advanced technologies to automatically perform inspection tasks. These inspection robots are widely used in many fields, including but not limited to facility maintenance, industrial inspection, safety monitoring, etc. They can automatically complete tedious, dangerous or highly repetitive tasks that originally required manpower, thereby improving efficiency, reducing costs and increasing operational safety. In the field of rail transit, inspection robots have been used to a certain extent. For example, the train inspection robot uses a wheeled motion platform combined with a robotic arm and a visual system. It runs in the maintenance trench of the vehicle depot and replaces manual labor to complete the inspection tasks of the bottom of the train during daily maintenance work.

Inspection robots have their own unique characteristics and working environments. First, these inspection robots are often large and heavy, which limits their maneuverability. Second, they need to operate in complex working environments, which are full of narrow and difficult-to-navigate areas. To meet these challenges, the operation mode of the inspection robot needs to strictly follow the preset path, while implementing a strict obstacle parking strategy while ensuring passability. This involves three key issues: high-precision trajectory tracking to ensure that the inspection robot can accurately follow the predetermined path; semantic-level environmental perception to help the inspection robot understand and adapt to its complex working environment; and environmentally adaptive obstacle parking strategy to ensure safe and effective parking when encountering spatial obstacles.

Existing inspection robots generally use an engineered single-point autonomous navigation obstacle avoidance algorithm, which uses radar to directly sense and set deceleration areas and parking areas to slowly pass through narrow spaces to achieve collision-free parking. An autonomous navigation obstacle avoidance algorithm based on Costmap2D is used to generate a spatial obstacle cost map that integrates the navigation map and real-time perception by setting the expansion radius, and a local planner is used to plan a low-cost obstacle avoidance path in real time based on the cost map. These methods only support single-point (straight path) navigation, and the workload of on-site deployment is large; the obstacle parking strategy does not integrate map information; it only works well in wide spaces, and in narrow environments, it will stop abnormally due to the environment triggering obstacle avoidance false alarms; the expansion radius of each area needs to be manually set; the cost map is large in scale, complex in calculation, and has low real-time performance.

Summary of the invention

In order to overcome the shortcomings and deficiencies of the prior art, the present invention provides an autonomous navigation method for a mobile inspection robot based on narrow space perception. The technical solution adopted by the present invention includes:

Step S1, generate a global path: use the SLAM algorithm for real-time positioning and mapping, generate a global map of the environment, specify the target location on the global map, and use the A* algorithm for global path planning. This method uses a heuristic search algorithm to measure the distance relationship between the real-time search position and the target position, so that the search direction is prioritized towards the direction of the target point, and ultimately achieves the effect of improving the search efficiency, thereby calculating an optimal and safe global path based on the starting point and the end point on the basis of the existing map.

Step S2, introduce a new spatial obstacle descriptor from the inspection robot's perspective: Based on the laser radar data and ultrasonic radar data, four spatial obstacle detection areas are formed in front, behind, left and right of the inspection robot, including N grids. When a spatial obstacle is detected in the grid, the spatial obstacle information is downsampled into a line segment in the spatial obstacle grid according to the minimum value of the parking distance (safety distance) and the closest scanning distance, with a total of 4N spatial obstacle line segments. The size and number of the grids are determined by the parking safety distance required in the actual environment.

Step S3, global path preprocessing: on the acquired global path, simulate the movement of the inspection robot, use the new spatial obstacle descriptor, and use the sliding window-computational geometry judgment method to perceive narrow spaces. The details are as follows: first, the global path is sampled, and the sampled path point set W = { _w1 , _w2 , ..., _wn } is set. With the sampled global path point as the center, a spatial obstacle search sliding window is established which is parallel to the coordinate system and has a side length of multiples of the parking distance; secondly, the spatial obstacle detection area is constructed in the inspection robot coordinate system using the new spatial obstacle descriptor, and is transformed into the grid navigation map according to the global path point; then, it is detected whether there is an occupied point in the sliding window grid of each sampling point in the simulated driving process within the spatial obstacle perception area polygon. If so, the point w _i is stored in the spatial obstacle spatial point set 0. After traversing all the path point sets, each point in 0 and its two preceding and following points are connected according to the global path, that is, w _i , w _i-1 , w _i-2 , w _i+1 , w _i+2 are taken, and the resulting line segment is the narrow spatial path. After preprocessing the global path, a global path is formed which distinguishes between narrow area paths and non-narrow area paths.

Step S4, moving in non-narrow areas: After global path preprocessing, in non-narrow areas, where there are fewer spatial obstacles, the local path planning of the inspection robot's formal movement adopts the speed sampling method. The core idea is to determine a sampling speed space that meets the hardware constraints of the mobile inspection robot in the speed space (v, w) according to the current position state and speed state of the mobile inspection robot, and then calculate the trajectory of the mobile inspection robot moving for a certain period of time under these speed conditions, and evaluate the trajectory through the evaluation function, and finally select the speed corresponding to the best evaluated trajectory as the movement speed of the mobile inspection robot. The speed sampling space is composed of various constraints, the basic ones include speed limit, acceleration limit and environmental space obstacle limit. Let the speed sampling space be V _s , and the speed space under speed limit be V _m , then

V _m ={(v, w)|v∈[v _min , v _max ], w∈[w _min , w _max ]}

Among them, v _min and v _max are the minimum linear velocity and maximum linear velocity of the inspection robot respectively, and w _min and w _max are the minimum angular velocity and maximum angular velocity of the inspection robot respectively.

At the same time, the acceleration of the inspection robot also has boundary limits. Suppose the speed space under the acceleration limit is V _d , then

Wherein, v _c and w _c are the linear velocity and angular velocity of the inspection robot at the current moment, respectively; a _vmax and a _wmax are the maximum linear acceleration and maximum angular acceleration of the inspection robot, respectively.

In addition, for local path planning, dynamic spatial obstacles need to be considered. Since there are more redundant spaces in non-narrow spaces, the perception map uses the general Costmap2d method, which can not only detect spatial obstacles, but also allow dynamic obstacle avoidance, and use redundant space to loosely track the path to complete the task of reaching the target point. Therefore, the speed constraint space considering the spatial obstacle factor is

Where dist(v, w) represents the shortest distance between the corresponding simulation trajectory and the spatial obstacle at the current speed. When there is no spatial obstacle, dist(v, w) will be a large constant value. This constraint can be used to safely decelerate and avoid spatial obstacles.

Therefore, combining the three types of speed restrictions, the speed sampling space can be obtained:

_Vs = _{Vm ∩Vd ∩Va}

After determining the velocity sampling space _Vs , the angular velocity and linear velocity are sampled. Let _Ew and _Ev represent the sampling rate. The linear velocity takes a value every _Ev , and the angular velocity takes a value every _Ew . When a group of (v, w) is sampled, the trajectory can be predicted through the motion model of the inspection robot. The inspection robot adopts the differential drive model here:

In the formula, (x, y, θ) represents the posture of the inspection robot, k represents the sampling time, and Δt represents the sampling interval.

By simulating the trajectory, we can get a trajectory group. Some trajectories are feasible for the inspection robot, while some are not. Therefore, we need to determine a standard to select which trajectory to use for actual movement. This standard is called the trajectory evaluation function. Here, the evaluation function is set as

Among them, heading(v, w) is the azimuth evaluation function, which is used to evaluate the angle error between the position direction of the trajectory end point and the target point connection line at the current sampling speed, and headdis _weight is the weight of heading(v, w); dist(v, w) is the distance evaluation function, which represents the inverse ratio of the distance between the corresponding simulated trajectory and the spatial obstacle at the current speed, and dist _weight is the weight of dist(v, w); velocity(v, w) represents the inverse ratio of the current speed, and velocity _weight is the weight of velocity(v, w). The larger the evaluation function value, the better the trajectory.

In this way, the inspection robot can move safely and avoid obstacles in non-narrow areas.

Traveling in narrow areas: Traveling in narrow areas uses an improved speed sampling algorithm. First, the detection mechanism for spatial obstacles is improved. A new spatial obstacle descriptor is used to detect from the perspective of the inspection robot. The speed constraint space at this time is still V _a in the formula, but the closest distance between the inspection robot and the spatial obstacle required by dist(v, w) is different. For the inspection robot, when the spatial obstacle does not enter the grid range set by the inspection robot, the spatial obstacle is not considered, and dist(v, w) is set to the maximum value. Within the grid range, dist(v, w) is the minimum distance to the spatial obstacle line segment. Secondly, the obstacle avoidance strategy is improved. Since the space is relatively narrow, when the inspection robot is completely unable to track the global path due to the spatial obstacle that directly blocks the global path, it chooses to stop and the speed is set to 0. Finally, the simulation of the trajectory is improved. At this time, not only the trajectory should be simulated, but the perception of the inspection robot's own perspective should be added when simulating the trajectory. That is, after simulating the inspection robot's motion trajectory in the same way as in a non-narrow space, the forward projection simulation method is used to detect whether the inspection robot will touch spatial obstacles or spatial obstacles when moving along the simulated trajectory, thereby forming a new evaluation function.

Among them, dist _point is the distance between the end point of the current trajectory and the path tracking point. The path tracking point is the point on the global path that is expected to be reached after the sampling interval Δt strictly following the global path at the current speed. Point _weight is the weight of dist _point . Factor _obstacle is the spatial obstacle collision factor, which is the inverse ratio of the distance between the inspection robot and the nearest spatial obstacle under the inspection robot's own perspective. Ov _weight is the weight of factor _obstacle . Dist _path is the distance from the current path to the global path, and path _weight is the weight of dist _path . The larger the value of the evaluation function, the better the trajectory.

This evaluation function tends to choose the method of tracking the global path rather than circumventing spatial obstacles. Therefore, the selected simulation projections are concentrated around the global path to ensure the safety of traveling in narrow spaces. And because the area is narrow, there are trajectories with forward simulation projections sampled at all speeds that cannot travel safely. At this time, the speed is reduced and the sampling process is repeated. When the speed is reduced to 0, it means the vehicle stops.

Step S5, navigation completed: different travel strategies for narrow space and non-narrow space are adopted so that the inspection robot can smoothly reach the target point, stop after reaching the target point, and the navigation is completed.

Beneficial effects:

The present invention proposes an autonomous navigation method for a mobile inspection robot based on narrow space perception. The method adopts a novel spatial obstacle descriptor and a global path preprocessing technology to perceive spatial obstacles and distinguish narrow spaces during the movement process, targeting the working scene characteristics of the inspection robot. The present invention reduces the burden on the inspection robot itself, ensures that the inspection robot moves precisely along the preset path, has a high trajectory tracking accuracy, and safely passes through narrow spaces, thereby smoothly completing the navigation task.

1. This invention proposes a new spatial obstacle descriptor, redefines the spatial obstacle perception method, generates a grid map around the inspection robot, and only focuses on the data around the grid map obtained by the radar. This data is used to determine whether the spatial obstacle is within the safe distance of the inspection robot, which simplifies the representation of the distance and position of the spatial obstacle, reduces the time and space complexity, and makes the algorithm suitable for embedded devices;

2. The present invention proposes a global path narrow space perception preprocessing method, which uses the sliding window method combined with a new spatial obstacle descriptor to perceive spatial obstacles from the perspective of the inspection robot itself, and effectively distinguishes between static narrow areas and non-narrow areas in the map before the inspection robot officially moves, making the inspection robot's navigation in complex environments more flexible and reliable.

3. The present invention proposes a trajectory tracking collision prediction and maximum speed adaptive scaling method, which adopts a forward simulation projection-computational geometry judgment method to sample points in the forward simulation trajectory, and based on the posture of each point, the inspection robot is projected and transformed into a grid navigation map to form a forward simulation of the inspection robot. The speed is scaled by whether the forward simulation interferes with spatial obstacles, thereby ensuring both the safety of travel and the fit of the travel path to the predetermined global path.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of the method steps of the present invention;

FIG2 is a schematic diagram of a novel spatial obstacle descriptor of the present invention;

FIG3 is a schematic diagram of a global path of sliding window processing according to the present invention;

FIG. 4 is a schematic diagram of forward projection simulation-trajectory tracking of the present invention.

DETAILED DESCRIPTION

It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

As shown in FIG1 , a method for autonomous navigation of a mobile inspection robot based on narrow space perception includes:

Step S1, generate a global path: use the Cartographer algorithm for real-time positioning and mapping, generate a global map of the environment, specify the target location on the global map, and use the A* algorithm for global path planning. This method uses a heuristic search algorithm to measure the distance relationship between the real-time search position and the target position, so that the search direction is prioritized towards the direction of the target point, and ultimately achieves the effect of improving the search efficiency, thereby calculating an optimal and safe global path based on the starting point and the end point on the basis of the existing map.

Step S2, introduce a new spatial obstacle descriptor from the inspection robot's perspective: based on the laser radar data and ultrasonic radar data, four spatial obstacle detection areas are formed in front, behind, left and right of the inspection robot, including N grids, as shown in Figure 2, where 1 is various spatial obstacles, 2 is the converted line segment information, and 3 is the inspection robot. When a spatial obstacle is detected in the grid, the spatial obstacle information is downsampled to the line segment in the spatial obstacle grid according to the minimum value of the parking distance (safety distance) and the closest scanning distance, for a total of 4N spatial obstacle line segments. The size and number of grids are determined by the parking safety distance required in the actual environment. Here N is set to 20, and the parking distance is set to 0.8m;

Step S3, global path preprocessing: on the acquired global path, simulate the movement of the inspection robot, that is, use the sliding window-computational geometry judgment method to perceive the narrow space. As shown in Figure 3, 1 is the spatial obstacle perception area, 2 is the global path, 3 is the inspection robot, 4 is the spatial obstacle search sliding window, 5 is the spatial obstacle, and 6 is the navigation map. First, the global path is sampled, and the sampled path point set W = {w ₁ , w ₂ , ..., w _n } is set. With the sampled global path point as the center, a spatial obstacle search sliding window is established that is parallel to the coordinate system and has a side length of the multiple of the parking distance. Secondly, the spatial obstacle detection area is constructed in the inspection robot coordinate system using the new spatial obstacle descriptor, and is transformed into the grid navigation map according to the global path point. Then, it is detected whether there is an occupied point in the sliding window grid of each sampling point in the simulated travel process within the spatial obstacle perception area polygon. If so, the point w _i is stored in the spatial obstacle spatial point set 0. After traversing all the path point sets, each point in 0 and its two preceding and following points are connected according to the global path, that is, w _i , w _i-1 , w _i-2 , w i ₊₁ , w _i+2 The resulting line segment is the narrow space path. After preprocessing the global path, a global path consisting of narrow area paths and non-narrow area paths is formed, making the actual driving more accurate.

Step S4, moving in non-narrow areas: After global path preprocessing, in non-narrow areas, where there are fewer spatial obstacles, the local path planning of the inspection robot's formal movement adopts a speed sampling algorithm, namely a dynamic window algorithm. The core idea is to determine a sampling speed space that meets the hardware constraints of the mobile inspection robot in the speed space (v, w) according to the current position state and speed state of the mobile inspection robot, and then calculate the trajectory of the mobile inspection robot moving for a certain period of time under these speed conditions, and evaluate the trajectory through an evaluation function, and finally select the speed corresponding to the best evaluated trajectory as the movement speed of the mobile inspection robot. The speed sampling space is composed of various constraints, the basic ones include speed limit, acceleration limit and environmental space obstacle limit. Let the speed sampling space be V _s , and the speed space under speed limit be V _m , then

V _m ={(v, w)|v∈[v _min , v _max ], w∈[w _min , w _max ]}

Where v _min and v _max are the minimum linear velocity and maximum linear velocity of the inspection robot, respectively, and w _min and w _{max are the minimum angular velocity and maximum angular velocity of the inspection robot, respectively} . Here, v _min and v _max are set to 0.02m/s and 0.8m/s, respectively, and w _min and w _max are set to 0.05rad/s and 0.3rad/s, respectively.

At the same time, due to the limitation of the driving motor, the acceleration of the inspection robot also has boundary limitations. Suppose the speed space under the acceleration limitation is V _d , then

Wherein, v _c and w _c are the current linear velocity and angular velocity of the inspection robot, respectively, and a _vmax and a _wmax are the maximum linear acceleration and maximum angular acceleration of the inspection robot, respectively. Here, a _vmax and a _wmax are set to 0.6 m/s ² and 0.3 rad/s ² , respectively.

In addition, for local path planning, dynamic spatial obstacles need to be considered. Since there is more redundant space in non-narrow spaces, the perception map uses a general costmap method, which can not only detect spatial obstacles, but also allow dynamic obstacle avoidance, and use redundant space to loosely track the path to complete the task of reaching the target point. Therefore, the speed constraint space considering the spatial obstacle factor is

Where dist(v, w) represents the shortest distance between the corresponding simulation trajectory and the spatial obstacle at the current speed. When there is no spatial obstacle, dist(v, w) will be a large constant value. This constraint can be used to safely decelerate and avoid spatial obstacles. Here dist(v, w) is set to 100.

Therefore, combining the three types of speed restrictions, the speed sampling space can be obtained:

_Vs = _{Vm ∩Vd ∩Va}

After determining the velocity sampling space _Vs , the angular velocity and linear velocity are sampled. Let _Ew and _Ev represent the sampling rate. The linear velocity takes a value every _Ev , and the angular velocity takes a value every _Ew . When a set of (v, w) is sampled, the trajectory can be predicted through the motion model of the inspection robot. The inspection robot adopts the differential drive model here:

In the formula, (x, y, θ) represents the posture of the inspection robot, k represents the sampling time, and Δt represents the sampling interval.

By simulating the trajectory, we can get a trajectory group. Some trajectories are feasible for the inspection robot, while some are not. Therefore, we need to determine a standard to select which trajectory to use for actual movement. This standard is called the trajectory evaluation function. Here, the evaluation function is set as

Among them, heading(v, w) is the azimuth evaluation function, which is used to evaluate the angle error between the position direction of the trajectory end point and the target point connection line at the current sampling speed. Heading _weight is the weight of heading(v, w); dist(v, w) is the distance evaluation function, which represents the inverse ratio of the distance between the corresponding simulated trajectory and the spatial obstacle at the current speed. Dist _weight is the weight of dist(v, w); velocity(v, w) represents the inverse ratio of the current speed. Velocity _weight is the weight of velocity(v, w). Here, the heading _weight is 15.

The dist _weight is 100, the velocity _weight is 30,

In this way, the inspection robot can move safely and avoid obstacles in non-narrow areas.

Traveling in narrow areas: Traveling in narrow areas uses an improved speed sampling algorithm. First, the detection mechanism for spatial obstacles is improved. A new spatial obstacle descriptor is used to detect from the perspective of the inspection robot. The speed constraint space at this time is still V _a in the formula, but the closest distance between the inspection robot and the spatial obstacle required by dist(v, w) is different. For the inspection robot, when the spatial obstacle does not enter the grid range set by the inspection robot, the spatial obstacle is not considered, and dist(v, w) is set to the maximum value. Within the grid range, dist(v, w) is the minimum distance to the spatial obstacle line segment. Secondly, the obstacle avoidance strategy is improved. Since the space is relatively narrow, when the inspection robot is completely unable to track the global path due to the spatial obstacle that directly blocks the global path, it chooses to stop and the speed is set to 0. Finally, the simulation of the trajectory is improved. At this time, not only the trajectory should be simulated, but also the perception of the inspection robot's own perspective should be added when simulating the trajectory. That is, after simulating the movement trajectory of the inspection robot in the same way as in a non-narrow space, the forward projection simulation method shown in Figure 4 is used to detect whether the inspection robot will touch the spatial obstacle or spatial obstacle when moving along the simulated trajectory, thereby forming a new evaluation function. Among them, 1 is the inspection robot, 2 is the forward simulation projection of the inspection robot, 3 is the spatial obstacle, 4 is the local path, and 5 is a schematic diagram of the intersection of the spatial obstacle and the projection.

Among them, dist _point is the distance between the end point of the current trajectory and the path tracking point. The path tracking point is the point on the global path that is expected to be reached after the sampling interval Δt in strict accordance with the global path at the current speed. Point _weight is the weight of dist _point ; factor _obstacle is the spatial obstacle collision factor, which is the inverse ratio of the distance between the inspection robot and the nearest spatial obstacle under the method of the inspection robot's own perspective. Ov _weight is the weight of factor _obstacle ; dist _path is the distance from the current path to the global path, and path _weight is the weight of dist _path . The larger the value of the evaluation function, the better the trajectory. Here, the point _weight is 1, the ov _weight is 100, and the path _weight is 40.

This evaluation function tends to choose the method of tracking the global path rather than circumventing spatial obstacles. Therefore, the selected simulation projections are concentrated around the global path to ensure the safety of traveling in narrow spaces. And because the area is narrow, there are trajectories with forward simulation projections sampled at all speeds that cannot travel safely. At this time, the speed is reduced and the sampling process is repeated. When the speed is reduced to 0, it means the vehicle stops.

Step S5, navigation completed: different travel strategies for narrow space and non-narrow space are adopted so that the inspection robot can smoothly reach the target point, stop after reaching the target point, and the navigation is completed.

Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various equivalent changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims (8)

1. The autonomous navigation method of the mobile inspection robot based on narrow space perception is characterized by comprising the following steps of:

Step S1, generating a global path: performing instant positioning and mapping by using a SLAM algorithm, generating a global map of the environment, and performing global path planning by using an A-type algorithm after a target place is designated on the global map;

Step S2, introducing a novel space obstacle descriptor of the inspection robot view angle: forming four space obstacle detection areas in front of, behind, left of and right of the inspection robot by utilizing laser radar data and ultrasonic radar data, wherein the four space obstacle detection areas comprise N grids, and when space obstacles in the grids are detected, the space obstacle information is downsampled into line segments in the space obstacle grids according to the minimum values of the distance between the space obstacle information and the parking distance and the scanning nearest distance to obtain 4N space obstacle line segments;

step S3, global path preprocessing: simulating the running of the inspection robot on the acquired global path, and using a sliding window-calculation geometric judgment method to sense a narrow space by using a novel space obstacle descriptor;

Step S4, advancing in a non-narrow area: in a non-narrow area, a local path planning of formal running of the inspection robot adopts a speed sampling method, a sampling speed space meeting the hardware constraint of the mobile inspection robot is determined in a speed space (v, omega) according to the current position state and the speed state of the mobile inspection robot, then the track of the mobile inspection robot moving for a certain time under the speed conditions is calculated, the track is evaluated through an evaluation function, and the speed corresponding to the track with the optimal evaluation is selected as the movement speed of the mobile inspection robot;

narrow area travel: the method comprises the steps that an improved speed sampling algorithm is adopted for narrow area advancing, a detection mechanism for space obstacles is improved, a novel space obstacle descriptor mode is adopted, detection is carried out in a view angle of a patrol robot, a speed constraint space is still V _a, dist (V, w) is set to be a maximum value, an obstacle avoidance strategy is improved, stopping is selected when the patrol robot cannot track a global path at all due to the fact that the global path is blocked by the space obstacles, the speed is set to be 0, the simulation of a track is improved, and whether the patrol robot touches the space obstacle or the space obstacle when advancing according to the simulated track is detected by adopting a forward projection simulation method, so that a new evaluation function is formed;

step S5, navigation is completed: by adopting different travelling strategies of the narrow space and the non-narrow space, the inspection robot can smoothly reach the target point, stop after reaching the target point, and complete navigation.

2. The autonomous navigation method of mobile inspection robot based on narrow space perception according to claim 1, wherein the size and number of grids in step S2 are determined by the parking safety distance required by the actual environment.

3. The autonomous navigation method of the mobile inspection robot based on the narrow space perception according to claim 1, wherein the step S3 includes:

s3.1, sampling a global path, setting a sampled path point set W= { W ₁,w₂,...,w_n }, taking the sampled global path point as a center, establishing a space obstacle search sliding window parallel to a coordinate system and taking a parking distance multiple as a side length;

s3.2, constructing a space obstacle detection area by using a novel space obstacle descriptor through a patrol robot coordinate system, and transforming the space obstacle detection area into a grid navigation map according to a global path point;

And S3.3, detecting whether occupied points exist in a sliding window grid of each sampling point in the simulation advancing process or not, if so, storing the points w _i into a space obstacle space point set O, after traversing the complete path point set, depending on a global path, connecting each point in the O and the front and rear points thereof according to the global path, namely taking w _i、w_i-1、w_i-2、w_i+1、w_i+2, and taking the formed line segment as a narrow space path, thereby preprocessing the global path and forming a global path for distinguishing a narrow area path and a non-narrow area path.

4. The autonomous navigation method of a mobile inspection robot based on narrow space perception according to claim 1, wherein the step S4, the velocity sampling space V _s includes a velocity limit, an acceleration limit and an environmental space obstacle limit,

At the speed limit, the speed sampling spatial expression is:

V_m＝{(ν,ω)|ν∈[ν_min,ν_max],ω∈[ω_min,ω_max]}

V _m is a speed sampling space under a speed limit, V _min、v_max is a minimum linear speed and a maximum linear speed of the inspection robot, and ω _min、ω_max is a minimum angular speed and a maximum angular speed of the inspection robot, respectively.

5. The autonomous navigation method of a mobile inspection robot based on narrow space perception according to claim 1, wherein,

At the acceleration limit, the velocity sampling spatial expression is:

V _d is a speed space under the limit of acceleration, V _c、ω_c is the linear speed and the angular speed of the inspection robot at the current moment, and a _vmax、a_ωmax is the maximum linear acceleration and the maximum angular acceleration of the inspection robot;

at the environmental spatial obstacle limit, the velocity sampling spatial expression is:

where V _a is the speed sampling space limited by the environmental spatial obstacle, dist (V, ω) represents the closest distance between the corresponding simulated trajectory and the spatial obstacle at the current speed.

6. The autonomous navigation method of the mobile inspection robot based on narrow space perception according to claim 1, wherein the speed sampling space V _s is obtained by integrating speed limitation, acceleration limitation and environmental space obstacle limitation:

V_s＝V_m∩V_d∩V_a

After the speed sampling space V _s is determined, the angular speed and the linear speed are sampled, E _ω、E_v is set to represent the sampling rate, the linear speed takes a value every other E _v, the angular speed takes a value every other E _ω, and when a group of (V, omega) is sampled, the track is predicted through the inspection robot motion model.

7. The autonomous navigation method of a mobile inspection robot based on narrow space perception according to claim 1, wherein the movement model of the inspection robot predicts the track by using a differential driving model, and the expression is:

Wherein, (x, y, θ) represents the pose of the inspection robot, k represents the sampling time, and Δt represents the sampling interval;

By simulating the track, a track group is obtained, and an evaluation function is established to evaluate the track travelling feasibility, wherein the evaluation function expression is as follows:

Wherein, the heading (v, ω) is an azimuth evaluation function, which is used to evaluate the error of the included angle between the track end position direction and the target point line at the current sampling speed, and the heading _weight is the weight of the heading (v, ω); dist (v, ω) is a distance evaluation function representing an inverse ratio of the distance between the corresponding simulated trajectory and the spatial obstacle at the current speed, dist _weight being a weight of dist (v, ω); the magnitude (v, ω) represents the inverse of the current speed magnitude, and the magnitude _weight is the magnitude (v, ω) of the magnitude, and the trajectory is more optimal as the evaluation function value is larger.

8. The autonomous navigation method of the mobile inspection robot based on narrow space perception according to claim 1, wherein the new evaluation function expression in step S4 is as follows:

wherein dist _point is the distance between the current track end point and the path tracking point, the path tracking point is the point on the global path which is expected to be reached by the global path through the sampling interval delta t under the current speed, and point _weight is the weight occupied by dist _point; factor _obstacle is a space obstacle collision factor, namely the inverse ratio of the distance between the inspection robot and the nearest space obstacle is perceived under the method of the inspection robot under the self view angle, and ov _weight is the weight of factor _obsiacle; dist _path is the distance from the current path to the global path, path _weight is the weight occupied by dist _path, and the track is more excellent as the evaluation function value is larger.