Robot environment perception and autonomous obstacle avoidance system based on data fusion

Abstract

The invention discloses a robot environment perception and autonomous obstacle avoidance system based on data fusion, and relates to the technical field of path planning. The method comprises the following steps: constructing a GIS distribution map and dividing working sub-regions, constructing image recognition models and point cloud twin spaces of different sensing objects, obtaining the sensing state of each working sub-region by using the image recognition models and the point cloud twin spaces, constructing an operation evaluation model, and obtaining bypassing energy consumption by using the operation evaluation model. Carrying out analogue simulation on the point cloud twinborn space to obtain the passing energy consumption, and selecting an optimal passing path according to the passing energy consumption and the bypassing energy consumption; whether the obstacle object in the working area can pass or not can be effectively judged, autonomous obstacle avoidance and energy-saving optimization can be comprehensively achieved, and the overall intelligent level of the robot is improved.

Description

TECHNICAL FIELD

[0001] The application relates to the path planning technical field, and particularly relates to a robot environment perception and autonomous obstacle avoidance system based on data fusion. BACKGROUND

[0002] The robot environment perception and autonomous obstacle avoidance based on data fusion is a key technology for improving the intelligent level of the robot, which can break through the limitation of a single sensor by intelligently integrating information from multiple sensors, and can provide strong support for the robot to realize safe, autonomous and efficient navigation and obstacle avoidance decision in a complex scene, however, it also faces various problems. In the prior art, the bypass mode is often selected for obstacle avoidance for different obstacle objects, and the bypass mode will inevitably increase the energy consumption of the robot, and many obstacle objects do not need to bypass and can pass through by taking the passing mode, so it is necessary to accurately identify whether the obstacle object can pass through and obtain the energy consumption in the two modes, so as to select the optimal passing path, and the application provides a robot environment perception and autonomous obstacle avoidance system based on data fusion. SUMMARY

[0003] The application aims to provide a robot environment perception and autonomous obstacle avoidance system based on data fusion.

[0004] The application can be realized by the following technical scheme: a robot environment perception and autonomous obstacle avoidance system based on data fusion, comprising the following modules: A region division module is used for constructing a GIS distribution map of a robot working area, and a plurality of working sub-areas are divided in the GIS distribution map. An environment perception module is used for acquiring image data of different perception objects, and constructing a corresponding image recognition model, acquiring point cloud data of the robot and different perception objects, and constructing a corresponding point cloud twin space. A region marking module is used for acquiring the perception state of each working sub-area by using the image recognition model and the point cloud twin space. A running evaluation module is used for acquiring historical running data of the robot, and constructing a corresponding running evaluation model by using an LSTM neural network. A path optimization module is used for simulating the point cloud twin space to acquire a passing energy consumption, acquiring a bypass energy consumption by using the running evaluation model, and selecting a corresponding optimal passing path according to the passing energy consumption and the bypass energy consumption.

[0005] Further, the process of constructing the GIS distribution map of the robot working area and dividing a plurality of working sub-areas in the GIS distribution map comprises: acquire distribution information of the working area, the distribution information including building contour, indoor wall and partition, room size, slope and height difference, facility location, furniture layout, passageway and entrance and exit, and construct a GIS distribution map of the working area according to each item of the distribution information by using GIS technology; acquire equipment size of the robot, the equipment size being used to reflect footprint of a single robot, and divide the GIS distribution map into a plurality of working sub-areas according to the equipment size of the robot in the GIS distribution map, and the area of each working sub-area being greater than the equipment size of the robot.

[0006] Further, the process of acquiring image data of different perception objects and constructing a corresponding image recognition model includes: The perception objects refer to different objects that can be interacted by the robot when it performs ground cleaning in its working area, and are divided into obstacle objects and cleaning objects, and the image data refer to pictures of different perception objects from multiple angles; generate an image recognition set corresponding to the image data of different perception objects and their object names, and divide the image recognition set into an image training set and an image test set; construct a convolutional neural network, use the image data of different perception objects in the image training set as input data of the convolutional neural network, use the corresponding object names in the image training set as output data of the convolutional neural network, train the convolutional neural network to obtain an initial convolutional neural network; use the image test set to verify the initial convolutional neural network, and output the initial convolutional neural network with an image test error less than or equal to a preset threshold as the image recognition model.

[0007] Further, the process of acquiring point cloud data of the robot and different perception objects and constructing a corresponding point cloud twin space includes: acquire point cloud data of the robot, construct a three-dimensional robot model according to the point cloud data, acquire point cloud data of different obstacle objects respectively, construct corresponding three-dimensional obstacle models respectively, and integrate the three-dimensional robot model and a single three-dimensional obstacle model to obtain a point cloud twin space of the two; in the point cloud twin space, fix the three-dimensional obstacle model, and move the three-dimensional robot model to determine whether it can pass through the bottom area of the three-dimensional obstacle model; if yes, mark the corresponding obstacle object as a passable obstacle, and if no, mark the corresponding obstacle object as an impassable obstacle, and mark each obstacle object respectively.

[0008] Further, the process of acquiring the perception state of each working sub-area by using the image recognition model and the point cloud twin space includes: An image acquisition unit and a point cloud acquisition unit are arranged on the robot respectively to acquire real-time image data around the robot and real-time point cloud data in a working area respectively; The real-time image data is input into an image recognition model to determine whether there is a perception object around the robot and to acquire an object name thereof, and the real-time point cloud data is used to determine whether there is a perception object in the working area and to acquire an object name thereof; Each identified perception object is synchronized to a GIS distribution map in real time, and the perception state includes a to-be-cleaned area, a passable area, an impassable area, and a passable area; The working sub-area where the cleaning object is located is marked as a to-be-cleaned area, the working sub-area where the obstacle object marked as a passable obstacle is located is marked as a passable area, the working sub-area where the obstacle object marked as an impassable obstacle is located is marked as an impassable area, and the other working sub-areas are marked as passable areas.

[0009] Further, the process of acquiring historical running data of the robot and constructing a corresponding running evaluation model using an LSTM neural network includes: The historical running data refers to the moving distance, cleaning object, and corresponding energy consumption of the robot during a single cleaning operation, a running evaluation set is generated according to the moving distance, cleaning object, and energy consumption in different historical running data, and the running evaluation set is divided into a running training set and a running test set; A convolutional neural network is constructed, different moving distances and cleaning objects in the running training set are used as input data of the convolutional neural network, and corresponding energy consumptions in the running training set are used as output data of the convolutional neural network, the convolutional neural network is trained to obtain an initial convolutional neural network; The running test set is used to verify the initial convolutional neural network, and the initial convolutional neural network with an output less than or equal to a preset running test error threshold is output as a running evaluation model.

[0010] Further, the process of simulating the point cloud twin space to obtain a pass-through energy consumption and using the running evaluation model to obtain a detour energy consumption includes: If there is a passable area between the real-time position of the robot and the nearest to-be-cleaned area, a pass-through path is constructed between the real-time position and the corresponding to-be-cleaned area using Dijkstra's algorithm, and if there is no passable area, a detour path is constructed between the two using Dijkstra's algorithm; For the corresponding to-be-cleaned area with only a detour path, the detour path is used as the optimal pass-through path, and the robot is controlled to perform cleaning work according to the optimal pass-through path; For the corresponding to-be-cleaned area existing the threading path and the bypass path at the same time, in the point cloud twin space of the robot and the single obstacle object on the corresponding threading path, the three-dimensional machine model and the three-dimensional obstacle model therein are adjusted to the corresponding direction and position in the actual application scene; The working process of the three-dimensional machine model is simulated by using the simulation software to obtain the simulation energy consumption required for passing through the three-dimensional obstacle model from the bottom area of the three-dimensional obstacle model; The total distance of the robot moving along the threading path to the corresponding to-be-cleaned area is taken as the threading distance of the threading path, and the threading distance is combined with the cleaning object of the corresponding to-be-cleaned area and input into the running evaluation model to obtain the moving energy consumption, and the sum of the simulation energy consumption and the moving energy consumption is taken as the threading energy consumption of the threading path; The total distance of the robot moving along the bypass path to the corresponding to-be-cleaned area is taken as the bypass distance of the bypass path, and the bypass distance is combined with the cleaning object of the corresponding to-be-cleaned area and input into the running evaluation model to obtain the bypass energy consumption of the bypass path.

[0011] Further, the process of selecting the corresponding optimal passing path according to the threading energy consumption and the bypass energy consumption comprises: The threading energy consumption and the bypass energy consumption of the robot reaching the corresponding to-be-cleaned area are compared, if the threading energy consumption is greater than or equal to the bypass energy consumption, the bypass path is taken as the optimal passing path, if the threading energy consumption is less than the bypass energy consumption, the threading path is taken as the optimal passing path; The real-time position of the robot and the threading energy consumption and the bypass energy consumption of the to-be-cleaned area closest to the robot are continuously obtained, and the optimal passing path is obtained, and the robot is controlled to clean according to the optimal passing path until all the to-be-cleaned areas are cleaned.

[0012] Compared with the prior art, the beneficial effects of the present application are: The present application can fuse and process the environment perception and autonomous obstacle avoidance of the robot based on different data by obtaining image data, point cloud data and historical running data, and can effectively judge whether the obstacle object in the working area can pass through by constructing an image recognition model and a point cloud twin space, which is beneficial to improve the accuracy of obstacle recognition; The simulation of the point cloud twin space can obtain the simulation energy consumption of the robot passing through the obstacle object in the threading mode, and the running evaluation model can obtain the moving energy consumption, so as to obtain the threading energy consumption of the robot in the threading mode, and by comparing with the bypass energy consumption in the bypass mode, the optimal passing path with the lowest energy consumption can be obtained, which is beneficial to comprehensively realize autonomous obstacle avoidance and energy saving optimization and improve the overall intelligent level of the robot. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 A schematic diagram of the module of the present application. DETAILED DESCRIPTION

[0014] As Figure 1 shown, a robot environment perception and autonomous obstacle avoidance system based on data fusion includes the following modules: A region division module is configured to construct a GIS distribution map of a robot working area and divide a plurality of working sub-areas in the GIS distribution map. An environment perception module is configured to acquire image data of different perception objects and construct corresponding image recognition models, acquire point cloud data of the robot and different perception objects, and construct corresponding point cloud twin spaces. A region marking module is configured to acquire perception states of each working sub-area by using the image recognition models and the point cloud twin spaces. A running evaluation module is configured to acquire historical running data of the robot and construct a corresponding running evaluation model by using an LSTM neural network. A path optimization module is configured to simulate the point cloud twin spaces to acquire a traversal energy consumption and acquire a detour energy consumption by using the running evaluation model, and select a corresponding optimal passing path according to the traversal energy consumption and the detour energy consumption.

[0015] It needs to be further explained that, in the specific implementation process, the process of constructing a GIS distribution map of a robot working area and dividing a plurality of working sub-areas in the GIS distribution map includes: In the embodiment of the present application, the robot refers to an intelligent mobile robot for floor cleaning, and the working area refers to an indoor area corresponding to the actual application scene of the intelligent mobile robot for floor cleaning. The distribution information of the working area is acquired, and the distribution information refers to various data necessary for constructing the GIS distribution map thereof, including building contours, indoor walls and partitions, room sizes, slopes and height differences, facility locations, furniture layouts, passages and entrances, and the GIS distribution map of the working area is constructed by using GIS technology according to the acquired various distribution information. The device size of the robot is acquired, and the device size is used to reflect the floor area of a single robot. In the GIS distribution map, the GIS distribution map is divided according to the device size of the robot, and the entire GIS distribution map is divided into a plurality of working sub-areas, and the area of each working sub-area is greater than the device size of the robot.

[0016] It needs to be further explained that, in the specific implementation process, the process of acquiring image data of different perception objects and constructing corresponding image recognition models includes: The perception object refers to different objects that the robot can interact with when it cleans the ground in its working area, and the perception object is divided into obstacle objects and cleaning objects, the obstacle object refers to objects that the robot cannot clean and will affect its passage, and the cleaning object refers to objects that the robot can clean and continue to pass through; For example, the obstacle object includes sofas, wardrobes, refrigerators, TV cabinets, toys, dolls, shoes, etc., and the cleaning object refers to paper scraps, residues, hair, garbage, etc. The image data refers to multi-angle pictures of different perception objects, which are used to reflect the shape characteristics of different perception objects, and the corresponding image recognition set is generated according to the image data and the object name of different perception objects, and is divided into an image training set and an image test set. A convolutional neural network is constructed, the image data of different perception objects in the image training set is used as the input data of the convolutional neural network, and the corresponding object name in the image training set is used as the output data of the convolutional neural network, and the convolutional neural network is trained to obtain an initial convolutional neural network. The initial convolutional neural network is verified by using the image test set, and the initial convolutional neural network with an output less than or equal to a preset image test error threshold is output as an image recognition model, and the image recognition model is used to output the corresponding object name according to the input image data.

[0017] It needs to be further explained that in the specific implementation process, the process of obtaining the point cloud data of the robot and different perception objects and constructing the corresponding point cloud twin space includes: The point cloud data is a set of vectors, usually represented in the form of x, y, z three-dimensional coordinates, which is a collection of mass points of object surface characteristics, the point cloud data of the robot is obtained, and a three-dimensional model thereof is constructed according to the point cloud data, which is denoted as a three-dimensional robot model. The same method is used to obtain the point cloud data of different obstacle objects respectively, and the corresponding three-dimensional models are constructed respectively, which are denoted as three-dimensional obstacle models, each three-dimensional obstacle model is bound with its corresponding object name, and the three-dimensional robot model and a single three-dimensional obstacle model are integrated into the same digital twin space, and are marked as the point cloud twin space of the two. In the point cloud twin space, the three-dimensional robot model and the corresponding three-dimensional obstacle model are simulated to move, the moving simulation refers to fixing the three-dimensional obstacle model, and moving the three-dimensional robot model to determine whether it can pass through the bottom area of the three-dimensional obstacle model, if it can, the corresponding obstacle object is marked as a passable obstacle, if it cannot, the corresponding obstacle object is marked as an impassable obstacle, and the same method is used to mark each obstacle object respectively.

[0018] It needs to be further explained that, in the specific implementation process, the process of acquiring the perception state of each working sub-region by using the image recognition model and the point cloud twin space includes: An image acquisition unit and a point cloud acquisition unit are respectively arranged on the robot, specifically a camera and a laser radar, the camera can acquire real-time image data around the robot, and the laser radar can acquire real-time point cloud data in the entire working area; The real-time image data is input into the image recognition model to determine whether there is a perception object around the robot, if there is, the object name is acquired, if not, no other operation is performed, whether there is a perception object in the working area is determined according to the real-time point cloud data, if there is, the object name is acquired, if not, no other operation is performed; Each perception object recognized by the image acquisition unit and the point cloud acquisition unit is synchronized to the GIS distribution map in real time, the perception state includes a to-be-cleaned area, a passable area, an impassable area, and a passable area, and the working sub-region where the cleaning object is located is marked as the to-be-cleaned area; The working sub-region where the obstacle object marked as a passable obstacle is located is marked as the passable area, the working sub-region where the obstacle object marked as an impassable obstacle is located is marked as the impassable area, and the other working sub-regions are marked as the passable area.

[0019] It needs to be further explained that, in the specific implementation process, the process of acquiring the historical running data of the robot and constructing a corresponding running evaluation model by using the LSTM neural network includes: The historical running data refers to the moving distance, cleaning object and corresponding energy consumption of the robot during a single cleaning work, and the single cleaning work refers to the process of moving the robot to a single to-be-cleaned area and completing the cleaning work; According to the moving distance, cleaning object and energy consumption in different historical running data, a corresponding running evaluation set is generated, and it is divided into a running training set and a running test set; A convolutional neural network is constructed, different moving distances and cleaning objects in the running training set are used as input data of the convolutional neural network, and corresponding energy consumptions in the running training set are used as output data of the convolutional neural network, and the convolutional neural network is trained to obtain an initial convolutional neural network; The running test set is used to verify the initial convolutional neural network, and the initial convolutional neural network with an output less than or equal to a preset running test error threshold is output as the running evaluation model.

[0020] It needs to be further explained that in the specific implementation process, the process of simulating the point cloud twin space to obtain the traversal energy consumption and using the running evaluation model to obtain the bypass energy consumption includes: If there is a traversable area between the real-time position of the robot and the closest to-be-cleaned area, a traversal path is constructed between the real-time position and the corresponding to-be-cleaned area using the Dijkstra algorithm, and the traversal path includes the to-be-cleaned area, the traversable area and the passable area; If not, a bypass path is constructed between the two using the Dijkstra algorithm, and the bypass path only includes the to-be-cleaned area and the passable area; For the corresponding to-be-cleaned area that only has a bypass path, the bypass path is taken as the optimal path, and the robot is controlled to clean according to the optimal path. For the corresponding to-be-cleaned area that has both a traversal path and a bypass path, the optimal path is selected by obtaining the traversal energy consumption and the bypass energy consumption respectively; In the point cloud twin space of the robot and the single obstacle object on the corresponding traversal path, the three-dimensional machine model and the three-dimensional obstacle model are adjusted to the corresponding direction and position in the actual application scene, and the working process of the three-dimensional machine model is simulated to obtain the simulation energy consumption required for passing through the bottom area of the three-dimensional obstacle model; The total distance of the robot moving along the traversal path to the corresponding to-be-cleaned area is taken as the traversal distance of the traversal path, and the moving energy consumption is obtained by inputting the traversal distance and the cleaning object of the corresponding to-be-cleaned area into the running evaluation model, and the sum of the simulation energy consumption and the moving energy consumption is taken as the traversal energy consumption of the traversal path; The total distance of the robot moving along the bypass path to the corresponding to-be-cleaned area is taken as the bypass distance of the bypass path, and the bypass energy consumption of the bypass path is obtained by inputting the bypass distance and the cleaning object of the corresponding to-be-cleaned area into the running evaluation model.

[0021] It needs to be further explained that in the specific implementation process, the process of selecting the corresponding optimal path according to the traversal energy consumption and the bypass energy consumption includes: The traversal energy consumption and the bypass energy consumption of the robot reaching the corresponding to-be-cleaned area are compared. If the traversal energy consumption is greater than or equal to the bypass energy consumption, the corresponding bypass path is taken as the optimal path. If the traversal energy consumption is less than the bypass energy consumption, the corresponding traversal path is taken as the optimal path; The traversal energy consumption and the bypass energy consumption of the robot reaching the corresponding to-be-cleaned area are continuously obtained, and the optimal path is obtained. The robot is controlled to clean according to the optimal path until all to-be-cleaned areas are cleaned.

[0022] The above examples are only used to illustrate the technical method of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present application can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present application.

Claims

1. A robot environment perception and autonomous obstacle avoidance system based on data fusion, characterized in that: Includes the following modules: The area division module is used to construct a GIS distribution map of the robot's working area and divide several working sub-areas in the GIS distribution map; The environmental perception module is used to obtain image data of different perception objects and build corresponding image recognition models, obtain point cloud data of the robot and different perception objects, and build corresponding point cloud twin spaces; The region labeling module is used to obtain the perception status of each working sub-region using the image recognition model and point cloud twin space; The operation evaluation module is used to obtain the robot's historical operation data and build a corresponding operation evaluation model using the LSTM neural network; The path optimization module is used to simulate the point cloud twin space to obtain the energy consumption of travel, and use the operation evaluation model to obtain the energy consumption of detour, and select the corresponding optimal travel path based on the energy consumption of travel and detour.

2. The robot environment perception and autonomous obstacle avoidance system based on data fusion according to claim 1 is characterized in that: The process of building a GIS distribution map and dividing the work sub-areas includes: Obtaining distribution information of the work area, including building outlines, interior walls and partitions, room dimensions, slopes and height differences, facility locations, furniture layouts, passageways and entrances and exits, and using GIS technology to construct a GIS distribution map of the work area based on this distribution information; Obtain the device size of the robot. The device size is used to reflect the area occupied by a single robot. In the GIS distribution map, the GIS distribution map is divided into several working sub-areas according to the device size of the robot. The area of ​​each working sub-area is larger than the device size of the robot.

3. The robot environment perception and autonomous obstacle avoidance system based on data fusion according to claim 2 is characterized in that: The process of acquiring image data and building an image recognition model includes: The perception objects refer to different objects that the robot can interact with when cleaning the ground in its working area, which are divided into obstacle objects and cleaning objects. The image data refers to multi-angle pictures of different perception objects; Generate corresponding image recognition sets based on the image data of different perceived objects and their object names, and divide them into image training sets and image test sets; Construct a convolutional neural network, use the image data of different perception objects in the image training set as the input data of the convolutional neural network, use the corresponding object names in the image training set as the output data of the convolutional neural network, and train the convolutional neural network to obtain an initial convolutional neural network; The initial convolutional neural network is model verified using the image test set, and the output of the initial convolutional neural network with a value less than or equal to the preset image test error threshold is used as the image recognition model.

4. The robot environment perception and autonomous obstacle avoidance system based on data fusion according to claim 3 is characterized in that: The process of acquiring point cloud data and constructing a point cloud twin space includes: Obtain the robot's point cloud data and build its 3D machine model based on the point cloud data. Obtain point cloud data for different obstacle objects and build corresponding 3D obstacle models for each. Integrate the 3D machine model with the individual 3D obstacle models to obtain a point cloud twin space of the two. In the point cloud twin space, the 3D obstacle model is fixed and the 3D machine model is moved to determine whether it can pass through the bottom area of ​​the 3D obstacle model. If it is possible, the corresponding obstacle object is marked as a passable obstacle; if it is not possible, the corresponding obstacle object is marked as an inpassable obstacle, and each obstacle object is marked separately.

5. The robot environment perception and autonomous obstacle avoidance system based on data fusion according to claim 4 is characterized in that: The process of obtaining the perception status of each working sub-area includes: An image acquisition unit and a point cloud acquisition unit are respectively set on the robot to obtain real-time image data around the robot and real-time point cloud data in the working area; Input real-time image data into the image recognition model to determine whether there are any sensed objects around the robot and obtain their names. Based on the real-time point cloud data, determine whether there are any sensed objects in the working area and obtain their names. The identified sensing objects are synchronized to the GIS distribution map in real time. The sensing status includes the area to be cleaned, the traversable area, the impassable area, and the traversable area. The working sub-area where the cleaning object is located is marked as the area to be cleaned, the working sub-area where the obstacle object marked as a passable obstacle is located is marked as a passable area, the working sub-area where the obstacle object marked as an inpassable obstacle is located is marked as an inpassable area, and all other working sub-areas are marked as passable areas.

6. The robot environment perception and autonomous obstacle avoidance system based on data fusion according to claim 5, characterized in that: The process of obtaining historical operation data and building an operation evaluation model includes: The historical operation data refers to the movement distance, cleaning objects and corresponding energy consumption of the robot during a single cleaning task. An operation evaluation set is generated based on the movement distance, cleaning objects and energy consumption in different historical operation data, and is divided into an operation training set and an operation test set; Construct a convolutional neural network, use different moving distances and cleaning objects in the running training set as input data of the convolutional neural network, use the corresponding energy consumption in the running training set as output data of the convolutional neural network, and train the convolutional neural network to obtain an initial convolutional neural network; The initial convolutional neural network is model verified using the running test set, and the initial convolutional neural network with a preset running test error threshold is output as the running evaluation model.

7. The robot environment perception and autonomous obstacle avoidance system based on data fusion according to claim 6, characterized in that: The process of obtaining the through-travel energy consumption and the detour energy consumption includes: If there is a traversable area between the robot's real-time position and the nearest area to be cleaned, the Dijkstra algorithm is used to construct a traversable path between the real-time position and the corresponding area to be cleaned. If there is no traversable area between the real-time position and the nearest area to be cleaned, the Dijkstra algorithm is used to construct a detour path between the two. For the area to be cleaned that only has a detour path, the detour path is used as the optimal path, and the robot is controlled to perform cleaning work according to the optimal path; For the corresponding cleaning area with both a through path and a detour path, in the point cloud twin space of the robot and a single obstacle object on the corresponding through path, the 3D machine model and the 3D obstacle model are adjusted to the direction and position corresponding to the actual application scenario; Use simulation software to simulate the working process of the three-dimensional machine model to obtain the simulated energy consumption required for it to pass through the bottom area of ​​the three-dimensional obstacle model; The total distance the robot travels along the travel path to the corresponding area to be cleaned is taken as the travel distance of the travel path. This is combined with the cleaning objects in the corresponding area to be cleaned and input into the operation evaluation model to obtain the movement energy consumption. The sum of the simulated energy consumption and the movement energy consumption is taken as the travel energy consumption of the travel path. The total distance the robot moves along the detour path to the corresponding area to be cleaned is taken as the detour distance of the detour path, and combined with the cleaning object of the corresponding area to be cleaned, they are input into the operation evaluation model to obtain the detour energy consumption of the detour path.

8. The robot environment perception and autonomous obstacle avoidance system based on data fusion according to claim 7 is characterized in that: The process of selecting the optimal route includes: Compare the energy consumption of the robot's travel and the energy consumption of its detour when it reaches the area to be cleaned. If the energy consumption of the travel is greater than or equal to the energy consumption of its detour, the detour path is taken as the optimal path. If the energy consumption of the travel is less than the energy consumption of its detour, the travel path is taken as the optimal path. The robot's real-time position and the energy consumption of traveling and bypassing corresponding to the area to be cleaned closest to it are continuously obtained, and the optimal passage path is obtained. The robot is controlled to perform cleaning work according to the optimal passage path until all areas to be cleaned are cleaned.

Images