Continuous autonomous driving perception and path planning method for unstructured scenes under limited computing power
By separating navigation control tasks and perception and obstacle avoidance planning tasks on mobile vehicles, and integrating rasterized digital elevation maps, the problems of small perception range and limited computing power on unstructured terrain are solved, and continuous travel is achieved Autonomous local obstacle avoidance improves mobility efficiency and safety.
Patent Information
- Application Number
- CN202210989378.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-17
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-08-17
AI Technical Summary
On unstructured terrain, due to the small perception range of the perception sensor and limited computing power of the computer, mobile vehicles lead to the safety risks and low movement efficiency of local obstacle avoidance planning, especially when patrolling the surface of extraterrestrial celestial bodies, environmental perception and obstacle avoidance are carried out. The planning and calculations are large, and usually require parking to be completed, which seriously affects the vehicle's movement efficiency.
Using the perception and path planning method of continuous autonomous driving in non-structural scenarios under limited computing power, the navigation control task and perception and obstacle avoidance planning tasks are separated, the navigation control task is executed in a fixed period, and the perception and obstacle avoidance planning tasks are executed in a non-determinal period, through the gate The grid-type digital elevation map and the suitability map are integrated to achieve continuous movement of vehicles and autonomous obstacle avoidance.
Improve the perception and movement efficiency of autonomous mobile vehicles, reduce the safety risks caused by short sensor perception distance, expand the effective coverage of the suitability map, improve the reliability of local obstacle avoidance planning, and realize the Continuous travel on unstructured terrain independently avoids obstacles.
Smart Images

Figure CN115454051B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a continuous autonomous driving perception and path planning method for unstructured scenarios under limited computing power, which is applied to mobile vehicles working on unstructured terrain, with a small sensor perception range and limited computer computing power, to achieve autonomous local obstacle avoidance during continuous travel. Background Art
[0002] For autonomous vehicles, in order to ensure the safety of the moving process, obstacle avoidance planning must be completed based on local perception.
[0003] Local perception is based on global planning and conducts detailed perception and modeling of the close-range local environment during movement. It also supplements / updates the perception information through fusion and completes obstacle extraction and trafficability analysis based on the modeling results.
[0004] Local obstacle avoidance planning relies on local perception information and follows a certain strategy to find a safe and collision-free path. Local obstacle avoidance planning takes safety and collision-free as its main goal, and the planning result is not required to be the optimal path.
[0005] The local obstacle avoidance planning of a mobile vehicle has a relatively obvious dependence on the perception range of the perception sensor and the computing resources on the vehicle. When the perception range of the perception sensor is difficult to cover the area around the vehicle, it will bring safety risks to the local obstacle avoidance; when the computing resources on the vehicle are limited and the computing power cannot meet the real-time requirements of the perception sensor data processing, it is generally necessary to reduce the vehicle's moving speed accordingly. Especially for exploration vehicles patrolling the surface of extraterrestrial bodies, since the working environment is an unknown and complex unstructured terrain, the terrain modeling and trafficability analysis are computationally intensive, and it is often necessary to stop to complete environmental perception and obstacle avoidance planning, which seriously affects the vehicle's mobility efficiency. Summary of the invention
[0006] The technical problem solved by the present invention is: In view of the shortcomings of the prior art, the present invention proposes a method for continuous autonomous driving perception and path planning in unstructured scenarios under limited computing power, which effectively overcomes the constraints of the small perception range of the sensor and the limited computing power of the on-board computer, and improves the perception and movement efficiency of a class of autonomous mobile vehicles.
[0007] The technical solution adopted by the present invention is: a method for continuous autonomous driving perception and path planning in unstructured scenes under limited computing power, including a navigation control task and a perception and obstacle avoidance planning task performed in sequence;
[0008] The navigation control task is a periodic task. It performs navigation and control operations in each cycle and determines whether to initiate the perception and obstacle avoidance tasks.
[0009] The perception and obstacle avoidance planning task is a non-periodic task initiated by the navigation control task; this task performs perception and planning operations and feeds back the status to the navigation control task after completion.
[0010] The navigation control tasks include:
[0011] Starting from the initial time t0, every T1 time period, that is, t k =t0+i*T1, i=0-k; k represents the total number of cycles in the process; perform one posture determination and motion control; the posture determination is used to determine the position and posture of the mobile vehicle in a specified navigation reference coordinate system, and the motion control is used to control the mobile vehicle to move according to the output result of the obstacle avoidance planning;
[0012] After completing the posture determination and motion control in each cycle, the starting conditions of the perception and obstacle avoidance tasks are judged, and based on the results, it is decided whether to start the perception and obstacle avoidance planning tasks.
[0013] The start conditions for the perception and obstacle avoidance tasks are as follows:
[0014] If: L u <V*T2,L u =LL e , then: start the perception and obstacle avoidance planning task; where L is the path length given by a perception planning task, L u is the remaining untraveled length of the planned path, L e is the length of the path traveled after the last perception planning task. At t0, L = L e =L u =0; V is the moving speed, T2 is the execution time of each perception planning task.
[0015] The perception and obstacle avoidance planning tasks include:
[0016] After receiving the start perception planning task instruction sent by the navigation control task, the configured perception sensor is controlled to collect data;
[0017] After completing data collection, the current position and attitude information obtained by the navigation control task is used to perform environmental modeling and obtain a rasterized digital elevation map;
[0018] After completing the environmental modeling, the accessibility analysis is completed on the digital elevation grid topographic map, and the accessibility suitability of each grid is calculated to obtain the current accessibility suitability map, that is, the local suitability map; when evaluating the accessibility suitability, if there is valid measurement data in the grid, the value range after the suitability evaluation is 0-1, 0 means impassable and 1 means safe to pass; if there is no valid measurement data in a grid, the value is assigned to 2, that is, invalid;
[0019] The historical trafficability analysis results obtained in the previous perception planning task are converted to the current location, and the suitability map is fused to obtain the fused traffic suitability map, that is, the global suitability map;
[0020] Path planning is performed on the fused global traffic suitability map, and the curvature and length of the planned path are given.
[0021] The suitability map is defined as follows:
[0022] Local suitability maps and global suitability maps are established separately. The local suitability map is established by each perception planning task based on the current perception data, and the global suitability map is obtained by fusing all previous local suitability maps.
[0023] The range sizes of the global suitability map and the local suitability map are determined in advance according to the effective perception range of the perception sensor, and both take the location of the autonomous driving unmanned system as the origin.
[0024] The local suitability map and the global suitability map are both in the form of grid maps, where the number of rows and columns of the local suitability map grid is r. M × M , the number of rows and columns of the global suitability map grid is r GM × GM , and r GM ≥n·r M , coefficient n>1; the traffic suitability evaluation value of each grid in the local suitability map and the global suitability map is represented in the array G cl |r M × M Sum G gl |r GM × GM In; each data in this array represents a size of r GM × GM The traffic suitability evaluation value of each grid in the grid map.
[0025] The local suitability map data G cl and global suitability map data G gl The initial value is impassable. After the perception planning task is executed, G is updated. cl and G gl Data in the area that overlaps with the local suitability map extent.
[0026] The fusion method of suitability map fusion is as follows:
[0027] Each perception planning task obtains the current local suitability map data G cl Then, define a new global suitability map data G gl_new, based on the current fitness map origin (x0, y0) and the fitness map origin (x L0 ,y L0 ), calculate the relative position relationship of the autonomous driving system during the two plannings, and then use G gl For G gl_new Assign values to the grid of the corresponding area, and finally let G gl =G gl_new , thus completing the translation of the global suitability map;
[0028] For the local suitability map data G cl Each grid (i l ,j l ), find its corresponding global suitability map data G gl The grid in (i g ,j g ), using G gl (i g ,j g ) for G cl (i l ,j l ) to obtain the fused local suitability map data G cl ;
[0029] Using the fused local suitability map data G cl The grid in (i l ,j l ), find its corresponding global suitability map data G gl The cells in (i g ,j g ), let G gl (i g ,j g )=G cl (i l ,j l ) to complete the global suitability map update.
[0030] The obtained fused local suitability map data G cl , as follows:
[0031] If G gl (i g ,j g ) is valid, that is, the array is in the range of 0 to 1, and G cl (i l ,j l ) is invalid, that is, the array is 2, then: G cl (i l ,j l )=m g0 Ggl (i g ,j g );
[0032] If G gl (i g ,j g ) is valid and G cl (i l ,j l ) is valid, then:
[0033] G cl (i l ,j l )=m g1 G gl (i g ,j g )+(1-m g1 )G cl (i l ,j l );
[0034] Among them, m g0 is the suitability forgetting factor, ranging from [0,1]; m g1 is the fitness weighting coefficient, and its value range is [0,1].
[0035] The advantages of the present invention compared with the prior art are:
[0036] (1) The present invention optimizes the local obstacle avoidance process of autonomous mobile vehicles, splits environmental perception and passability analysis and obstacle avoidance path selection into two independent links, and records the corresponding vehicle position and posture data respectively, and separates the suitability map origin and the planned starting position point, so as to achieve continuous mobile autonomous obstacle avoidance under conditions of limited computing power, effectively improving the vehicle's mobility efficiency.
[0037] (2) The present invention improves the suitability map fusion algorithm, making the map coordinates independent of the vehicle's heading and avoiding posture conversion during the fusion process, so that the fusion of suitability maps generated by different sensing positions can be easily realized, thereby expanding the effective coverage range of the suitability map, reducing the safety risks of the vehicle due to the short sensing distance of the sensor, and improving the reliability of local obstacle avoidance planning.
[0038] (3) The present invention is also applicable to mobile vehicles operating on unstructured terrain, with small sensor sensing range and limited computer computing power, and can achieve autonomous local obstacle avoidance during continuous movement. It has a wide range of applications and strong practicality. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION
[0040] The present invention discloses a method for continuous autonomous driving perception and path planning in unstructured scenes under limited computing power. The method is applicable to mobile vehicles working on unstructured terrain, where both the perception sensor and the computer computing power are constrained. The perception sensor is generally a stereo camera with a large field of view and a short focal length or a depth camera. The effective perception distance S meters of the sensor, the vehicle moving speed V meters per second, and the time T seconds for the computer to complete environmental perception and trafficability analysis satisfy the relationship S>6VT, where the effective perception distance refers to the distance within which the accuracy of the perception data of the sensor can meet the accuracy requirements of environmental modeling and trafficability analysis. The specific implementation method is as follows:
[0041] (i) Decompose the local obstacle avoidance scheme into two parallel tasks: navigation control task and perception and obstacle avoidance planning task.
[0042] (II) The navigation control task is as follows:
[0043] (1) The navigation control task periodically performs posture determination and motion control functions. Posture determination is used to determine the position and posture of the mobile vehicle in a specified navigation reference coordinate system, and motion control is used to control the mobile vehicle to move according to the output results of the obstacle avoidance plan.
[0044] (2) The navigation control task needs to notify the perception and obstacle avoidance planning task to perform "data collection" before starting to move, or T seconds before the end of the current obstacle avoidance planning movement time, or when the remaining mileage of the current obstacle avoidance planning movement is V*T meters, and at the same time pass the current vehicle position and posture information to the perception and obstacle avoidance planning task.
[0045] (3) After receiving the "passability analysis" completion signal from the perception and obstacle avoidance planning task, the navigation control task needs to notify the perception and obstacle avoidance planning task, execute the "planning path selection", and at the same time pass the current vehicle position and posture information to the perception and obstacle avoidance planning task.
[0046] (4) After receiving the “planning path selection” result of the perception and obstacle avoidance planning task, the navigation control task needs to update the motion control strategy and control the vehicle to move according to the new planning output result.
[0047] (III) The tasks of perception and obstacle avoidance planning are as follows:
[0048] (1) After receiving the "data collection" instruction of the navigation control task, control the configured perception sensor to collect data.
[0049] (2) After completing "data collection", the data is processed. According to the collection time position and attitude information sent by the navigation control task, the processed three-dimensional point cloud data is converted to the coordinate system required for planning, and environmental modeling is performed to establish a rasterized digital elevation map.
[0050] (3) After completing the "environmental modeling", the accessibility analysis is completed on the digital elevation grid topographic map, and the suitability evaluation value of each grid is calculated. At the same time, the historical accessibility analysis results are converted to the current planning coordinate system and the suitability map is fused. After the suitability map fusion is completed, the "accessibility analysis" completion signal is sent to the navigation control task.
[0051] (4) After receiving the "planning path selection" instruction from the navigation control task, on the current suitability map, the current planning position sent by the navigation control task is used as the starting point, combined with the posture information, an appropriate local obstacle avoidance planning algorithm is selected to generate the current planning path result, which is fed back to the navigation control task. The longest planned moving distance generally does not exceed S / 3.
[0052] (IV) The specific strategies for suitability map fusion are as follows:
[0053] (1) Establish local suitability maps and global suitability maps respectively. The local suitability map is used for trafficability analysis and local obstacle avoidance planning based on current perception data, and the global suitability map is used to realize the fusion of historical suitability maps.
[0054] (2) Generation of local suitability map, the specific contents are as follows:
[0055] Local suitability map data G cl For a size of r M × M The physical meaning of this array is an array centered at the origin of the suitability map coordinate system and containing r M × M The local suitability map of each grid stores an evaluation value (i.e., suitability) representing whether the corresponding terrain is suitable for vehicle traffic.
[0056] The direction of the local suitability map is consistent with the navigation reference coordinate system and does not change with the vehicle heading. Assume that the resolution of each grid in the local suitability map array is l c , then r M × c Generally, it is not less than twice the effective sensing distance S of the perception sensor. The origin of the local suitability map is the location of the vehicle at the moment of environmental perception. Therefore, no matter which direction the vehicle is heading when the environment is perceived, the local suitability map can cover the effective sensing range of the perception sensor.
[0057] For each grid in the local suitability map, a terrain block with the grid as the center and the size of the vehicle envelope is taken. The terrain feature data in the terrain block is calculated using the terrain feature data output by the environmental modeling, and the suitability evaluation values such as slope, roughness, and height difference of the terrain block are calculated respectively to obtain the suitability of the grid in the local suitability map.
[0058] (3) Suitability map fusion, the specific contents are as follows:
[0059] This section uses the global suitability map data G gl The local suitability map data G generated during the current planning cl Fusion, while using the fused G cl For G gl to update.
[0060] Global suitability map data G gl For a size of r GM × GM The physical meaning of this array is the same as that of the local suitability map data G cl Same, but the map size is larger than G cl Large, so it can be used to store G generated by multiple planning cl , generally GM Should be no less than 2r M .
[0061] First, define a new global suitability map data G gl_new , based on the current fitness map origin (x0, y0) and the fitness map origin (x L0 ,y L0 ), calculate the relative position relationship of the vehicle during the two previous plannings, and then use G gl For G gl_new Assign values to the grid of the corresponding area, and finally let G gl =G gl_new , thus completing the translation of the global suitability map.
[0062] Secondly, for the local suitability map data G cl The grid in (i l ,j l ), find its corresponding global suitability map data G gl The cells in (i g ,j g ), using G gl For G cl To integrate, specifically:
[0063] If G gl (ig ,j g ) is valid and G cl (i l ,j l ) is invalid, then: G cl (i l ,j l )=m g0 G gl (i g ,j g );
[0064] If G gl (i g ,j g ) is valid and G cl (i l ,j l ) is valid, then:
[0065] G cl (i l ,j l )=m g1 G gl (i g ,j g )+(1-m g1 )G cl (i l ,j l );
[0066] Among them, m g0 is the suitability forgetting factor, the value range is [0,1], and it can generally be taken as 0.95; m g1 is the suitability weighted coefficient, the value range is [0,1], and it can generally be taken as 0.9.
[0067] Finally, using the fused local suitability map data G cl The grid in (i l ,j l ), find its corresponding global suitability map data G gl The cells in (i g ,j g ), let G gl (i g ,j g )=G cl (i l ,j l ) to complete the global suitability map update.
[0068] According to the following Figure 1The present invention is described in detail. A planetary surface patrol probe is equipped with a pair of wide-angle fisheye obstacle avoidance cameras for local perception. The effective perception distance is 3m. The maximum moving speed of the designed vehicle is 200m / h, which is about 0.056m / s. The total time required to complete environmental perception and trafficability analysis does not exceed 5s.
[0069] The specific implementation method of the patrol detector's continuous movement local obstacle avoidance is as follows:
[0070] (i) Decompose the local obstacle avoidance scheme into two parallel tasks: navigation control task and perception and obstacle avoidance planning task.
[0071] (II) The navigation control task is as follows:
[0072] (1) The navigation control task periodically performs the posture determination and motion control functions. The posture determination is used to determine the position and posture of the patrol probe in the specified north-east-ground coordinate system, and the motion control is used to control the patrol probe to move according to the output results of the obstacle avoidance plan.
[0073] (2) Before the navigation control task starts moving, or when the remaining mileage of the currently executed obstacle avoidance planning is 0.3 meters, the perception and obstacle avoidance planning task is notified to perform "data collection" and transmit the current patrol detector position and posture information to the perception and obstacle avoidance planning task.
[0074] (3) After receiving the “passability analysis” completion signal from the perception and obstacle avoidance planning task, the navigation control task needs to notify the perception and obstacle avoidance planning task, execute the “planning path selection”, and at the same time pass the patrol probe position and attitude information to the perception and obstacle avoidance planning task.
[0075] (4) After receiving the “planning path selection” result of the perception and obstacle avoidance planning task, the navigation control task needs to update the motion control strategy and control the patrol detector to move according to the new planning output result.
[0076] (II) The tasks of perception and obstacle avoidance planning are as follows:
[0077] (1) After receiving the "data collection" instruction of the navigation control task, control the obstacle avoidance camera to collect images.
[0078] (2) After completing "data collection", the image is processed. According to the collection time position and attitude information sent by the navigation control task, the processed three-dimensional point cloud data is converted to the north-east-ground coordinate system required for planning, and environmental modeling is performed to establish a rasterized digital elevation map.
[0079] (3) After completing the "environmental modeling", the accessibility analysis is completed on the digital elevation grid topographic map, and the suitability evaluation value of each grid is calculated. At the same time, the historical accessibility analysis results are converted to the current planning coordinate system and the suitability map is fused. After the suitability map fusion is completed, the "accessibility analysis" completion signal is sent to the navigation control task.
[0080] (4) After receiving the "planning path selection" instruction from the navigation control task, on the current suitability map, the current planning time position sent by the navigation control task is used as the starting point, combined with the attitude information, the GESTALT algorithm is selected, and the curvature and arc length of the optimal arc path are selected from a set of alternative arc paths as the current planning path result, which is fed back to the navigation control task. The planned arc length does not exceed 1 meter.
[0081] (III) The specific strategies for suitability map fusion are as follows:
[0082] (1) Establish local suitability maps and global suitability maps respectively. The local suitability map is used for trafficability analysis and local obstacle avoidance planning based on current perception data, and the global suitability map is used to realize the fusion of historical suitability maps.
[0083] (2) Generation of local suitability map, the specific contents are as follows:
[0084] Local suitability map data G cl is an array of size 50×50, and the resolution of each grid is 0.2 meters.
[0085] The direction of the local suitability map is consistent with the navigation reference coordinate system and does not change with the vehicle heading.
[0086] For each grid in the local suitability map, a terrain block with the grid as the center and the size of the vehicle envelope is taken. The terrain feature data in the terrain block is calculated using the terrain feature data output by the environmental modeling, and the suitability evaluation values such as slope, roughness, and height difference of the terrain block are calculated respectively. The minimum value of the three is taken as the suitability of the grid.
[0087] (3) Suitability map fusion, the specific contents are as follows:
[0088] This section uses the global suitability map data G gl The local suitability map data G generated during the current planning cl Fusion, while using the fused G cl For G gl to update.
[0089] Global suitability map data G gl is an array of size 100×100.
[0090] First, define a new global suitability map data G gl_new , based on the current fitness map origin (x0, y0) and the fitness map origin (x L0 ,y L0 ), calculate the relative position relationship of the vehicle during the two previous plannings, and then use G gl For G gl_new Assign values to the grid of the corresponding area, and finally let G gl =G gl_new , thus completing the translation of the global suitability map.
[0091] Secondly, for the local suitability map data G cl The grid in (i l ,j l ), find its corresponding global suitability map data G gl The cells in (i g ,j g ), using G gl For G cl To integrate, specifically:
[0092] If G gl (i g ,j g ) is valid and G cl (i l ,j l ) is invalid, then: G cl (i l ,j l )=m g0 G gl (i g ,j g );
[0093] If G gl (i g ,j g ) is valid and G cl (i l ,j l ) is valid, then:
[0094] G cl (i l ,j l )=m g1 G gl (i g ,j g )+(1-m g1 )G cl (i l ,j l );
[0095] Among them, mg0 is the suitability forgetting factor, which is set to 0.95; m g1 is the suitability weighting coefficient, and its value is 0.9.
[0096] Finally, using the fused local suitability map data G cl The grid in (i l ,j l ), find its corresponding global suitability map data G gl The cells in (i g ,j g ), let G gl (i g, j g )=G cl (i l, j l ) to complete the global suitability map update.
[0097] The method of the present invention is also applicable to mobile vehicles working on unstructured terrain, with small sensor perception range and limited computer computing power, to achieve autonomous local obstacle avoidance during continuous movement.
[0098] The above description is only the best specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.
Claims
1. A method for continuous autonomous driving perception and path planning in unstructured scenes under limited computing power, characterized in that: It includes navigation control tasks and perception and obstacle avoidance planning tasks that are performed sequentially; The navigation control task is a periodic task. It performs navigation and control operations in each cycle and determines whether to initiate the perception and obstacle avoidance tasks. The perception and obstacle avoidance planning task is a non-periodic task initiated by the navigation control task; this task performs perception and planning operations and feeds back the status to the navigation control task after completion.
2. According to claim 1, a method for continuous autonomous driving perception and path planning in unstructured scenes under limited computing power, characterized in that: The navigation control tasks include: Starting from the initial time t0, every T1 time period, that is, t k =t0+i*T1, i=0-k; k represents the total number of cycles in the process; perform one posture determination and motion control; the posture determination is used to determine the position and posture of the mobile vehicle in a specified navigation reference coordinate system, and the motion control is used to control the mobile vehicle to move according to the output result of the obstacle avoidance planning; After completing the posture determination and motion control in each cycle, the starting conditions of the perception and obstacle avoidance tasks are judged, and based on the results, it is decided whether to start the perception and obstacle avoidance planning tasks.
3. The method for continuous autonomous driving perception and path planning in unstructured scenes under limited computing power according to claim 2 is characterized in that: The start conditions for the perception and obstacle avoidance tasks are as follows: If: L u <V*T2,L u =LL e , then: start the perception and obstacle avoidance planning task; where L is the path length given by a perception planning task, L u is the remaining untraveled length of the planned path, L e is the length of the path traveled after the last perception planning task. At t0, L = L e =L u =0; V is the moving speed, T2 is the execution time of each perception planning task.
4. According to claim 2, a method for continuous autonomous driving perception and path planning in unstructured scenes under limited computing power, characterized in that: The perception and obstacle avoidance planning tasks include: After receiving the start perception planning task instruction sent by the navigation control task, the configured perception sensor is controlled to collect data; After completing data collection, the current position and attitude information obtained by the navigation control task is used to perform environmental modeling and obtain a rasterized digital elevation map; After completing the environmental modeling, the accessibility analysis is completed on the rasterized digital elevation map, and the accessibility suitability of each grid is calculated to obtain the current accessibility suitability map, that is, the local suitability map; the accessibility suitability evaluation, if there is valid measurement data in the grid, the value range after the suitability evaluation is 0-1, 0 means impassable and 1 means safe to pass; if there is no valid measurement data in a grid, the value is assigned to 2, that is, invalid; The historical trafficability analysis results obtained in the previous perception planning task are converted to the current location, and the suitability map is fused to obtain the fused traffic suitability map, that is, the global suitability map; Path planning is performed on the fused global suitability map, and the status is fed back to the navigation control task after completion.
5. The method for continuous autonomous driving perception and path planning in unstructured scenes under limited computing power according to claim 4, characterized in that: The suitability map is defined as follows: The local suitability map is established by each perception planning task based on the current perception data, and the global suitability map is obtained by fusing all the local suitability maps.
6. The method for continuous autonomous driving perception and path planning in unstructured scenes under limited computing power according to claim 5, characterized in that: The range sizes of the global suitability map and the local suitability map are determined in advance according to the effective perception range of the perception sensor, and both take the location of the autonomous driving unmanned system as the origin.
7. The method for continuous autonomous driving perception and path planning in unstructured scenes under limited computing power according to claim 6, characterized in that: The local suitability map and the global suitability map are both in the form of grid maps, where the number of rows and columns of the local suitability map grid is r. M × M , the number of rows and columns of the global suitability map grid is r GM × GM , and r GM ≥n·r M , coefficient n>1; the traffic suitability evaluation value of each grid in the local suitability map and the global suitability map is represented in the array G cl |r M × M Sum G gl |r GM × GM In; each data in this array represents a size of r GM × GM The traffic suitability evaluation value of each grid in the grid map.
8. The method for continuous autonomous driving perception and path planning in unstructured scenes under limited computing power according to claim 7, characterized in that: The local suitability map data G cl and global suitability map data G gl The initial value is impassable. After the perception planning task is executed, G is updated. cl and G gl Data in the area that overlaps with the local suitability map extent.
9. The method for continuous autonomous driving perception and path planning in unstructured scenes under limited computing power according to claim 8, characterized in that: The fusion method of suitability map fusion is as follows: Each perception planning task obtains the current local suitability map data G cl Then, define a new global suitability map data G gl_new , based on the current fitness map origin (x0, y0) and the fitness map origin (x L0 ,y L0 ), calculate the relative position relationship of the autonomous driving system during the two plannings, and then use G gl For G gl_new Assign values to the grid of the corresponding area, and finally let G gl =G gl_new , thus completing the translation of the global suitability map; For the local suitability map data G cl Each grid (i l ,j l ), find its corresponding global suitability map data G gl The grid in (i g ,j g ), using G gl (i g ,j g ) for G cl (i l ,j l ) to obtain the fused local suitability map data G cl ; Using the fused local suitability map data G cl The grid in (i l ,j l ), find its corresponding global suitability map data G gl The cells in (i g ,j g ), let G gl (i g ,j g )=G cl (i l ,j l ) to complete the global suitability map update.
10. The method for continuous autonomous driving perception and path planning in unstructured scenes under limited computing power according to claim 9, characterized in that: The obtained fused local suitability map data G cl , as follows: If G gl (i g ,j g ) is valid, that is, the array is in the range of 0 to 1, and G cl (i l ,j l ) is invalid, that is, the array is 2, then: G cl (i l ,j l )=m g0 G gl (i g ,j g ); If G gl (i g ,j g ) is valid and G cl (i l ,j l ) is valid, then: G cl (i l ,j l )=m g1 G gl (i g ,j g )+(1-m g1 )G cl (i l ,j l ); Among them, m g0 is the suitability forgetting factor, ranging from [0,1]; m g1 is the fitness weighting coefficient, and its value range is [0,1].
