High-real-time autonomous obstacle avoidance method for outdoor unmanned tractor

Abstract

The invention belongs to the technical field of outdoor tractor autonomous navigation, and particularly relates to a high-real-time outdoor unmanned tractor autonomous obstacle avoidance method which comprises the following steps: acquiring a local planning curve and a collision detection constraint table; after an obstacle avoidance demand is triggered, transverse and longitudinal sampling is carried out along a lane line, polynomial planning is generated from the current position of the tractor to a sampling point, a standby obstacle avoidance guide line along a driving reference line within a set lane width range is generated, and obstacle interference detection is carried out; selecting an optimal reference line as a lane guide line for the lane passing through obstacle detection according to the distance from the reference line and the driving cost, and selecting a guide point on the selected lane guide line; a local planning line is screened out; and according to the guide point and the current attitude of the tractor, evaluating and scoring the local planning line generated offline according to the transverse and longitudinal cost of driving, and according to the obtained local planning line to be executed, obtaining the angular velocity of the center of the vehicle body in the next planning period and sending the angular velocity to the tractor for execution.

Description

Technical Field

[0001] The invention belongs to the technical field of outdoor tractor autonomous navigation, in particular to a highly real-time outdoor unmanned tractor autonomous obstacle avoidance method. Background Art

[0002] Autonomous obstacle avoidance technology is an important part of the robot's autonomous navigation. It is currently mainly used in autonomous driving and service robots. However, with the rapid development of artificial intelligence, the intelligent attributes of unmanned tractors are becoming stronger and stronger, the application scenarios are increasing, and various customized needs are also increasing. In order to balance safety and efficiency, the demand for autonomous obstacle avoidance functions of unmanned tractors is becoming increasingly urgent. Domestic and foreign scholars have done a lot of work in the research of trailer trajectory planning, but all of them complete the planning by solving nonlinear optimization models based on trailer kinematics and dynamics models.

[0003] However, due to the need to tow multiple material trucks behind the tractor, its overall size is long, its running trajectory is difficult to estimate, and there are many uncertain factors in the outdoor environment. As a result, the collision constraints between the trailer and obstacles in the environment are large in scale and highly non-convex and nonlinear, making the solution efficiency extremely low. With limited computing power, it is difficult to apply it to real vehicles. To solve this problem, a high-real-time outdoor unmanned tractor autonomous obstacle avoidance method needs to be proposed. Summary of the invention

[0004] The purpose of the present invention is to provide a high real-time outdoor unmanned tractor autonomous obstacle avoidance method based on the consideration of the safety, efficiency, algorithm real-time and computing power constraints of the outdoor tractor autonomous obstacle avoidance, so as to improve the safety and efficiency of the outdoor unmanned tractor operation and meet the autonomous obstacle avoidance function requirements of the unmanned tractor.

[0005] The technical solution adopted by the present invention to achieve the above-mentioned purpose is: a high-real-time outdoor unmanned tractor autonomous obstacle avoidance method, comprising the following steps:

[0006] 1) The offline calculation module obtains the local planning curve and the collision detection constraint table according to the vehicle parameters and vehicle operating environment information in the call database, and adds the local planning curve and the collision detection constraint table as a configuration file to the online working module;

[0007] 2) After the obstacle avoidance requirement is triggered, based on the driving reference line and the road width limit, horizontal and vertical sampling is performed along the lane line; then a polynomial planning is generated from the current position of the tractor to the sampling point, and finally a backup obstacle avoidance guide line is generated along the driving reference line within the set lane width;

[0008] 3) Perform obstacle interference detection on the generated road guide line to exclude lanes occupied by obstacles; select the lanes that pass the obstacle detection according to the distance from the reference line and the driving cost, select the optimal reference line as the lane guide line, and select the guide point on the selected lane guide line;

[0009] 4) Look up the table and perform coordinate transformation according to the current position of the tractor to obtain the estimated running trajectory of the local plan in the world coordinate system. Combined with the obstacle information collected by the perception module, the local planning line that passes the collision detection constraint is selected through the obstacle detection constraint table;

[0010] 5) Evaluate and score the offline generated local planning line according to the lateral and longitudinal costs of the guide point and the current posture of the tractor, and obtain the local planning line to be executed;

[0011] 6) Finally, according to the local planning line, the angular velocity of the vehicle body center in the next planning period is obtained and sent to the tractor body control module for execution.

[0012] The vehicle parameters include: vehicle body motion parameters, vehicle type, outline shape and number of trailers;

[0013] The vehicle operating environment information includes: map data, road information, and obstacle locations.

[0014] The offline calculation of the local planning curve and the collision detection constraint table is specifically as follows:

[0015] 1-1) The offline calculation module plans the path of the vehicle according to the vehicle parameters and the vehicle operating environment information. First, the vehicle is sampled at the local planning end in the vehicle body coordinate system. The terminal sampling is divided into distance and angle. The length of the corresponding local planning line is selected according to the parameters such as the expected running speed of the vehicle. Then, the terminal vehicle head direction is sampled according to the vehicle's movement ability and other restrictions. In this way, a series of terminal state points can be obtained.

[0016] The local plan is a series of position points that the vehicle follows in a short period of time;

[0017] 1-2) Generate a local planning curve: Based on the sampling set at the end of the local planning path, use polynomial interpolation or optimization algorithm to generate a smooth curve so that the vehicle can travel along this curve;

[0018] 1-3) Collision detection constraint table: Based on vehicle parameters and environmental information, interaction with map data, obstacle detection, and planning curve constraints, a collision detection constraint table is calculated, i.e., obstacles or collision risk areas that the vehicle should avoid during driving;

[0019] 1-4) Performance optimization: Optimize the generated local planning curves and collision detection constraint tables to ensure that the vehicle has good performance and safety in actual operation.

[0020] The polynomial planning is generated from the current position of the tractor to the sampling point, and finally a spare obstacle avoidance guide line is generated along the driving reference line within the set lane width, specifically:

[0021] 2-1) Obtain the road reference line to be driven in the Cartesian coordinate system, and then obtain the frenet coordinate system of the driving reference line according to the conversion formula from the frenet coordinate system to the Cartesian coordinate system. In the frenet coordinate system, the current position of the tractor is taken as the state starting point, the sampling point is taken as the state end point, and the speed and acceleration of the end point are both set to 0;

[0022] 2-2) Generate a 5th-order polynomial trajectory based on the starting and ending point states. After obtaining the lateral and longitudinal trajectories, perform the conversion from the Frenet coordinate system to the Cartesian coordinate system, and finally generate a backup obstacle avoidance guide line along the driving reference line within the set lane width.

[0023] The Frenet coordinate system is used to describe the position of the vehicle along the path, and the Cartesian coordinate system describes the absolute position of the vehicle on the plane;

[0024] The step 3) is specifically:

[0025] 3-1) Perform obstacle interference detection on the generated road guide lines to exclude lanes occupied by obstacles. If all the generated local guide lines pass the interference detection, it is considered that the obstacle avoidance is completed and a completion signal is returned to the vehicle control program. Otherwise, continue to execute;

[0026] 3-2) The lane guide line that passes the obstacle detection is selected according to the distance from the reference line, the distance obstacle cost and the driving cost, and the optimal reference line is selected as the lane guide line, and the guide point is selected on the selected lane guide line.

[0027] The step 3-2) comprises the following steps:

[0028] 3-2-1) Calculate the lateral offset cost: The absolute value of the lateral offset of the end sampling position corresponding to the local obstacle avoidance reference line from the reference line is taken as the lateral cost. The greater the lateral distance from the reference line, the greater the lateral cost.

[0029] 3-2-2) Calculate the collision risk cost: The minimum distance derivative of the local obstacle avoidance line from the obstacle is taken as the collision cost. The closer the distance to the obstacle, the greater the collision risk.

[0030] 3-2-3) Calculate the driving cost: The driving cost is determined based on whether the current position of the vehicle is consistent with the direction of the driving reference line. If it is consistent, the cost is 0, otherwise it is 1;

[0031] 3-2-4) According to actual needs, appropriate weights are assigned to different costs and then added to obtain the final local obstacle avoidance reference line evaluation function, and the local obstacle avoidance reference line with the minimum cost is selected as the final driving guide line.

[0032] The step 4) is specifically:

[0033] 4-1) Retrieve the optional partial executable trajectory from the database;

[0034] 4-2) According to the current position of the vehicle and the obstacle information, the coordinate position of the obstacle point cloud and shape information is discretized and converted to the vehicle coordinate system according to actual needs to obtain the grid coordinates in the vehicle coordinate system;

[0035] 4-3) The corresponding obstacle coordinates are projected to the offline calculation module, and the alternative local planning lines passing through the grid occupied by the obstacle are excluded from the obtained collision detection table, and finally a feasible local planning line without collision risk is obtained.

[0036] The offline generated local planning line is evaluated and scored according to the lateral and longitudinal costs of the guide point and the current posture of the tractor, and the local planning line to be executed is obtained, specifically:

[0037] 5-1) Calculate the absolute value of the deviation between the end position angle of all candidate local planning lines that pass the collision detection and the current vehicle head direction as the lateral driving cost 1;

[0038] 5-2) Calculate the absolute value of the angle deviation between the end position of all candidate local planning lines that pass the collision detection and the optimal local planning line selected in the previous cycle as the lateral cost 2;

[0039] 5-3) Calculate the next cycle state value corresponding to all candidate local planning lines that pass the collision detection, and use the linear acceleration and the absolute value of the jerk as the longitudinal driving cost;

[0040] 5-4) The horizontal cost 1, horizontal cost 2 and vertical cost are added together as the cost evaluation function of the overall local planning line. All candidate lines are scored according to the evaluation function. The higher the cost value, the lower the score. The local planning line with the highest score is selected.

[0041] The step 6) is specifically:

[0042] 6-1) According to the guide point, the current posture of the tractor and the local curve of the previous cycle, the local curve with the smallest angle cost is selected as the local planning line to be executed;

[0043] 6-2) Finally, the angular velocity of the vehicle body center in the next planning period is obtained based on the local planning line and sent to the tractor body control module for execution.

[0044] The present invention has the following beneficial effects and advantages:

[0045] 1. The present invention uses lane-based obstacle avoidance guide lines and offline local planning to achieve autonomous obstacle avoidance, avoiding the complexity of solving using a nonlinear optimization model, without the risk of falling into local optimality, and is safer and more reliable;

[0046] 2. The present invention uses offline calculation and table lookup to generate local planning lines, which greatly reduces the algorithm's demand for computing power and reduces the cost of implementation;

[0047] 3. The present invention uses offline calculation of the running area occupied by the local planning line for collision detection. Since collision detection requires real-time calculation, this operation can effectively reduce the time required for collision detection and meet the real-time requirements of the algorithm;

[0048] 4. The present invention adopts the operation of expanding the collision detection operation area according to the number and size of trailers, which is simple and easy to operate and can meet the needs of trailer obstacle avoidance;

[0049] 5. The present invention adopts an offline calculation method to perform spatial sampling of different vehicle head directions and target postures to generate a second-order smooth curve. The parameters can be configured according to the movement capabilities of different vehicle models (whether they can spin and reverse, etc.) to generate local planning lines that meet different needs, and the deployment is flexible;

[0050] 6. The present invention uses a lane guide line generated along the driving path based on the Frenet coordinate system to perform lane-based obstacle interference detection, which is very consistent with the behavioral habits of outdoor vehicles driving along the lanes. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Autonomous obstacle avoidance workflow diagram of the present invention;

[0052] Figure 2 The display diagram of the offline calculation result of the local planning line when the vehicle head is facing at 0° of the present invention;

[0053] Figure 3 Schematic diagram of the lane obstacle avoidance guide line of the present invention. DETAILED DESCRIPTION

[0054] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0055] like Figure 1As shown, it is a flowchart of the autonomous obstacle avoidance process of the present invention. The present invention provides a high-real-time outdoor unmanned tractor autonomous obstacle avoidance method, which is characterized by comprising the following steps:

[0056] 1) The offline calculation module obtains the local planning curve and the collision detection constraint table according to the vehicle parameters and vehicle operating environment information in the call database, and adds the local planning curve and the collision detection constraint table as a configuration file to the online working module;

[0057] Among them, vehicle parameters include: vehicle body motion parameters, vehicle type, outline shape and number of trailers; vehicle operating environment information includes: map data, road information, and obstacle locations.

[0058] The local planning curve and collision detection constraint table are calculated offline, specifically:

[0059] 1-1) The offline calculation module plans the path of the vehicle according to the vehicle parameters and the vehicle operating environment information. First, the vehicle is sampled at the local planning end in the vehicle body coordinate system. The terminal sampling is divided into distance and angle. The length of the corresponding local planning line is selected according to the parameters such as the expected running speed of the vehicle. Then, the terminal vehicle head direction is sampled according to the vehicle's movement ability and other restrictions, such as Figure 2 As shown, this is a display diagram of the offline calculation results of the local planning line when the vehicle head is facing 0°, so that a series of terminal state points can be obtained;

[0060] The local plan is a series of position points that the vehicle follows in a short period of time;

[0061] 1-2) Generate a local planning curve: Based on the sampling set at the end of the local planning path, use polynomial interpolation or optimization algorithm to generate a smooth curve so that the vehicle can travel along this curve;

[0062] 1-3) Collision detection constraint table: Based on vehicle parameters and environmental information, interaction with map data, obstacle detection, and planning curve constraints, a collision detection constraint table is calculated, i.e., obstacles or collision risk areas that the vehicle should avoid during driving;

[0063] 1-4) Performance optimization: Optimize the generated local planning curves and collision detection constraint tables to ensure that the vehicle has good performance and safety in actual operation.

[0064] like Figure 3 , which is a schematic diagram of the lane obstacle avoidance guide line of the present invention;

[0065] 2) After the obstacle avoidance requirement is triggered, based on the driving reference line and the road width limit, horizontal and vertical sampling is performed along the lane line; then a polynomial planning is generated from the current position of the tractor to the sampling point, and finally a backup obstacle avoidance guide line is generated along the driving reference line within the set lane width;

[0066] Generate a polynomial plan from the current position of the tractor to the sampling point, and finally generate a backup obstacle avoidance guide line along the driving reference line within the set lane width, specifically:

[0067] The Frenet coordinate system is used to describe the position of the vehicle along the path, and the Cartesian coordinate system describes the absolute position of the vehicle on the plane;

[0068] 2-1) Obtain the road reference line to be driven in the Cartesian coordinate system, and then obtain the frenet coordinate system of the driving reference line according to the conversion formula from the frenet coordinate system to the Cartesian coordinate system. In the frenet coordinate system, the current position of the tractor is taken as the state starting point, the sampling point is taken as the state end point, and the speed and acceleration of the end point are both set to 0;

[0069] 2-2) Generate a 5th-order polynomial trajectory based on the starting and ending point states. After obtaining the lateral and longitudinal trajectories, perform the conversion from the Frenet coordinate system to the Cartesian coordinate system, and finally generate a backup obstacle avoidance guide line along the driving reference line within the set lane width.

[0070] 3) Perform obstacle interference detection on the generated road guide line to exclude lanes occupied by obstacles; select the lanes that pass the obstacle detection according to the distance from the reference line and the driving cost, select the optimal reference line as the lane guide line, and select the guide point on the selected lane guide line;

[0071] 3-1) Perform obstacle interference detection on the generated road guide lines to exclude lanes occupied by obstacles. If all the generated local guide lines pass the interference detection, it is considered that the obstacle avoidance is completed and a completion signal is returned to the vehicle control program. Otherwise, continue to execute;

[0072] 3-2) The lane guide line that passes the obstacle detection is selected according to the distance from the reference line, the distance obstacle cost and the driving cost, and the optimal reference line is selected as the lane guide line. The guide point is selected on the selected lane guide line, such as Figure 3 As shown, the pink line.

[0073] 3-2-1) Calculate the lateral offset cost: The absolute value of the lateral offset of the end sampling position corresponding to the local obstacle avoidance reference line from the reference line is taken as the lateral cost. The greater the lateral distance from the reference line, the greater the lateral cost.

[0074] 3-2-2) Calculate the collision risk cost: The minimum distance derivative of the local obstacle avoidance line from the obstacle is taken as the collision cost. The closer the distance to the obstacle, the greater the collision risk.

[0075] 3-2-3) Calculate the driving cost: The driving cost is determined based on whether the current position of the vehicle is consistent with the direction of the driving reference line. If it is consistent, the cost is 0, otherwise it is 1;

[0076] 3-2-4) According to actual needs, appropriate weights are assigned to different costs and then added to obtain the final local obstacle avoidance reference line evaluation function, and the local obstacle avoidance reference line with the minimum cost is selected as the final driving guide line.

[0077] 4) Look up the table and perform coordinate transformation according to the current position of the tractor to obtain the estimated running trajectory of the local plan in the world coordinate system. Combined with the obstacle information collected by the perception module, the local planning line that passes the collision detection constraint is selected through the obstacle detection constraint table;

[0078] 4-1) Retrieve the optional partial executable trajectory from the database;

[0079] 4-2) According to the current position of the vehicle and the obstacle information, the coordinate position of the obstacle point cloud and shape information is discretized and converted to the vehicle coordinate system according to actual needs to obtain the grid coordinates in the vehicle coordinate system;

[0080] 4-3) The corresponding obstacle coordinates are projected to the offline calculation module, and the alternative local planning lines passing through the grid occupied by the obstacle are excluded from the obtained collision detection table, and finally a feasible local planning line without collision risk is obtained.

[0081] 5) Evaluate and score the offline generated local planning line according to the lateral and longitudinal costs of the guide point and the current posture of the tractor, and obtain the local planning line to be executed;

[0082] 5-1) Calculate the absolute value of the deviation between the end position angle of all candidate local planning lines that pass the collision detection and the current vehicle head direction as the lateral driving cost 1;

[0083] 5-2) Calculate the absolute value of the angle deviation between the end position of all candidate local planning lines that pass the collision detection and the optimal local planning line selected in the previous cycle as the lateral cost 2;

[0084] 5-3) Calculate the next cycle state value corresponding to all candidate local planning lines that pass the collision detection, and use the linear acceleration and the absolute value of the jerk as the longitudinal driving cost;

[0085] 5-4) The horizontal cost 1, horizontal cost 2 and vertical cost are added together as the cost evaluation function of the overall local planning line. All candidate lines are scored according to the evaluation function. The higher the cost value, the lower the score. The local planning line with the highest score is selected.

[0086] 6) Finally, according to the local planning line, the angular velocity of the vehicle body center in the next planning period is obtained and sent to the tractor body control module for execution.

[0087] 6-1) According to the guide point, the current posture of the tractor and the local curve of the previous cycle, the local curve with the smallest angle cost is selected as the local planning line to be executed;

[0088] 6-2) Finally, the angular velocity of the vehicle body center in the next planning period is obtained based on the local planning line and sent to the tractor body control module for execution.

[0089] Embodiment 1:

[0090] This embodiment is mainly divided into the following steps:

[0091] (1) According to parameters such as vehicle type, profile shape, and number of trailers, the local planning curve and collision detection constraint table are calculated offline, and the local planning curve and collision detection constraint table are added to the online module as a configuration file;

[0092] (2) After the obstacle avoidance requirement is triggered, based on the driving reference line and road width limit, horizontal and vertical sampling is performed along the lane line, and then a polynomial planning is generated from the current position of the tractor to the sampling point, and finally a backup obstacle avoidance guide line is generated along the driving reference line within the set lane width;

[0093] (3) Obstacle interference detection is performed on the generated road guide lines to exclude lanes occupied by obstacles; lanes that pass the obstacle detection are selected according to the distance from the reference line and the driving cost, and the optimal reference line is selected as the lane guide line, and a guide point is selected on the selected lane guide line;

[0094] (4) Look up the table and perform coordinate transformation according to the current position of the tractor to obtain the estimated running trajectory of the local plan in the world coordinate system. Combined with the obstacle information collected by the perception module, the local planning line that passes the collision detection constraint is selected through the obstacle detection constraint table;

[0095] (5) Evaluate and score the offline generated local planning line according to the lateral and longitudinal costs of the guide point and the current posture of the tractor, and obtain the local planning line to be executed;

[0096] (6) Finally, the angular velocity of the vehicle center in the next planning period is obtained based on the local planning line and sent to the tractor body control module for execution.

[0097] In summary, the present invention uses lane-based obstacle avoidance guide lines and offline local planning to achieve autonomous obstacle avoidance, avoiding the complexity of solving with a nonlinear optimization model, without the risk of falling into local optimality, and is safer and more reliable; generating local planning lines by offline calculation and table lookup greatly reduces the algorithm's demand for computing power and reduces the cost of implementation; offline calculation of the operating area occupied by the local planning line is used for collision detection. Since collision detection requires real-time calculation, this operation can effectively reduce the time required for collision detection and meet the real-time requirements of the algorithm;

[0098] The present invention adopts the operation of expanding the collision detection operation area according to the number and size of trailers, which is simple and easy to operate and can meet the needs of trailer obstacle avoidance; the present invention adopts the offline calculation method to perform spatial sampling of different vehicle head directions and target postures to generate a second-order smooth curve, and can generate local planning lines that meet different needs by parameter configuration according to the movement capabilities of different vehicle models (whether they can spin and reverse, etc.), and the deployment is flexible;

[0099] The present invention uses a lane guide line generated along a driving path based on a frenet coordinate system to perform lane-based obstacle interference detection, which is very consistent with the behavioral habit of outdoor vehicles driving along lanes.

[0100] Those skilled in the art will appreciate that the above are only preferred embodiments of the present invention, and the various embodiments of the present disclosure and / or the features described in the claims may be combined or combined in various ways, even if such combinations or combinations are not explicitly described in the present disclosure. It is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art may still modify the technical solutions described in the aforementioned embodiments, or perform equivalent substitutions on some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

[0101] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A highly real-time outdoor unmanned tractor autonomous obstacle avoidance method, characterized in that: The following steps are involved: 1) The offline calculation module obtains the local planning curve and the collision detection constraint table according to the vehicle parameters and vehicle operating environment information in the call database, and adds the local planning curve and the collision detection constraint table as a configuration file to the online working module; 2) After the obstacle avoidance requirement is triggered, based on the driving reference line and the road width limit, horizontal and vertical sampling is performed along the lane line; then a polynomial planning is generated from the current position of the tractor to the sampling point, and finally a backup obstacle avoidance guide line is generated along the driving reference line within the set lane width; 3) Obstacle interference detection is performed on the generated road guide line to exclude lanes occupied by obstacles; lanes that pass the obstacle detection are selected according to the distance from the reference line and the driving cost, the optimal reference line is selected as the lane guide line, and the guide point is selected on the selected lane guide line; 4) Look up the table and perform coordinate transformation according to the current position of the tractor to obtain the estimated running trajectory of the local plan in the world coordinate system. Combined with the obstacle information collected by the perception module, the local planning line that passes the collision detection constraint is selected through the obstacle detection constraint table; 5) According to the guidance point and the current posture of the tractor, the local planning line generated offline is evaluated and scored according to the lateral and longitudinal costs of driving to obtain the local planning line to be executed; 6) Finally, according to the local planning line, the angular velocity of the vehicle body center in the next planning period is obtained and sent to the tractor body control module for execution.

2. A high-real-time outdoor unmanned tractor autonomous obstacle avoidance method according to claim 1, characterized in that: The vehicle parameters include: vehicle body motion parameters, vehicle type, outline shape and number of trailers; The vehicle operating environment information includes: map data, road information, and obstacle locations.

3. A high-real-time outdoor unmanned tractor autonomous obstacle avoidance method according to claim 1, characterized in that: The offline calculation of the local planning curve and the collision detection constraint table is specifically as follows: 1-1) The offline calculation module plans the path of the vehicle according to the vehicle parameters and the vehicle operating environment information. First, the vehicle is locally planned and sampled at the end in the vehicle body coordinate system; The terminal sampling is divided into distance and angle. The length of the corresponding local planning line is selected according to the parameters such as the expected running speed of the vehicle. Then, the terminal head direction is sampled according to the limitations such as the vehicle's movement ability. In this way, a series of terminal state points can be obtained. The local plan is a series of position points that the vehicle follows in a short period of time; 1-2) Generate a local planning curve: Based on the sampling set at the end of the local planning path, use polynomial interpolation or optimization algorithm to generate a smooth curve so that the vehicle can travel along this curve; 1-3) Collision detection constraint table: Based on vehicle parameters and environmental information, interaction with map data, obstacle detection, and planning curve constraints, a collision detection constraint table is calculated, i.e., obstacles or collision risk areas that the vehicle should avoid during driving; 1-4) Performance optimization: Optimize the generated local planning curves and collision detection constraint tables to ensure that the vehicle has good performance and safety in actual operation.

4. A high-real-time outdoor unmanned tractor autonomous obstacle avoidance method according to claim 1, characterized in that: The polynomial planning is generated from the current position of the tractor to the sampling point, and finally a spare obstacle avoidance guide line is generated along the driving reference line within the set lane width, specifically: 2-1) Obtain the road reference line to be driven in the Cartesian coordinate system, and then obtain the frenet coordinate system of the driving reference line according to the conversion formula from the frenet coordinate system to the Cartesian coordinate system. In the frenet coordinate system, the current position of the tractor is taken as the state starting point, the sampling point is taken as the state end point, and the speed and acceleration of the end point are both set to 0; 2-2) Generate a 5th-order polynomial trajectory based on the starting and ending point states. After obtaining the lateral and longitudinal trajectories, perform the conversion from the Frenet coordinate system to the Cartesian coordinate system, and finally generate a backup obstacle avoidance guide line along the driving reference line within the set lane width.

5. A high-real-time outdoor unmanned tractor autonomous obstacle avoidance method according to claim 4, characterized in that: The Frenet coordinate system is used to describe the position of the vehicle along the path, and the Cartesian coordinate system describes the absolute position of the vehicle on the plane.

6. A high-real-time outdoor unmanned tractor autonomous obstacle avoidance method according to claim 1, characterized in that: The step 3) is specifically: 3-1) Perform obstacle interference detection on the generated road guide lines to exclude lanes occupied by obstacles. If all the generated local guide lines pass the interference detection, it is considered that the obstacle avoidance is completed and a completion signal is returned to the vehicle control program. Otherwise, continue to execute; 3-2) The lane guide line that passes the obstacle detection is selected according to the distance from the reference line, the distance obstacle cost and the driving cost, and the optimal reference line is selected as the lane guide line, and the guide point is selected on the selected lane guide line.

7. A high-real-time outdoor unmanned tractor autonomous obstacle avoidance method according to claim 1, characterized in that: The step 3-2) comprises the following steps: 3-2-1) Calculate the lateral offset cost: The absolute value of the lateral offset of the end sampling position corresponding to the local obstacle avoidance reference line from the reference line is taken as the lateral cost. The greater the lateral distance from the reference line, the greater the lateral cost. 3-2-2) Calculate the collision risk cost: The minimum distance derivative of the local obstacle avoidance line from the obstacle is taken as the collision cost. The closer the distance to the obstacle, the greater the collision risk. 3-2-3) Calculate the driving cost: The driving cost is determined based on whether the current position of the vehicle is consistent with the direction of the driving reference line. If consistent, the cost is 0, otherwise it is 1; 3-2-4) According to actual needs, appropriate weights are assigned to different costs and then added to obtain the final local obstacle avoidance reference line evaluation function, and the local obstacle avoidance reference line with the minimum cost is selected as the final driving guide line.

8. The high-real-time outdoor unmanned tractor autonomous obstacle avoidance method according to claim 1, characterized in that: The step 4) is specifically: 4-1) Retrieve the optional partial executable trajectory from the database; 4-2) According to the current position of the vehicle and the obstacle information, the coordinate position of the obstacle point cloud and shape information is discretized and converted to the vehicle coordinate system according to actual needs to obtain the grid coordinates in the vehicle coordinate system; 4-3) The corresponding obstacle coordinates are projected to the offline calculation module, and the alternative local planning lines passing through the grid occupied by the obstacle are excluded from the obtained collision detection table, and finally a feasible local planning line without collision risk is obtained.

9. A high-real-time outdoor unmanned tractor autonomous obstacle avoidance method according to claim 1, characterized in that: The offline generated local planning line is evaluated and scored according to the lateral and longitudinal costs of the guide point and the current posture of the tractor, and the local planning line to be executed is obtained, specifically: 5-1) Calculate the absolute value of the deviation between the end position angle of all candidate local planning lines that pass the collision detection and the current vehicle head direction as the lateral driving cost 1; 5-2) Calculate the absolute value of the angle deviation between the end position of all candidate local planning lines that pass the collision detection and the optimal local planning line selected in the previous cycle as the lateral cost 2; 5-3) Calculate the next cycle state value corresponding to all candidate local planning lines that pass the collision detection, and use the linear acceleration and the absolute value of the jerk as the longitudinal driving cost; 5-4) The horizontal cost 1, horizontal cost 2 and vertical cost are added together as the cost evaluation function of the overall local planning line. All candidate lines are scored according to the evaluation function. The higher the cost value, the lower the score. The local planning line with the highest score is selected.

10. The high-real-time outdoor unmanned tractor autonomous obstacle avoidance method according to claim 1, characterized in that: The step 6) is specifically: 6-1) According to the guide point, the current posture of the tractor and the local curve of the previous cycle, the local curve with the smallest angle cost is selected as the local planning line to be executed; 6-2) Finally, the angular velocity of the vehicle body center in the next planning period is obtained based on the local planning line and sent to the tractor body control module for execution.

Images