Method and apparatus for controlling non-road vehicle

Abstract

The invention provides a method for controlling a non-road vehicle. The method comprises the following steps: acquiring current operation data of the non-road vehicle; determining an identified job content of the non-road vehicle based on current operating data of the non-road vehicle, the identified job content representing a job content being performed by the non-road vehicle determined by means of the trained machine learning model; determining a current work content of the non-road vehicle based at least on the identified work content; and controlling the operation of the non-road vehicle according to the determined current work content. The invention further provides a device for controlling the non-road vehicle and a machine readable storage medium.

Description

Technical Field

[0001] The invention relates to a method for controlling an off-road vehicle. The invention also relates to a device for controlling an off-road vehicle and a machine-readable storage medium. Background Art

[0002] With the rapid development of intelligent manufacturing technology, more intelligent operating machines are urgently needed in various industries to improve production efficiency and achieve effective use of resources.

[0003] At present, off-road vehicles mainly rely on operators to manually switch operating modes, which requires operators to have proficient operating techniques and work experience, and to be highly focused during work. Otherwise, off-road vehicles cannot adjust their working status in time according to the changing working scenes, making it impossible to exert optimal performance during operation or difficult to meet actual production needs.

[0004] In the existing solution, it is proposed to collect the operation data of off-road vehicles through remote servers, use data proximity analysis to identify the working conditions, and regularly return calibration data to the off-road vehicles. However, this solution needs to process a large amount of data, has low recognition efficiency, and has special requirements for network coverage and data transmission bandwidth, so its applicability is limited.

[0005] In this context, it is expected to provide an improved off-road vehicle control solution to more accurately and efficiently identify the operation content of the off-road vehicle, thereby improving the operation quality and work efficiency. Summary of the invention

[0006] An object of the present invention is to provide a method for controlling an off-road vehicle, an apparatus for controlling an off-road vehicle and a machine-readable storage medium, so as to at least solve part of the problems in the prior art.

[0007] According to a first aspect of the present invention, there is provided a method for controlling an off-road vehicle, the method comprising the following steps:

[0008] Obtain current operating data of off-road vehicles;

[0009] Determine the identified operation content of the off-road vehicle based on the current operation data of the off-road vehicle, wherein the identified operation content represents the operation content being performed by the off-road vehicle determined by means of the trained machine learning model;

[0010] determining a current operation content of the off-road vehicle based at least on the identified operation content; and

[0011] The operation of the off-road vehicle is controlled according to the determined current operation content.

[0012] The present invention particularly includes the following technical concepts: by introducing a machine learning model, a more reliable determination of the current operation content of the off-road vehicle can be achieved, the data processing process is simplified, and the efficiency of operation recognition is improved. In addition, since the operation scene is identified online using artificial intelligence technology, the change of working conditions can always be recognized at the first time, so the off-road vehicle can be better controlled to perform operations, which largely avoids the situation where the working parameters do not match the actual operating conditions due to the operator's lack of experience or forgetting to select the operation mode.

[0013] Optionally, the identification operation content of the off-road vehicle and the determination of the current operation content are performed locally on the off-road vehicle.

[0014] As a result, off-road vehicles do not need to be equipped with communication units, reducing manufacturing costs. In addition, the identification process of the operation content is not restricted by geography (such as network coverage conditions), overcoming the problem of long-term mismatch between working parameters and actual operation scenarios due to communication delays.

[0015] Optionally, determining the identification operation content of the off-road vehicle based on the current operating data of the off-road vehicle includes: analyzing the current operating data in terms of the distribution matching degree of the sample operating data corresponding to the preset operation content with the help of a trained machine learning model, and determining the preset operation content corresponding to the sample operating data with the highest distribution matching degree with the current operating data as the identification operation content.

[0016] Therefore, we can make full use of the self-learning ability of the machine learning model to reduce the labor and time costs of summarizing data distribution patterns, and effectively improve the efficiency of job content recognition.

[0017] Optionally, the method further includes the following steps: in addition to determining the identification work content of the off-road vehicle, additionally allocating confidence to the identification work content using a trained machine learning model; and, additionally determining the current work content of the off-road vehicle based on the confidence assigned to the identification work content.

[0018] Thus, the reliability level of the job content recognition results can also be understood, thereby providing a valuable reference item for the final determination of the current job content.

[0019] Optionally, the machine learning model is trained by using sample operation data annotated with preset job content.

[0020] Therefore, the large amount of accumulated prior knowledge can be utilized to make the machine learning model more versatile and make the operation scenario determination process easier.

[0021] Optionally, the sample operating data is collected when the off-road vehicle or other off-road vehicles perform different preset work contents within a historical time period; and / or, the sample operating data is hypothetically generated by a mathematical or physical simulation model of the off-road vehicle.

[0022] Therefore, by collecting and continuously updating training samples through big data, the output results of the machine learning model are more in line with the actual situation. In addition, by generating sample data with the help of simulation models, the sample richness can be easily expanded, the generalization ability of the model is improved, and the training effect is optimized.

[0023] Optionally, controlling the operation of the off-road vehicle according to the determined current work content includes: obtaining target operating parameters adapted to the current work content, wherein the target operating parameters enable the off-road vehicle to have optimal performance in terms of energy consumption, driving dynamics, safety and / or post-processing efficiency when executing the current work content; and providing the target operating parameters for the operation of the off-road vehicle.

[0024] As a result, the working state of the off-road vehicle can be adjusted in time when the working conditions of the off-road vehicle change, thereby improving working efficiency and saving energy consumption.

[0025] Optionally, the off-road vehicle includes an agricultural machinery vehicle and / or an engineering machinery vehicle, wherein the agricultural machinery vehicle includes a tractor and a harvester, and the engineering machinery vehicle includes an excavator, a bulldozer and a loader; and / or

[0026] The operation contents of off-road vehicles are divided into light-load operation, medium-load operation and heavy-load operation according to the load level; in the case of off-road vehicles involving tractors, the operation contents are divided into deep plowing, rotary tillage, sowing and field transportation according to the type of work or production scene; and / or

[0027] The current operating data includes the throttle opening, engine speed, engine torque, hydraulic signal of the hydraulic suspension mechanism, posture signal of the agricultural machinery towed or pushed by the off-road vehicle and / or the type of the agricultural machinery towed or pushed by the off-road vehicle.

[0028] Therefore, the required model input data is easy to obtain, making the entire scheme more applicable.

[0029] Optionally, the method further comprises the following steps: obtaining a selected work content of the off-road vehicle, wherein the selected work content represents the work content being performed by the off-road vehicle determined by an operator of the off-road vehicle; and additionally determining a current work content of the off-road vehicle based on the selected work content.

[0030] As a result, the operator's operating experience is advantageously taken into account, making the decision results of the current work content more accurate.

[0031] Optionally, additionally determining the current working content of the off-road vehicle based on the selected working content includes: in the absence of the selected working content, allowing the identified working content to be determined as the current working content of the off-road vehicle; and / or in the presence of the selected working content:

[0032] - determining the selected work content as the current work content of the off-road vehicle; or

[0033] - When there is a conflict between the selected work content and the recognized work content, the selected work content is determined as the current work content, and a prompt is output to the operator.

[0034] In this way, the operator's subjective intentions can be fully followed, reducing the safety hazards caused by inaccurate recognition of machine learning models under complex working conditions.

[0035] Optionally, the method further includes the following steps: obtaining previous work content of the off-road vehicle, wherein the previous work content represents the work content being performed that has been determined to be performed for the off-road vehicle most recently so far; and additionally determining the current work content of the off-road vehicle based on the previous work content.

[0036] As a result, historical work records are advantageously taken into account when determining the current work content of the off-road vehicle, thereby facilitating more consistent and smooth switching of work scenes.

[0037] Optionally, additionally controlling the operation of the non-road vehicle based on previous work content includes: determining the identified work content as the current work content only when it is confirmed that the identified work content meets the predefined output condition, otherwise continuing to determine the previous work content as the current work content.

[0038] As a result, the working state is adaptively adjusted only when necessary (i.e., when the working scene switches), avoiding frequent changes in working parameters that may cause pauses or interference in the normal operation of the off-road vehicle.

[0039] Optionally, it is confirmed that the recognized job content meets the predefined output condition in the following cases: the recognized job content and the previous job content point to job content of different categories; and / or the confidence assigned to the recognized job content is higher than a preset threshold.

[0040] Therefore, by setting the above filtering conditions, the accuracy of determining the current working content of the off-road vehicle can be improved, and too frequent adjustment of working parameters can be avoided.

[0041] According to a second aspect of the present invention, there is provided a device for controlling an off-road vehicle, the device being used to execute the method according to the first aspect of the present invention, the device comprising:

[0042] An acquisition module configured to acquire current operation data of the off-road vehicle;

[0043] An identification module configured to determine an identification operation content of the off-road vehicle based on current operation data of the off-road vehicle, wherein the identification operation content represents an operation content being performed by the off-road vehicle determined by means of a trained machine learning model;

[0044] a decision module configured to determine a current working content of the off-road vehicle based at least on the identified working content; and

[0045] A control module is configured to control the operation of the off-road vehicle according to the determined current work content.

[0046] According to a third aspect of the present invention, there is provided a machine-readable storage medium, on which a computer program is stored, wherein the computer program is used to execute the method according to the first aspect of the present invention when executed on a computer. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present invention will be described in more detail below with reference to the accompanying drawings, so that the principles, features and advantages of the present invention can be better understood. The accompanying drawings include:

[0048] Figure 1 A block diagram showing an apparatus for controlling an off-road vehicle according to an exemplary embodiment of the present invention;

[0049] Figure 2 Shows Figure 1 A schematic diagram of the training process of the machine learning model in the recognition module in FIG.

[0050] Figure 3 A flow chart showing a method for controlling an off-road vehicle according to an exemplary embodiment of the present invention;

[0051] Figure 4 Shows Figure 3 A flow chart according to an exemplary embodiment of two method steps of the method shown;

[0052] Figure 5 Shows Figure 3 A flow chart according to a further exemplary embodiment shows two method steps of a method. DETAILED DESCRIPTION

[0053] In order to make the technical problems, technical solutions and beneficial technical effects to be solved by the present invention more clearly understood, the present invention will be further described in detail below in conjunction with the accompanying drawings and multiple exemplary embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the scope of protection of the present invention.

[0054] In this article, non-road vehicles may include various non-road vehicles. Such non-road vehicles may generally be equipped with a power source (e.g., fuel, electricity, wind power, water power, etc.), be movable (e.g., traction type, self-propelled power type, etc.), and have or not have a body, but are not intended to travel on roads (e.g., highways). For example, in some application scenarios, non-road vehicles may include non-road mobile mechanical vehicles. For example, non-road vehicles may include various agricultural mechanical vehicles, such as tractors, harvesters, etc. Non-road vehicles may also include various engineering mechanical vehicles, such as excavators, bulldozers, loaders, etc.

[0055] Figure 1 A block diagram of an apparatus for controlling an off-road vehicle according to an exemplary embodiment of the present invention is shown.

[0056] like Figure 1 As shown, the device 1 for controlling an off-road vehicle may include, for example, an acquisition module 10 , an identification module 20 , a decision module 30 and a control module 40 , which are connected to one another via communication technology.

[0057] In one embodiment, in order to realize real-time online recognition of the working content of the off-road vehicle, the device 1 can be arranged locally in the off-road vehicle as a whole, or at least some modules thereof (such as the recognition module 20 and the decision module 30). For example, the device 1 can be constructed as a part of the engine control unit (ECU) of the off-road vehicle or integrated therewith.

[0058] The acquisition module 10 is used to acquire the current operation data of the off-road vehicle. To this end, the acquisition module 10 can be connected to various state sensors of the off-road vehicle, such as a throttle position sensor, a fuel level sensor, a speed sensor, a torque sensor, a rotation speed sensor, a hydraulic sensor, a light sensor, etc. In some embodiments, the acquisition module 10 can also be connected to a position sensor and an environmental sensor of the off-road vehicle to acquire the geographical location data of the off-road vehicle and understand environmental data such as air pressure, temperature, humidity, altitude, etc. In addition, the acquisition module 10 can also collect current operation data, geographical location data, and environmental data through the bus (such as a CAN bus) of the off-road vehicle. Of course, in other cases, the acquisition module 10 can also collect other data related to the off-road vehicle, which is not limited herein.

[0059] The recognition module 20 receives the current operation data from the acquisition module 10, and determines the recognition operation content Ci of the non-road vehicle based on the current operation data. The recognition module 20 is provided with a machine learning model, for example, which has been trained by sample operation data marked with preset operation content. The specific training method and tuning process of the model will be described in detail below. Figure 2 After the training is completed, by inputting the current operation data acquired in real time into the machine learning model, the operation content being performed by the off-road vehicle can be obtained as an output result. The machine learning model is, for example, an artificial neural network, a support vector machine, a decision tree, etc., which is not specifically limited in the present invention.

[0060] The decision module 30 receives the identified operation content Ci determined by the recognition module 20, and determines the current operation content of the off-road vehicle at least based on the identified operation content Ci. In one embodiment, the decision module 30 can directly determine the identified content Ci as the current operation content Cn of the off-road vehicle. In another embodiment, other factors can also be additionally considered, and the decision module 30 can select the operation content determined by which method to obtain the final decision result through logical arbitration methods such as priority allocation and weight analysis. For example, in Figure 1 As shown in , the decision module 30 can also receive as input the selected work content Cj from the operator of the off-road vehicle and the previous work content Cn-1 that has been determined for the off-road vehicle, and simultaneously determine the current work content Cn of the off-road vehicle based on the identified work content Ci, the selected work content Cj and the previous work content Cn-1.

[0061] After the current operation content Cn is determined for the off-road vehicle, this can be provided to the control module 40 so that the operation of the off-road vehicle can be controlled there according to the determined current operation content Cn. In the control module 40, for example, a variety of target operating parameters can be pre-stored, which can be in the form of a data set or data characteristic curve (MAP diagram) consisting of engine torque, engine speed, gear, engine power, accelerator pedal opening, and brake pedal opening. For example, a variety of target operating parameters can be determined by big data analysis for off-road vehicles. Different target operating parameters can correspond to different usage scenarios of off-road vehicles, such as at least corresponding to different operation contents. In this way, based on the current operation content, an adapted target operating parameter can be selected from a variety of target operating parameters and applied to the off-road vehicle.

[0062] It should be noted that Figure 1The functional components shown in the figure are only block diagrams of some structures related to the present invention, and do not constitute a limitation on all functional components used in the scheme of the present invention. Figure 1 The embodiments shown may exist in more, less or in combination, and the present invention is not limited thereto.

[0063] Figure 2 Shows Figure 1 Schematic diagram of the training process of the machine learning model in the recognition module in .

[0064] like Figure 2 As shown, in order to train the machine learning model 202, Figure 1 The recognition module 20 in the embodiment includes, in addition to the machine learning model 202, a sample collection unit 201, a comparison unit 203, and an update unit 204. In this embodiment, the recognition module 20 includes a neural network as the machine learning model 202, but is not limited thereto, and the machine learning model 202 may also be a support vector machine, a decision tree, etc.

[0065] With the help of the sample collection unit 201, sample operation data of the off-road vehicle can be collected and annotated to add classification labels to the collected sample operation data. Taking a tractor as an example, the sample operation data may include a data set consisting of multiple sets of engine torque values ​​and engine speed values ​​collected when the off-road vehicle performs corresponding operations. Figure 2 Specifically, Figure 2 As shown, assuming that the horizontal axis represents the normalized speed (in Figure 2 The vertical axis represents the normalized torque (in Figure 2 It can be seen that different data points of the sample running data may fall into different regions.

[0066] Then, the collected sample operation data can be divided into multiple categories according to the distribution of data points in different areas. Each category corresponds to a typical operation content summarized based on experience or experimental rules, which is also defined as "preset operation content" in the context. For example, for the case where the data points are mainly distributed in the upper right interval, it can be indicated that the preset operation content of the tractor corresponding to this sample operation data is arable land. For another example, for the case where the data points are mainly distributed in the lower right interval, it can be indicated that the preset operation content corresponding to the sample operation data is sowing. For the case where the data points are mainly distributed in several intervals in the lower left, it can be indicated that the preset operation content of the tractor is field transportation operation.

[0067] It should be noted that the types of sample run data are not limited to Figure 2The engine torque value and engine speed value shown in the figure may also include other forms of data such as throttle opening. It is necessary to ensure that the data type and format of the current operating data collected in actual applications and provided as input to the machine learning model should be consistent with the data type and format of the sample operating data used to train the machine learning model.

[0068] In one embodiment, the sample operation data is collected by the off-road vehicle or other off-road vehicles when performing different preset operation contents in a historical time period, and is centrally stored in a storage unit located locally on the off-road vehicle or in a cloud server (not specifically shown). In another embodiment, the sample operation data is hypothetically generated by a mathematical or physical simulation model of the off-road vehicle.

[0069] This means that the sample operation data can be either real operation data detected by physical sensors installed on off-road vehicles, or hypothetical operation data generated by simulating the posture and operation status of off-road vehicles under different working conditions based on pre-built mathematical or physical simulation models. Hypothetically generating sample operation data has the following advantages: it can simulate the behavior of off-road vehicles under special weather conditions or extreme geographical conditions, and it can also artificially add outliers and abnormal points to the sample operation data to affect the identifiability of the preset operation content, thereby improving the generalization ability of the model, optimizing the training effect, and improving the training efficiency.

[0070] The sample operation data collected by the sample collection unit 201 can be provided to the machine learning model 202 to establish a mapping relationship between input and output, and continuously learn the correspondence between the sample operation data and the corresponding preset operation content through the self-learning ability of the model. After each training, the model 202 will obtain an output result C1' for the input sample operation data, and the output result is, for example, the preset operation content predicted by the model 202. Alternatively, the output result may also include a corresponding confidence level (which may be in the form of a percentage, for example).

[0071] The comparison unit 203 simultaneously receives the (annotated) classification label C1 of the sample operation data and the output result C1' of the machine learning model 202, and compares the two. Specifically, the error ΔC between the true value C1 and the predicted value C1' of the preset operation content can be used as the comparison result. For example, the error ΔC between the true value C1 and the predicted value C1' can be calculated by an appropriate loss function.

[0072] The comparison result is back-propagated to the update unit 204, where the error ΔC between the output result C1' of the model 202 and the actual label category C1 can be used as an indicator to judge the quality of the training of the model 202, and the updated internal parameters of the machine learning model 202 can be calculated accordingly. For example, the gradient descent algorithm / error back propagation algorithm can be used to calculate the updated weights of the neural network so that the error ΔC is continuously reduced.

[0073] The updated model internal parameters are assigned to the machine learning model 202. In this way, through repeated iterative processing, the internal parameters of the machine learning model 202 are continuously adjusted and optimized. Finally, until the error ΔC calculated by the comparison unit 203 is less than the predefined limit value or does not decrease significantly in multiple consecutive trainings, or until the model 202 reaches a preset number of training steps or iterations, the training of the machine learning model 202 is considered to be completed.

[0074] Figure 3 A flow chart of a method for controlling an off-road vehicle according to an exemplary embodiment of the present invention is shown. The method exemplarily comprises steps S1-S4 and can be used, for example, in Figure 1 The device 1 shown is implemented in the case.

[0075] In step S1 , current operating data of the off-road vehicle is acquired.

[0076] As already introduced above, for example, the current operation data can be collected with the help of various state sensors of the off-road vehicle, which include, for example, the throttle opening, engine speed and / or engine torque of the off-road vehicle. In addition, when the off-road vehicle involves a working mechanical vehicle such as a tractor, an excavator or a harvester that can be connected to agricultural machinery through a hydraulic suspension mechanism, the current operation data may also include the hydraulic signal of the hydraulic suspension mechanism of the off-road vehicle, the posture signal of the agricultural machinery towed or pushed by the off-road vehicle, and the category of the agricultural machinery towed or pushed. For example, the size and change trend of the hydraulic pressure fed back by the hydraulic sensor can reflect the height of the agricultural machinery being lifted or lowered, the depth of the agricultural machinery embedded in the soil, and the weight difference of the materials grabbed or harvested by the agricultural machinery, so that the operation content can be more accurately inferred. For another example, the height of the plow body connected to the off-road vehicle can also be measured by the signal of the optical sensor set on the off-road vehicle, so as to assist in judging whether the off-road vehicle is in a farming operation scene. In addition, depending on the working nature of the off-road vehicle, the current operation data may also include the action behavior data or posture data of the off-road vehicle during the operation, which may include, for example, the swing angle of the robot arm, the height from the ground, the position relative to the vehicle body or the base, the working state of the bucket, etc. In addition, when necessary, the environmental data and geographic location data of the off-road vehicle may also be obtained and provided to the machine learning model to improve the accuracy of the recognition of the working content.

[0077] In step S2, the identified operation content of the off-road vehicle is determined based on the current operation data of the off-road vehicle, wherein the identified operation content represents the operation content being performed by the off-road vehicle determined by means of the trained machine learning model.

[0078] As described above, the machine learning model can be in various forms. However, no matter which specific form is involved, the effect expected to be achieved through the self-learning ability of the machine learning model and the necessary training process is to make the internal decision logic of the machine learning model follow the following mechanism: determine which of the preset job contents the current operation data corresponds to by the distribution match between the current operation data and the sample operation data. Specifically, the machine learning model is trained, for example, to analyze the current operation data in terms of the distribution match of the sample operation data corresponding to the preset job content, and determine the preset job content corresponding to the sample operation data with the highest distribution match with the current operation data as the recognition job content.

[0079] In this step, in addition to outputting the off-road vehicle identification operation content through the trained machine learning model, the corresponding confidence level can also be output with the help of the machine learning model. For example, if the machine learning model determines that the data distribution of the current operating data meets the "deep plowing" scenario, it can directly output "deep plowing" as the off-road vehicle identification operation content. At the same time, for example, it is also known from the machine learning model that the confidence level of this recognition result is "90%", which means that the probability or credibility that the off-road vehicle is currently performing the "deep plowing" operation is 90%.

[0080] In addition, it is also possible that the machine learning model simultaneously outputs multiple categories of recognition task content and the corresponding confidence level for each category.

[0081] In this step, the current operation data collected in real time can be continuously input into the machine learning model. However, at this time, the machine learning model does not necessarily have a corresponding instantaneous output result for each input at each moment, but may need to identify the operation content currently being performed by the off-road vehicle after a preset time. For example, the preset time can be 5 minutes, 10 minutes, 30 minutes, etc., which may be related to the sampling density and frequency of the current operation data and the training method of the machine learning model itself.

[0082] In another embodiment, the current operation data may also be input into the machine learning model at predetermined time intervals (such as periodically). Thus, the single data points accumulated over a period of time may be formed into a data set, and the current operation data in the form of a data set may be provided to the machine learning model as input. In some implementations, the current operation data may also be input into the machine learning model non-periodically. For example, the current operation data may be input every 3 minutes over a period of time, and every 6 minutes over another period of time, and so on. In either case, for the machine learning model, when outputting the content of the recognition job, it may not only take into account the only input once, but also additionally take into account multiple inputs that are performed successively in time (such as the logical correlation and time correlation between multiple inputs).

[0083] It is understandable that before the collected current operation data is input into the machine learning model, the current operation data may be preprocessed in a step not shown, which may include, for example, removing outliers, denoising, smoothing, averaging, data aggregation operations, etc., so as to make the recognition effect of the model more accurate. In addition, during the actual use of the trained machine learning model, the machine learning model may be retrained using the current operation data collected in real time and the selected operation content given by the operator to further optimize the model performance, or to make it closer to the characteristics of the actual working area of ​​the off-road vehicle or the operator's operating preferences.

[0084] In step S3, the current working content of the off-road vehicle is determined based at least on the identified working content.

[0085] In the embodiments of this invention, the operation content being performed by the off-road vehicle may refer to the operation performed by the off-road vehicle for purposes other than driving on the road, such as the operation performed by the off-road vehicle for production or engineering purposes. For example, the off-road vehicle may have various functions such as production operations or engineering operations, such as agricultural production, engineering construction, etc. For a tractor, the operation content may be divided into deep plowing, rotary tillage, sowing, field transportation, etc. according to the type of work or production scene. For another example, the operation content of the off-road vehicle may also be divided into light load operation, medium load operation, heavy load operation, etc. according to the load level.

[0086] In this step, as described above, Figure 1 As has been explained, not only can the identified operation content output by the machine learning model be directly determined as the current operation content of the off-road vehicle, but the subjective intention of the operator and the historical operating data of the off-road vehicle can also be additionally considered, which will be combined with Figure 4 and Figure 5 Further elaboration.

[0087] In addition, in addition to considering the identification operation content, the confidence assigned to the identification operation content can also be considered when determining the current operation content. For example, only when the confidence is higher than a certain value (70%), the current operation content is determined according to the identification operation content. Otherwise (when it is lower than 70%), it is considered that the machine learning model is not credible in determining the operation content currently being performed by the off-road vehicle. Therefore, when determining the current operation content, the output from the machine learning model can be abandoned, and thus the current operation content of the off-road vehicle can be changed, or the current operation content can be determined through other channels.

[0088] In step S4, the operation of the off-road vehicle is controlled according to the determined current work content.

[0089] In this step, for example, target operating parameters adapted to the current work content can be obtained, wherein the target operating parameters enable the off-road vehicle to have optimal performance in terms of energy consumption, driving dynamics, safety, and / or post-processing efficiency when performing the current work content. Then, the target operating parameters are provided for the operation of the off-road vehicle.

[0090] As mentioned above, "optimal performance" is reflected in the following aspects, for example: energy consumption, driving dynamics, safety, and post-processing efficiency. Accordingly, the target operating parameters may include, for example, maximum fuel limit, driving characteristic parameters (accelerator pedal sensitivity, brake pedal sensitivity, engine torque, engine speed, etc.), and enable and disable instructions for specific functions (such as regeneration operations). The target operating parameters may be calibrated in advance based on experience, big data, and experimental processes, and may be adjusted as needed in actual applications.

[0091] In a specific example, the maximum fuel limit of the non-road vehicle under different categories of current operation content can be set to be different. When the non-road vehicle performs heavy-load operations such as deep plowing, the maximum fuel limit can be set to be larger, and when performing light-load operations such as field transportation, the maximum fuel limit can be set to be smaller. For another example, the driving characteristic parameters of the non-road vehicle under different categories of current operation content can also be set to be different. When performing heavy-load operations such as deep plowing, the accelerator pedal can be adjusted to be more sensitive to achieve higher power performance, and when performing light-load operations such as field transportation, the accelerator pedal can be adjusted to be more gentle to achieve higher driving comfort. For another example, when the non-road vehicle performs different categories of current operation content, a determined function can be enabled or disabled. When performing heavy-load operations such as deep plowing, the regeneration function is allowed to be triggered to achieve the most adequate exhaust gas after-treatment, and when performing light-load operations such as field transportation, the regeneration function can be disabled to avoid the exhaust gas after-treatment regeneration efficiency being too low, the regeneration time being too long, resulting in excessive exhaust gas pollutants and increased fuel consumption.

[0092] It can be seen that in this way, the working parameters suitable for the current working content can be dynamically determined for the off-road vehicle, thereby improving the performance of the off-road vehicle in different working scenarios (for example, reducing energy consumption, providing high power, high speed, etc.), thereby flexibly meeting the current application needs.

[0093] In one embodiment, the target operating parameters may be stored in a designated storage area of ​​the off-road vehicle, and then when a change has been determined in the operating scenario (i.e., when the operating parameters of the off-road vehicle need to be updated), the corresponding target operating parameters are retrieved from the storage area according to the determined current operating content, so that the off-road vehicle operates according to the target operating parameters.

[0094] In one embodiment, the current operation content of the off-road vehicle and the determination of the target operating parameters are performed locally in the off-road vehicle (especially in the ECU). In this case, the target operating parameters do not need to be flashed to the ECU as a whole to replace all the current operating parameters of the off-road vehicle. Instead, the current operating parameters of the off-road vehicle can be compared with the target operating parameters, and only the part of the current operating parameters that deviates from the target operating parameters can be adjusted to make the current operating parameters consistent with the target operating parameters.

[0095] Figure 4 Shows Figure 3 A flow chart of two method steps of the method shown according to an exemplary embodiment. Figure 4 In the embodiment shown, Figure 3 Step S2 in the method includes sub-steps S21-S22, and step S3 includes sub-steps S31-S34.

[0096] In step S21, it is checked whether the preset conditions for identifying the operation content by means of the machine learning model are met. The purpose of this is to avoid collecting the current operation data of the off-road vehicle under extreme working conditions, because the current operation data collected in this way may be seriously deviated from the normal operation data of the off-road vehicle when performing the operation due to environmental factors (such as temperature, altitude, air pressure, slope, etc.), and thus cannot reflect the actual operation scene in which it is located. If the machine learning model is allowed to identify the operation content based on such "non-universal" current operation data, it is very likely to obtain an erroneous output result, or it may also cause the output result of the machine learning model to be invalid.

[0097] In a specific example, it can be checked in step S21 whether the ambient temperature of the geographical area where the off-road vehicle is located is lower than a preset temperature (e.g., -10 degrees), and it can also be checked whether the off-road vehicle has just started from a shutdown state. Since the engine is started in extremely cold weather, factors such as fuel viscosity, friction resistance, and water temperature will be inconsistent with the corresponding factors when most sample operating data (the sample operating data is used to train the machine learning model) are collected, resulting in the current operating data collected in this case being unrepresentative, which may cause the output result of the machine learning model to be "distorted". Therefore, if this is the case, it is determined that the preset conditions for identification with the help of the machine learning model are not met, and then the current operating data is not collected in this case, or the collected current operating data is not provided to the machine learning model. For example, if it is determined that the preset conditions are not met, this check can be continued in step S21, or it can be directly jumped to step S31 to determine the current operation content according to operator input or other methods.

[0098] If it is determined in step S21 that the preset conditions for identifying the work content with the help of the machine learning model are met, the current operating data of the acquired off-road vehicle can be input into the machine learning model in step S22, and the corresponding identification work content can be obtained with the help of the machine learning model.

[0099] In step S31, the selected operation content of the off-road vehicle is obtained, and the selected operation content represents the operation content of the off-road vehicle being performed, which is manually input by the operator of the off-road vehicle. For example, a gear shift switch in the form of a knob may be arranged on the off-road vehicle, and the operator can select the load gear of the off-road vehicle by manipulating the knob. For example, it is possible to switch between "heavy load operation", "medium load operation" and "light load operation". In addition, according to the specific configuration of the off-road vehicle, the input from the operator can also be received through voice recognition, screen touch, etc.

[0100] In one embodiment, the selected operation content of the off-road vehicle is simpler than the identified operation content, that is, the selected operation content may only have a rough load classification, but cannot reflect the specific type of work (such as deep plowing, rotary tillage, sowing, field transportation, etc.). In this case, there may not be a clear one-to-one correspondence between the selected operation content and the identified operation content, but there is a "conflict" or "non-conflict" relationship.

[0101] In another embodiment, the selected operation content and the identified operation content of the off-road vehicle respectively include the types of work defined in the same manner, for example, both may involve specific operations such as deep plowing, rotary tillage, sowing, and field transportation. In this case, there is a clear consistency relationship between the selected operation content and the identified operation content, that is, they may point to the same category of operation content or to different categories of operation content.

[0102] In step S32, it may be checked whether there is a selected work content. For example, it may be checked whether a specific input operation of the operator is detected in the previous step S31.

[0103] The absence of the selected work content means, for example: the off-road vehicle does not provide a switch for the operator to input the selected work content; or the switch for inputting the selected work content is placed in a default state (for example, the knob is adjusted to the extreme position and therefore does not work), which means that although the off-road vehicle provides the corresponding switch, the operator has not made any input.

[0104] The existence of selected working content means, for example: the operator proactively adjusts the knob position in a timely manner according to the real-time working scene of the off-road vehicle, so that the gear indicated by the knob matches the working content being performed by the off-road vehicle; there is also a situation where the selected working content input by the operator remains in the previous process stage or working scene, but the off-road vehicle has changed the working content at this time, so in this case it is possible that the selected working content does not match the working content actually being performed by the off-road vehicle.

[0105] If it is determined in step S32 that there is no selected operation content, the identified operation content previously determined in step S22 can be determined as the current operation content of the off-road vehicle in step S33. In this way, the operation scene can be quickly identified with the help of artificial intelligence technology in the absence of operator input, thereby improving the performance of the off-road vehicle in different operation scenes.

[0106] If it is determined in step S32 that there is a selected operation content, in one embodiment, the selected operation content can be directly determined as the current operation content of the off-road vehicle in step S34. The advantage of this is that under some complex working conditions, the environmental conditions are changeable and emergencies occur frequently. Due to the limited generalization ability of the machine learning model, its output results may not be able to accurately reflect the actual operation content, and the decision made by the operator after considering multiple factors is often more flexible. Therefore, the subjective intention of the operator can be fully followed to ensure the accuracy of the determined current operation content.

[0107] In another embodiment, if it is determined in step S32 that the selected operation content exists, it can be further distinguished in step S34 whether the operation content selected by the operator for the off-road vehicle is updated in time (matching the operation content actually being performed) or expired (the switch position has not changed for a long time, or it does not match the operation content actually being performed). In the case where the selected operation content has been updated in time, the selected operation content can be directly determined as the current operation content. In the case where the selected operation content has not been updated in time, the selected operation content can be temporarily still determined as the current operation content. At the same time, a prompt is issued to the operator in the off-road vehicle (for example, by optical, acoustic and / or tactile means) to inform the operator that the selected operation content currently recorded by the system may be invalid due to long-term non-update, or to inquire whether it is necessary to determine the current operation content according to the identified operation content.

[0108] In another embodiment, if it is determined in step S32 that there is a selected operation content, it is further possible to distinguish whether there is a "conflict" between the selected operation content and the identified operation content in step S34, and determine the current operation content according to the result of the distinction. As described above, the existence of a "conflict" may mean that the selected operation content and the identified operation content point to different categories of operation content (for example, one points to "deep plowing" and the other points to "seeding"), or it may mean that the selected operation content and the identified operation content do not match in terms of their categories (for example, one points to "heavy load operation" and the other points to "field transportation" which is a light load operation).

[0109] Specifically, when there is no conflict between the two, either the selected work content or the identified work content can be determined as the current work content, or, if the selected work content only refers to heavy-duty work, and the identified work content refers to a more refined work category "deep ploughing", the identified work content can be determined as the current work content. On the contrary, if there is a conflict, the current work content can be determined preferentially according to the selected work content, and a prompt can be output to the operator to notify him of the conflict and ask the operator to confirm whether there may be an error in the selected work content.

[0110] Figure 5Shows Figure 3 A flow chart of two method steps of the method shown according to another exemplary embodiment. Figure 5 In the embodiment shown, the implementation process of steps S21, S22, S31, S32, and S34 is the same as Figure 4 The corresponding steps are the same and have the same reference numerals, and will not be repeated here. Figure 5 and Figure 4 The difference.

[0111] like Figure 5 As shown, in addition to obtaining the selected operation content of the off-road vehicle in step S31, the previous operation content of the off-road vehicle can also be obtained in step S31'. Here. The previous operation content represents the operation content that has been determined for the off-road vehicle and is being performed the most recently so far. For example, the system records can be read from the local storage area of ​​the off-road vehicle, and the latest operation content determined for the off-road vehicle (for example, determined by means of a trained machine learning model and / or manually input by an operator) can be viewed. Specifically, for example, the off-road vehicle has just switched from the "deep plowing" operation mode to the "seeding" operation mode. At this time, the "current operation content" recorded in the system has not been updated but still points to "deep plowing". In this case, it can be determined that the "previous operation content" of the non-road vehicle is "deep plowing".

[0112] Next, if it is determined in step S32 that there is no selected work content from the operator, the current work content is not determined directly according to the recognized work content, but it is checked in step S331 whether the output condition of the recognized work content is met.

[0113] In one embodiment, it can be checked in step S331 whether the confidence assigned to the identification of the work content is higher than a preset threshold (e.g., 70%), and the preset threshold can be set based on experience or dynamically adjusted based on actual needs. If the confidence is higher than the preset threshold, it means that the output result of the machine learning model is more credible, and the possibility that the recognition result is consistent with the actual scene is higher, so it can be determined that the output condition for identifying the work content is met. If the confidence is lower than the preset threshold, it means that the output result of the machine learning model is relatively vague, or in other words, the current cumulative operating data provided to the machine learning model is relatively small. For the machine learning model, there are multiple possibilities for the working conditions of the off-road vehicle at this time, and it is not clear enough to identify what specific work content the off-road vehicle is performing. In this case, it is determined that the output condition for identifying the work content is not met.

[0114] In another embodiment, it is also possible to check in step S331 whether the identified operation content and the previous operation content point to different categories of operation content. If the two point to different categories, it means that the operation scene of the off-road vehicle has changed, and it is meaningful to redetermine the current operation content at this time, so it is determined that the output condition of the identified operation content is met. If the two point to the same category, it means that the operation scene of the off-road vehicle has not been switched, so up to now, the operation content most recently determined for the off-road vehicle is still valid. In this case, there is no need to repeatedly determine the current operation content for the off-road vehicle according to the identified operation content, so it can be determined that the output condition of the identified operation content is not met.

[0115] In another embodiment, it is also possible to check in step S331 whether the identified operation content and the previous operation content point to different categories of operation content, and whether the two categories of operation content are temporally associated and logically associated. If the identified operation content and the previous operation content point to different categories of operation content, and the two categories of operation content are temporally associated and logically associated, it is determined that the output condition of the identified operation content is met, otherwise it is determined that it is not met. In this way, it is also possible to additionally verify whether the output results of the machine learning model are temporally consistent with the historical operation records and whether such operation task switching is logically meaningful. For example, if it is found that the operation task of the tractor suddenly changes from deep plowing to sowing, it is considered to be in line with the normal logical order. However, if it is found that the task type of the tractor suddenly changes from sowing to deep plowing, it is necessary to consider whether there may be an identification error.

[0116] If it is determined in step S331 that the output condition of the identified operation content is met, the identified operation content can be determined as the current operation content of the off-road vehicle in step S332. Thus, when a potential operation scene switch is confirmed, the current operation content can be updated immediately to ensure that the working performance of the off-road vehicle dynamically adapts to the changing working conditions.

[0117] If it is determined in step S331 that the output condition of the recognized operation content is not met, then in step S333, the previous operation content is continuously determined as the current operation content, that is, the current operation content is not updated according to the recognized operation content. As a result, there is no need to repeatedly adjust and update the working parameters of the off-road vehicle in a short period of time, which greatly reduces the calculation burden of the system.

[0118] It should be noted that although the reference Figure 5 It is described that the previous operation content is further obtained under the premise of considering the selected operation content input by the operator, but it is also conceivable that only the previous operation content is additionally considered. In other words, step S31' can not only be as in Figure 5Instead of being performed after step S31 as shown in FIG, it may be performed before, alternately with, or in parallel with step S31 ′, or only step S31 ′ may be performed while steps S31 , S32 , and S34 are omitted.

[0119] It can be understood that the methods of various embodiments of the present disclosure can be implemented by computer programs / software. These software can be loaded into the working memory of the processor and used to execute the methods according to various embodiments of the present disclosure when running.

[0120] According to another embodiment of the present disclosure, a machine-readable storage medium is provided, such as a CD-ROM, which includes a computer program, which, when executed, causes a computer or a processor to perform a method according to each embodiment of the present disclosure. The machine-readable storage medium is, for example, an optical storage medium or a solid-state medium supplied together with other hardware or as part of other hardware.

[0121] Although specific embodiments of the present invention are described in detail herein, they are provided for the purpose of explanation only and should not be considered to limit the scope of the present invention. Various substitutions, changes and modifications may be conceived without departing from the spirit and scope of the present invention.

Claims

1. A method for controlling an off-road vehicle, the method comprising the following steps: Obtain current operating data of off-road vehicles; Determine the identified operation content of the off-road vehicle based on the current operation data of the off-road vehicle, wherein the identified operation content represents the operation content being performed by the off-road vehicle determined by means of the trained machine learning model; determining a current operation content of the off-road vehicle based at least on the identified operation content; as well as The operation of the off-road vehicle is controlled according to the determined current operation content.

2. The method according to claim 1, wherein: The identification operation content of the off-road vehicle and the determination of the current operation content are performed locally on the off-road vehicle.

3. The method according to claim 1 or 2, wherein: The method further comprises the following steps: In addition to determining the identification task content of the off-road vehicle, a confidence level is additionally assigned to the identification task content by means of a trained machine learning model, wherein the machine learning model is trained by sample operating data annotated with preset task content; and Additionally, a current task content of the off-road vehicle is determined based on the confidence level assigned to the identified task content.

4. The method according to claim 3, wherein: The sample operation data is collected when the off-road vehicle or other off-road vehicles perform different preset operation contents within a historical time period; and / or The sample operating data is hypothetically generated by a mathematical or physical simulation model regarding the off-road vehicle.

5. The method according to any one of claims 1 to 4, wherein: Controlling the operation of the off-road vehicle according to the current work content determined includes: Obtaining target operating parameters adapted to the current operation content, wherein the target operating parameters enable the off-road vehicle to have optimal performance in terms of energy consumption, driving dynamics, safety, and / or post-processing efficiency when performing the current operation content; and Target operating parameters are provided for operation of the off-highway vehicle.

6. The method according to any one of claims 1 to 5, wherein: Off-road vehicles include agricultural machinery vehicles and / or engineering machinery vehicles, wherein the agricultural machinery vehicles include tractors and harvesters, and the engineering machinery vehicles include excavators, bulldozers and loaders; and / or The operation contents of off-road vehicles are divided into light-load operation, medium-load operation and heavy-load operation according to the load level; in the case of off-road vehicles involving tractors, the operation contents are divided into deep plowing, rotary tillage, sowing and field transportation according to the type of work or production scene; and / or The current operating data includes the throttle opening, engine speed, engine torque, hydraulic signal of the hydraulic suspension mechanism, posture signal of the agricultural machinery towed or pushed by the off-road vehicle and / or the type of the agricultural machinery towed or pushed by the off-road vehicle.

7. The method according to any one of claims 1 to 6, wherein: The method further comprises the following steps: obtaining a selected work content of the off-road vehicle, the selected work content representing work content being performed by the off-road vehicle as determined by an operator of the off-road vehicle; and A current working content of the off-road vehicle is additionally determined based on the selected working content.

8. The method according to claim 7, wherein: Determining the current operation content of the off-road vehicle additionally based on the selected operation content includes: In the absence of a selected working content, allowing the identified working content to be determined as the current working content of the off-road vehicle; and / or In case of selected job content: - determining the selected work content as the current work content of the off-road vehicle; or - When there is a conflict between the selected work content and the recognized work content, the selected work content is determined as the current work content, and a prompt is output to the operator.

9. The method according to any one of claims 1 to 8, wherein: The method further comprises the following steps: Acquire a previous operation content of the off-road vehicle, wherein the previous operation content represents the most recently determined and being executed operation content for the off-road vehicle up to now; and Additionally, a current work content of the off-road vehicle is determined based on previous work content.

10. The method according to claim 9, wherein: Controlling the operation of the off-road vehicle based additionally on the previous work content includes: Only when it is confirmed that the recognized job content meets the predefined output condition, the recognized job content is determined as the current job content, otherwise the previous job content is continuously determined as the current job content.

11. A device for controlling an off-road vehicle, the device being used to perform the method according to any one of claims 1 to 10, the device comprising: An acquisition module configured to acquire current operation data of the off-road vehicle; An identification module configured to determine an identification operation content of the off-road vehicle based on current operation data of the off-road vehicle, wherein the identification operation content represents an operation content being performed by the off-road vehicle determined by means of a trained machine learning model; a decision module configured to determine a current operation content of the off-road vehicle based at least on the identified operation content; as well as A control module is configured to control the operation of the off-road vehicle according to the determined current work content. 12 . A machine-readable storage medium having a computer program stored thereon, the computer program being configured to execute the method according to claim 1 when executed on a computer.

Images