CN110488821B - Method and device for determining unmanned vehicle motion strategy - Google Patents
Method and device for determining unmanned vehicle motion strategy Download PDFInfo
- Publication number
- CN110488821B CN110488821B CN201910741637.9A CN201910741637A CN110488821B CN 110488821 B CN110488821 B CN 110488821B CN 201910741637 A CN201910741637 A CN 201910741637A CN 110488821 B CN110488821 B CN 110488821B
- Authority
- CN
- China
- Prior art keywords
- unmanned vehicle
- road condition
- attention
- feature vector
- weighted
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course or altitude of land, water, air, or space vehicles, e.g. automatic pilot
- G05D1/02—Control of position or course in two dimensions
Abstract
The present specification discloses a method and an apparatus for determining a motion strategy of an unmanned vehicle, wherein the unmanned vehicle continuously acquires images during the driving process, so that the images acquired at the current moment can be determined, and then the images are input to a coding end of a pre-trained decision model to obtain a road condition feature vector corresponding to the current moment. And then, inputting the traffic characteristic vectors corresponding to the historical moments, the traffic characteristic vectors corresponding to the current moment and the motion strategy of the unmanned vehicle at the current moment, which are obtained historically through a coding end of the decision model, into a decoding end of the decision model to obtain the motion strategy of the unmanned vehicle at the next moment.
Description
Technical Field
The application relates to the technical field of unmanned vehicles, in particular to a method and a device for determining a motion strategy of an unmanned vehicle.
Background
The unmanned vehicle is an intelligent vehicle which senses the surrounding road environment through a sensing system carried by the unmanned vehicle, automatically plans a driving route and controls the vehicle to reach a preset target.
In the prior art, a path planning method for an unmanned vehicle uses a Long Short-Term Memory network (LSTM) -based encoder-decoder structure, outputs predicted coordinates of obstacles at N moments in the future by inputting actual coordinates of the obstacles at M moments in history, and controls the unmanned vehicle to travel according to a preset obstacle avoidance method.
The method has the greatest advantages that the model is simple, and meanwhile, the obstacle avoidance mode of the unmanned vehicle is determined. The defects are obvious, and the obstacle avoidance and path selection effects are poor due to the fact that only the position information of each obstacle is used.
Moreover, the influence of the difference of the traffic information at each time, for example, the influence of the change of the number of roads, the influence of the type of the vehicle, and the like, is not considered. The existing method for determining the unmanned vehicle motion strategy is not flexible and accurate enough.
Disclosure of Invention
The embodiment of the specification provides a method and a device for determining an unmanned vehicle motion strategy, so as to partially solve the problems in the prior art.
The embodiment of the specification adopts the following technical scheme:
the method for determining the unmanned vehicle motion strategy provided by the specification comprises the following steps:
the method comprises the following steps that images are continuously acquired during the driving process of the unmanned vehicle, and comprises the following steps:
determining an image acquired at the current moment of the unmanned vehicle;
inputting the image to a coding end of a pre-trained decision model to obtain a road condition feature vector corresponding to the current moment;
and inputting the road condition characteristic vector corresponding to each historical moment obtained by the coding end, the road condition characteristic vector corresponding to the current moment and the motion strategy of the unmanned vehicle at the current moment into a decoding end of the decision model to obtain the motion strategy of the unmanned vehicle at the next moment.
Optionally, the encoding end includes a convolutional neural network CNN, a first long-short term memory network LSTM;
inputting the image to a coding end of a pre-trained decision model to obtain a road condition feature vector corresponding to the current moment, specifically comprising:
inputting the image at the current moment into the CNN to obtain an image characteristic vector;
and inputting the obtained image feature vector and the road condition feature vector corresponding to the last moment of the current moment into the first LSTM to obtain the road condition feature vector corresponding to the current moment.
Optionally, the decoding end comprises an attention layer, a second long-short term memory network LSTM;
inputting the road condition characteristic vector corresponding to each historical moment obtained by the encoding end, the road condition characteristic vector corresponding to the current moment and the motion strategy of the unmanned vehicle at the current moment into the decoding end of the decision model to obtain the motion strategy of the unmanned vehicle at the next moment, specifically comprising:
determining an attention matrix of the attention layer according to the motion strategy of the unmanned vehicle at the current moment and the position of the unmanned vehicle at the current moment;
inputting the road condition feature vectors corresponding to the historical moments and the road condition feature vector corresponding to the current moment into the attention layer to obtain a road condition feature vector weighted by attention;
and inputting the attention weighted road condition feature vector into the second LSTM to obtain the motion strategy of the unmanned vehicle at the next moment.
Optionally, the step of inputting the traffic characteristic vector corresponding to each historical time and the traffic characteristic vector corresponding to the current time into the attention layer to obtain the attention-weighted traffic characteristic vector includes:
determining a road condition feature matrix according to the road condition feature vector corresponding to each historical moment and the road condition feature vector corresponding to the current moment;
obtaining a road condition characteristic matrix weighted by attention according to the road condition characteristic matrix and the attention matrix;
and determining the attention weighted road condition feature vector by a maximum pooling method according to the attention weighted road condition feature matrix.
Optionally, the decoding end further includes: a road constraint layer;
inputting the attention weighted road condition feature vector into the second LSTM to obtain a motion strategy of the unmanned vehicle at the next moment, specifically comprising:
determining at least one planned path according to the target position of the unmanned vehicle in running and the position of the unmanned vehicle at the current moment;
collecting coordinates of designated points in each planned path, and determining a path characteristic matrix according to the collected coordinates;
determining an attention weighted path feature matrix according to the path feature matrix and the attention matrix of the road constraint layer;
determining attention weighted road condition feature vectors by a maximum pooling method according to the attention weighted path feature matrix;
and inputting the attention weighted road condition feature vector and the attention weighted road condition feature vector into the second LSTM to obtain the motion strategy of the unmanned vehicle at the next moment.
The present specification provides an apparatus for determining a motion strategy of an unmanned vehicle, the apparatus continuously acquiring images during the driving of the unmanned vehicle, the apparatus comprising:
the acquisition module is used for determining an image acquired by the unmanned vehicle at the current moment;
the coding module is used for inputting the image to a coding end of a pre-trained decision model so as to obtain a road condition characteristic vector corresponding to the current moment;
and the strategy determining module is used for inputting the road condition characteristic vector corresponding to each historical moment obtained by the coding end, the road condition characteristic vector corresponding to the current moment and the motion strategy of the unmanned vehicle at the current moment into the decoding end of the decision model so as to obtain the motion strategy of the unmanned vehicle at the next moment.
Optionally, the encoding end includes a convolutional neural network CNN and a first long-short term memory network LSTM, and the encoding module inputs the image at the current time to the CNN to obtain an image feature vector, and inputs the obtained image feature vector and a road condition feature vector corresponding to the previous time at the current time to the first LSTM to obtain the road condition feature vector corresponding to the current time.
Optionally, the decoding end includes an attention layer and a second long-short term memory network LSTM, and the policy determining module determines the attention matrix of the attention layer according to the motion policy of the unmanned vehicle at the current time and the position of the unmanned vehicle at the current time, inputs the road condition feature vectors corresponding to the historical times and the road condition feature vectors corresponding to the current time into the attention layer to obtain an attention-weighted road condition feature vector, and inputs the attention-weighted road condition feature vector into the second LSTM to obtain the motion policy of the unmanned vehicle at the next time.
Optionally, the policy determining module determines a traffic characteristic matrix according to the traffic characteristic vectors corresponding to the historical moments and the traffic characteristic vector corresponding to the current moment, obtains an attention weighted traffic characteristic matrix according to the traffic characteristic matrix and the attention matrix, and determines the attention weighted traffic characteristic vector according to the attention weighted traffic characteristic matrix by a maximum pooling method.
Optionally, the decoding end further includes: the strategy determination module determines at least one planned path according to the target position of the unmanned vehicle in running and the position of the unmanned vehicle at the current moment, acquires coordinates of designated points in each planned path, determines a path feature matrix according to the acquired coordinates, determines an attention weighted path feature matrix according to the path feature matrix and an attention matrix of the road constraint layer, determines an attention weighted road condition feature vector according to the attention weighted path feature matrix through a maximum pooling method, and inputs the attention weighted road condition feature vector and the attention weighted road condition feature vector into the second LSTM to obtain the motion strategy of the unmanned vehicle at the next moment.
The present specification provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program, when executed by a processor, implements the above-mentioned method for determining an unmanned vehicle motion policy.
The unmanned vehicle provided by the specification comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, and is characterized in that the processor implements the method for determining the unmanned vehicle motion strategy when executing the program.
The embodiment of the specification adopts at least one technical scheme which can achieve the following beneficial effects:
firstly, the unmanned vehicle continuously acquires images in the driving process, so that the images acquired at the current moment can be determined, and then the images are input to a coding end of a pre-trained decision model to obtain a road condition feature vector corresponding to the current moment. And then, inputting the traffic characteristic vectors corresponding to the historical moments, the traffic characteristic vectors corresponding to the current moment and the motion strategy of the unmanned vehicle at the current moment, which are obtained historically through a coding end of the decision model, into a decoding end of the decision model to obtain the motion strategy of the unmanned vehicle at the next moment. And determining the input of a decoding end of the decision model according to road condition characteristics obtained by a coding end of the decision model according to the images acquired at the current moment and the images acquired at all historical moments, and determining the motion strategy of the unmanned vehicle at the next moment through the decoding end of the decision model based on the motion strategy of the vehicle at the current moment. Different from the prior art that the movement strategy is determined only by using the position information of each obstacle, the position information of the obstacles can be determined through the acquired images, and the road condition on the current unmanned vehicle driving route can also be determined, so that the determined movement strategy is more flexible. And the change trend of the road condition can be determined by fully utilizing the collected images at the historical moment. The defects in the prior art are avoided, and the determined motion strategy is more accurate.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the application and together with the description serve to explain the application and not to limit the application. In the drawings:
FIG. 1 is a process for determining an unmanned vehicle motion strategy provided herein;
FIG. 2 is a block diagram of a coding end of a decision model provided in the present specification;
fig. 3 is a structure of an encoding end and a decoding end of a decision model provided in the present specification;
FIG. 4 is a schematic diagram providing different importance at different times of the present specification;
fig. 5 is a schematic diagram of a road condition feature vector input decoding end provided in the present specification;
FIG. 6 is a schematic illustration of an alternative path provided herein;
fig. 7 is a schematic structural diagram of an apparatus for determining an unmanned vehicle motion strategy according to an embodiment of the present disclosure;
fig. 8 is a schematic view of an unmanned vehicle corresponding to fig. 1 provided in an embodiment of the present disclosure.
Detailed Description
In order to make the objects, technical solutions and advantages of the present disclosure more apparent, the technical solutions of the present disclosure will be clearly and completely described below with reference to the specific embodiments of the present disclosure and the accompanying drawings. It should be apparent that the described embodiments are only some of the embodiments of the present application, and not all of the embodiments. All other embodiments obtained by a person skilled in the art based on the embodiments in the present specification without any inventive step are within the scope of the present application.
The technical solutions provided by the embodiments of the present application are described in detail below with reference to the accompanying drawings.
Fig. 1 is a process for determining an unmanned vehicle motion strategy according to an embodiment of the present disclosure, which may specifically include the following steps:
s102: and determining the image acquired by the unmanned vehicle at the current moment.
In this specification, the unmanned vehicle may continuously acquire images during the driving process of the unmanned vehicle through an image sensor carried by the unmanned vehicle, and may determine the images acquired at the current time in order to determine the motion strategy of the unmanned vehicle at the next time. The process may be specifically executed by a device for determining a motion strategy in the unmanned vehicle, and this specification is not limited thereto, and for convenience of description, the unmanned vehicle is used as an execution subject for the following description.
Specifically, the unmanned vehicle can acquire images in the advancing direction through the sensor. In addition, other vehicles, obstacles and road conditions on the road are gradually changed along with time, so that the sensor can continuously acquire images. The interval for acquiring the images can be set according to the needs, and the description is not limited, for example, the images are acquired at 1/24 seconds intervals, the images are acquired at 1/60 seconds intervals, and the like. The time difference between the current time and the previous time is the interval time for acquiring the image.
S104: and inputting the image to a coding end of a pre-trained decision model to obtain the road condition characteristic vector corresponding to the current moment.
In this specification, after the acquired image at the current time, the unmanned vehicle may input the image to a coding end of a pre-trained decision model to obtain a road condition feature vector corresponding to the current time.
Specifically, in the present specification, the structure of the decision model may be as shown in fig. 2, where the left side is the encoding side and the right side is the decoding side. It can be seen that the encoding end comprises: convolutional Neural Networks (CNN) and first LSTM. The unmanned vehicle can input the image at the current moment into the CNN to obtain the image characteristic vector output by the CNN, and then the obtained image characteristic vector and the road condition characteristic vector corresponding to the previous moment at the current moment are input into the first LSTM to obtain the road condition characteristic vector at the current moment.
That is to say, by using the LSTM, important information can be memorized for a long time, and the characteristics of useless information are forgotten, and the road condition feature vector at each time is not determined based on the image acquired at the time, but also each road condition feature vector determined at a plurality of historical times participates in the process of determining the road condition feature vector.
For example, as shown in fig. 2, the numerical value of t indicates different times, t ═ 0s indicates the current time, and times smaller than 0s are all history times. For example, t-0.2 s represents a history time 0.2s before the current time. After the encoder collects the image every time, the same CNN is adopted to extract the image feature vector. And according to the time sequence of the collected images, inputting the image characteristic vectors corresponding to the historical moments into the LSTM, and continuously outputting the coding result at each moment through the LSTM, wherein the coding result is the road condition characteristic vector output by the LSTM at each moment.
Of course, the data source of the whole decision model is the collected image, and the image includes road signs (e.g., lane lines, road traffic signs, etc.) and other environmental information (e.g., road shoulders, guardrails, light poles), etc., in addition to obstacles (e.g., other vehicles) on the road. Therefore, the information included in the image feature vector obtained by CNN also includes the features of these road signs and other environmental information. The road condition feature vector obtained by the LSTM of the encoder includes not only the feature vector of the obstacle in the prior art, but also other feature vectors in the road surface. In addition, since the candidate needs to determine the motion strategy of the unmanned vehicle through a decoder of the decision model, all the road condition feature vectors are feature vectors for determining the motion strategy of the unmanned vehicle.
It should be noted that, in this specification, the decision model may be a Sequence to Sequence (Seq 2Seq) model, and fig. 2 may be regarded as an encoding end and a decoding end of the Seq2Seq model.
S106: and inputting the road condition characteristic vector corresponding to each historical moment obtained by the coding end, the road condition characteristic vector corresponding to the current moment and the motion strategy of the unmanned vehicle at the current moment into a decoding end of the decision model to obtain the motion strategy of the unmanned vehicle at the next moment.
In this specification, after determining the road condition feature vector at the current time through the encoding end of the decision model, the unmanned vehicle may input the obtained road condition feature vectors corresponding to the historical times, the road condition feature vector at the current time obtained in the previous step, and the motion strategy of the unmanned vehicle at the current time into the decoding end of the decision model, so as to obtain the motion strategy of the unmanned vehicle at the next time output by the decoding end. The decoding end outputs the motion strategy at the next moment, so the motion strategy at the current moment is the motion strategy output by the decoding end at the previous historical moment of the current moment.
Specifically, the decoding end of the decision model includes the attention layer and the second LSTM, and the decision model can be as shown in fig. 3. The input of the second LSTM is the road condition feature vector of each time outputted by the encoding end, wherein each time comprises the current time and each historical time. Since the importance of the road condition feature vectors at different times to determine the motion strategy is different, the attention layer is required to perform attention weighting on the road condition feature vectors, and then the second LSTM is input.
Fig. 4 is a schematic diagram of the difference in importance of different times provided in this specification, in which there are 4 historical times and a current time in total, a solid rectangle represents other vehicles on the road, an open rectangle represents the unmanned vehicle, a solid line represents shoulders on both sides of the road, a broken line represents a yellow broken line in the road sign, and an arrow represents the traveling direction of each vehicle. In the above figure, the other vehicles and the unmanned vehicle are both traveling in the same direction, the importance of the historical time t-1.5 s and the importance of the historical time t-1.0 s are relatively low, it is determined that the unmanned vehicle needs to follow the vehicle, the importance of the historical time t-0.5 s and the current time t-0 s is relatively high, it is determined that the motion strategy of the unmanned vehicle at the next time (for example, whether to overtake or continue following the vehicle) is determined, and similarly, the importance of the historical time t-1.5 s and the historical time t-1.0 s is relatively high when the other vehicles travel in the same direction in the same drawing. The attention layer performs attention weighting on the road condition feature vectors at different moments, so as to distinguish the importance of the different road condition feature vectors.
In this specification, first, the unmanned vehicle may determine an attention matrix of an attention layer of the decoding end according to a motion strategy of the unmanned vehicle at the current time and a position of the unmanned vehicle at the current time.
And then, inputting the road condition characteristic vectors corresponding to the historical moments and the road condition characteristic vector corresponding to the current moment into the attention layer, and obtaining the attention weighted road condition characteristic vector through the attention matrix.
And finally, inputting the road condition feature vector weighted by attention into a second LSTM to obtain the motion strategy of the unmanned vehicle at the next moment.
Specifically, the number of road condition feature vectors input into the attention layer is predetermined, and the historical time required to be input can be updated continuously according to a period. For example, assuming that the number of the least required road condition feature vectors at the historical time is 10, the number of the road condition feature vectors input into the attention layer for the first time in each update period is 11 (i.e., the road condition feature vectors at the historical time before 10 current periods and the road condition feature vectors at 1 current time of the current period). Assuming that the update period is 10 times, the traffic characteristic vectors of the last time attention layer is input in each update period are 20 (i.e., the traffic characteristic vectors of the historical times before 10 current periods, the traffic characteristic vectors of the historical times of 9 current periods, and the traffic characteristic vectors of 1 current time in the current period). For example, fig. 5 shows a schematic road condition feature vector input diagram, where arrows represent a time axis, in a current update cycle, a vector input at a first time (i.e., t10) includes road condition feature vectors at 10 historical times before the current time and the road condition feature vector at the first time, and a vector input at a second time (i.e., t11) includes road condition feature vectors at 10 historical times before the current time, and road condition feature vectors at the first time and the second time until a last time (i.e., t19) of the current update cycle, and road condition feature vectors at 20 times are input. The first time (i.e., t20) of the next update cycle thereafter has t10 to t19 as the history time. Of course, the remaining traffic condition feature vectors are historical traffic condition feature vectors except the current traffic condition feature vector. Of course, the number of the attention layers is input into the specific road condition feature vector, and the specific road condition feature vector is updated according to what period, which may be set as required, and the description is not limited herein.
Further, since the number of the road condition feature vectors input into the attention layer at different times is not exactly the same, and the lengths of the attention weighted road condition feature vectors input into the second LSTM at different times need to be the same, the attention layer also needs to cross-multiply each road condition feature vector input into the decoder with the attention moment matrix, and unify the lengths of the attention weighted road condition feature vectors input into the second LSTM. That is, the attention matrix includes, in addition to the attention weighting function, a function of unifying the lengths of the attention weighted road condition feature vectors.
Specifically, firstly, the unmanned vehicle can determine a road condition feature matrix according to the road condition feature vectors corresponding to the historical moments and the road condition feature vectors corresponding to the current moment. And then, cross-multiplying the road condition characteristic matrix and the attention matrix to obtain an attention weighted road condition characteristic matrix. Finally, the attention weighted road condition feature vector, that is, the vector of the input second LSTM, is determined from the attention weighted road condition feature matrix by the maximum pooling method.
The unmanned vehicle can be generated through a mapping function according to the motion strategy of the current moment and the position of the unmanned vehicle at the current moment.
Specifically, the unmanned vehicle may determine a change value of a position at the present time with respect to a position at a time immediately before the present time by dXt＝0And dYt＝0Where X and Y represent two coordinate values (e.g., longitude and latitude) of horizontal and vertical, t ═ 0 represents the current time, d represents a variation value, and, for example, if an unmanned vehicle acquires an image every 0.5s, the interval between each historical time is 0.5s, and dX represents the time at which the vehicle has been started, and the time at which the vehicle has been started is 0.5st＝0The displacement on the X coordinate axis at the time t-0 s with respect to the t-05 s unmanned vehicle is shown, and dY is the samet＝0The displacement of the unmanned vehicle on the Y coordinate axis at the time t-0 s is shown with respect to t-05 s.
By mapping function g, the unmanned vehicle can be according to Xt，Yt，dXt，dYtAn attention matrix for the current time instance is determined. The formula is expressed specifically as g (X)t，Yt，dXt，dYt) Wherein the g-function is used to map the 4-dimensional data to a multi-dimensional space, wherein the dimension of the multi-dimensional space is determined according to the requirements of the attention matrix. For example, the attention moment matrix needs to convert the feature vectors of each road condition into a matrix composed of the feature vectors of attention weighted road conditions with a length of 32 bits after weighting, and then the g function needs to be according to Xt，Yt，dXt，dYtA corresponding attention matrix is generated.
Furthermore, after the unmanned vehicle obtains the attention-weighted road condition feature matrix, the attention-weighted road condition feature vector to be input into the second LSTM can be determined from the attention-weighted road condition feature matrix by the max-pool method. Of course, since the maximum pooling method is already mature and is used for determining the maximum result from multiple data (or features or vectors), the process of specifically determining the attention-weighted road condition feature vector is not repeated in this specification.
In addition, like the first LSTM, the data input each time also includes, in addition to the newly generated attention weighted road condition feature vector, the motion strategy output by the second LSTM at the time immediately before the current time, and outputs the unmanned vehicle motion strategy at the time immediately after the current time.
That is, the decoding end of the decision model determines the attention matrix at the current time according to the motion state of the unmanned vehicle at the current time, determines the attention weighted road condition feature vector from the road condition feature vectors corresponding to each time (including the historical time and the current time) by the maximum pooling method, inputs the determined attention weighted road condition feature vector into the second LSTM, and determines the motion strategy of the unmanned vehicle at the next time by integrating the motion strategies output at each historical time by using the characteristics of the LSTM.
It should be noted that, in this specification, the CNN, the first LSTM, the g function, and the second LSTM are mathematical expressions of portions of the whole decision model that play different roles, and may be regarded as a whole model in practice. And therefore is also trained as a whole when training the decision model.
Specifically, the structure of the decision model is as shown above, the training sample is an image continuously acquired in a plurality of driving processes of a vehicle driven by a person, the image acquired in one driving process is used as one training sample, and the actual route of the driver in the driving process is used as supervision. Namely, the decision model is trained according to the motion route of the vehicle under the actual operation of the driver at each moment and the motion strategy output by the decision model at each time corresponding to the moment. It is also understood that each training is a training of the CNN, the first LSTM, the g-function, and the second LSTM together.
Based on the method for determining the unmanned vehicle motion strategy shown in fig. 1, the unmanned vehicle continuously acquires images during the driving process, so that the images acquired at the current moment can be determined, and then the images are input to the coding end of the pre-trained decision model to obtain the road condition feature vector corresponding to the current moment. And then, inputting the traffic characteristic vectors corresponding to the historical moments, the traffic characteristic vectors corresponding to the current moment and the motion strategy of the unmanned vehicle at the current moment, which are obtained historically through a coding end of the decision model, into a decoding end of the decision model to obtain the motion strategy of the unmanned vehicle at the next moment. And determining the input of a decoding end of the decision model according to road condition characteristics obtained by a coding end of the decision model according to the images acquired at the current moment and the images acquired at all historical moments, and determining the motion strategy of the unmanned vehicle at the next moment through the decoding end of the decision model based on the motion strategy of the vehicle at the current moment. Different from the prior art that the movement strategy is determined only by using the position information of each obstacle, the position information of the obstacles can be determined through the acquired images, and the road condition on the current unmanned vehicle driving route can also be determined, so that the determined movement strategy is more flexible. And the change trend of the road condition can be determined by fully utilizing the collected images at the historical moment. The defects in the prior art are avoided, and the determined motion strategy is more accurate.
In addition, in this specification, the decoding end of the decision model may further include: and (6) a road constraint layer. The road constraint layer is configured to constrain the motion strategy output by the second LSTM to generally direct the unmanned vehicle to the destination based on the destination to which the unmanned vehicle is to travel.
Specifically, first, the unmanned vehicle may determine at least one planned route according to a target position of travel and a position of the unmanned vehicle at the present time. Since there are usually many paths from one point to another, no one can plan the road conditions at any time and determine a plurality of paths to the target location. It should be noted that the path planning described in this specification includes the selection of lanes, for example, if the unmanned vehicle runs on a road with six bidirectional lanes, the path planning may at least include the path where the unmanned vehicle continues to run on the current lane and the path where the unmanned vehicle runs to the other two lanes in parallel, as shown in fig. 6.
Fig. 6 is a schematic diagram of the selectable paths provided in the present specification, in which it can be seen that from the current position of no person to the target position, although there are only two paths, since one path has 3 lanes and the other has 2 lanes, 5 paths can be determined.
And then, the unmanned vehicle can acquire the coordinates of the designated points in each planned path and determine a path characteristic matrix according to the acquired coordinates. Wherein, the number of the designated points and the spacing between the designated points can be set according to the requirement. For example, the specified number of points is 10, the distance between the points is 10 meters, and for one path, the unmanned vehicle can acquire the coordinates of 10 points with the distance between the points in the advancing direction of the unmanned vehicle on the path. Therefore, the unmanned vehicle can obtain a path feature matrix, each row in the path feature matrix corresponds to a plurality of coordinate values in one path, and actually represents a path which can be selected at the next moment of the unmanned vehicle, so that the motion strategy of the unmanned vehicle is constrained.
Then, the unmanned vehicle can determine an attention weighted path feature matrix according to the path feature matrix and the attention matrix of the road constraint layer. This process is consistent with the process of determining the attention weighted road condition feature matrix described in step S104. The attention moment matrix can be determined by the same method, and certainly, the attention weighted road condition feature matrix and the attention weighted path feature matrix have different requirements, so that the two trained attention matrixes may not be identical.
And finally, determining the road condition feature vector weighted by attention by a maximum pooling method according to the attention weighted path feature matrix. And similarly, determining the attention weighted road condition feature vector by adopting a maximum pooling method.
At this point, the unmanned vehicle can input the obtained attention weighted road condition feature vector and the attention weighted road condition feature vector into the second LSTM to obtain the motion strategy of the unmanned vehicle with path constraint at the next moment.
Specifically, the unmanned vehicle may add the products of the two attention-weighted vectors and the preset coefficients, respectively, according to the preset coefficients, to obtain a vector that is finally input to the second LSTM. The two coefficients are normalized, and specific values can be set according to needs, which is not limited in the present application. For example, if the path constraint needs to be stronger, the coefficient values of the attention weighted path feature vector are increased.
In addition, in this specification, if the unmanned vehicle starts to acquire images after being started, at the time of the unmanned vehicle starting, the unmanned vehicle does not acquire images at historical times yet, and therefore the unmanned vehicle cannot acquire the road condition feature vectors corresponding to the historical times obtained by the encoding end in step S106, because there is no image acquired at the historical time, there is no road condition feature vector corresponding to the historical time. At this time, the unmanned vehicle may first move according to a preset motion strategy (e.g., advance at 0.1km/hde1 for 10 seconds), and then determine the motion strategy through the method described herein. Alternatively, the unmanned vehicle may not move first, but wait for a preset time period before determining the movement strategy according to the method described in the present specification. That is, the motion strategy output by the decision model in this specification requires images acquired at a plurality of historical times, and when the historical times are few, the accuracy of the motion strategy output by the decision model may be reduced, so that the unmanned vehicle can wait for a period of time, and the motion strategy output by the decision model is utilized to control unmanned motion after the decision model is sufficiently warmed up.
Based on the method for determining the unmanned vehicle motion strategy shown in fig. 1, an embodiment of the present specification further provides a schematic structural diagram of an apparatus for determining the unmanned vehicle motion strategy, as shown in fig. 7.
Fig. 7 is a schematic structural diagram of an apparatus for determining an unmanned vehicle motion strategy according to an embodiment of the present disclosure, where the apparatus continuously acquires images during the driving of the unmanned vehicle, and the apparatus includes:
the acquisition module 200 is used for determining the image acquired by the unmanned vehicle at the current moment;
the encoding module 202 is configured to input the image to an encoding end of a pre-trained decision model to obtain a road condition feature vector corresponding to the current time;
the policy determining module 204 is configured to input the traffic characteristic vector corresponding to each historical time obtained by the encoding end, the traffic characteristic vector corresponding to the current time, and the motion policy of the unmanned vehicle at the current time into the decoding end of the decision model, so as to obtain the motion policy of the unmanned vehicle at the next time.
Optionally, the encoding end includes a convolutional neural network CNN and a first long-short term memory network LSTM, and the encoding module 202 inputs the image at the current time to the CNN to obtain an image feature vector, and inputs the obtained image feature vector and a road condition feature vector corresponding to the previous time at the current time to the first LSTM to obtain the road condition feature vector corresponding to the current time.
Optionally, the decoding end includes an attention layer and a second long-short term memory network LSTM, and the policy determining module 204 determines the attention matrix of the attention layer according to the motion policy of the unmanned vehicle at the current time and the position of the unmanned vehicle at the current time, inputs the road condition feature vectors corresponding to the historical times and the road condition feature vector corresponding to the current time into the attention layer to obtain an attention-weighted road condition feature vector, and inputs the attention-weighted road condition feature vector into the second LSTM to obtain the motion policy of the unmanned vehicle at the next time.
Optionally, the policy determining module 204 determines a traffic characteristic matrix according to the traffic characteristic vectors corresponding to the historical moments and the traffic characteristic vector corresponding to the current moment, obtains an attention weighted traffic characteristic matrix according to the traffic characteristic matrix and the attention matrix, and determines the attention weighted traffic characteristic vector according to the attention weighted traffic characteristic matrix by a maximum pooling method.
Optionally, the decoding end further includes: the policy determination module 204 is configured to determine at least one planned path according to the target position where the unmanned vehicle travels and the position of the unmanned vehicle at the current time, acquire coordinates of designated points in each planned path, determine a path feature matrix according to the acquired coordinates, determine an attention weighted path feature matrix according to the path feature matrix and an attention matrix of the road constraint layer, determine an attention weighted road condition feature vector according to the attention weighted path feature matrix by a maximum pooling method, input the attention weighted road condition feature vector and the attention weighted road condition feature vector into the second LSTM, and obtain a motion policy of the unmanned vehicle at the next time.
The embodiment of the specification also provides a computer readable storage medium, and the storage medium stores a computer program, and the computer program can be used for executing the method for determining the unmanned vehicle motion strategy provided by the above-mentioned fig. 1.
Based on the method for determining the unmanned vehicle motion strategy shown in fig. 1, the embodiment of the present specification further proposes a schematic structure diagram of the unmanned vehicle shown in fig. 8. As shown in fig. 8, at the hardware level, the unmanned vehicle includes a processor, an internal bus, a network interface, a memory, and a nonvolatile memory, but may also include hardware required for other services. The processor reads a corresponding computer program from the nonvolatile memory into the memory and then runs the computer program to implement the method for determining the unmanned vehicle motion strategy described in fig. 1.
Of course, besides the software implementation, the present specification does not exclude other implementations, such as logic devices or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, and may be hardware or logic devices.
In the 90 s of the 20 th century, improvements in a technology could clearly distinguish between improvements in hardware (e.g., improvements in circuit structures such as diodes, transistors, switches, etc.) and improvements in software (improvements in process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical modules. For example, a Programmable Logic Device (PLD), such as a Field Programmable Gate Array (FPGA), is an integrated circuit whose Logic functions are determined by programming the Device by a user. A digital system is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Furthermore, nowadays, instead of manually making an Integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development and writing, but the original code before compiling is also written by a specific Programming Language, which is called Hardware Description Language (HDL), and HDL is not only one but many, such as abel (advanced Boolean Expression Language), ahdl (alternate Hardware Description Language), traffic, pl (core universal Programming Language), HDCal (jhdware Description Language), lang, Lola, HDL, laspam, hardward Description Language (vhr Description Language), vhal (Hardware Description Language), and vhigh-Language, which are currently used in most common. It will also be apparent to those skilled in the art that hardware circuitry that implements the logical method flows can be readily obtained by merely slightly programming the method flows into an integrated circuit using the hardware description languages described above.
The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic for the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller as pure computer readable program code, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be considered a hardware component, and the means included therein for performing the various functions may also be considered as a structure within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.
The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. One typical implementation device is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functions of the various elements may be implemented in the same one or more software and/or hardware implementations of the present description.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
This description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
The above description is only an example of the present specification, and is not intended to limit the present specification. Various modifications and alterations to this description will become apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification.
Claims (12)
1. A method of determining an unmanned vehicle motion strategy, wherein images are continuously acquired by an unmanned vehicle during travel, the method comprising:
determining an image acquired at the current moment of the unmanned vehicle;
inputting the image to a coding end of a pre-trained decision model to obtain a road condition feature vector corresponding to the current moment;
and inputting the road condition characteristic vector corresponding to each historical moment obtained by the coding end, the road condition characteristic vector corresponding to the current moment and the motion strategy of the unmanned vehicle at the current moment into a decoding end of the decision model to obtain the motion strategy of the unmanned vehicle at the next moment.
2. The method of claim 1, wherein the encoding end comprises a Convolutional Neural Network (CNN), a first long short term memory network (LSTM);
inputting the image to a coding end of a pre-trained decision model to obtain a road condition feature vector corresponding to the current moment, specifically comprising:
inputting the image at the current moment into the CNN to obtain an image characteristic vector;
and inputting the obtained image feature vector and the road condition feature vector corresponding to the last moment of the current moment into the first LSTM to obtain the road condition feature vector corresponding to the current moment.
3. The method of claim 1, wherein the decoding end comprises an attention layer, a second long short term memory network (LSTM);
inputting the road condition characteristic vector corresponding to each historical moment obtained by the encoding end, the road condition characteristic vector corresponding to the current moment and the motion strategy of the unmanned vehicle at the current moment into the decoding end of the decision model to obtain the motion strategy of the unmanned vehicle at the next moment, specifically comprising:
determining an attention matrix of the attention layer according to the motion strategy of the unmanned vehicle at the current moment and the position of the unmanned vehicle at the current moment;
inputting the road condition feature vectors corresponding to the historical moments and the road condition feature vector corresponding to the current moment into the attention layer to obtain a road condition feature vector weighted by attention;
and inputting the attention weighted road condition feature vector into the second LSTM to obtain the motion strategy of the unmanned vehicle at the next moment.
4. The method as claimed in claim 3, wherein the step of inputting the traffic condition feature vector corresponding to each historical time and the traffic condition feature vector corresponding to the current time into the attention layer to obtain the attention-weighted traffic condition feature vector comprises:
determining a road condition feature matrix according to the road condition feature vector corresponding to each historical moment and the road condition feature vector corresponding to the current moment;
obtaining a road condition characteristic matrix weighted by attention according to the road condition characteristic matrix and the attention matrix;
and determining the attention weighted road condition feature vector by a maximum pooling method according to the attention weighted road condition feature matrix.
5. The method of claim 3, wherein the decoding side further comprises: a road constraint layer;
inputting the attention weighted road condition feature vector into the second LSTM to obtain a motion strategy of the unmanned vehicle at the next moment, specifically comprising:
determining at least one planned path according to the target position of the unmanned vehicle in running and the position of the unmanned vehicle at the current moment;
collecting coordinates of designated points in each planned path, and determining a path characteristic matrix according to the collected coordinates;
determining an attention weighted path feature matrix according to the path feature matrix and the attention matrix of the road constraint layer;
determining attention weighted road condition feature vectors by a maximum pooling method according to the attention weighted path feature matrix;
and inputting the attention weighted road condition feature vector and the attention weighted road condition feature vector into the second LSTM to obtain the motion strategy of the unmanned vehicle at the next moment.
6. An apparatus for determining an unmanned vehicle motion strategy, the apparatus continuously acquiring images during the unmanned vehicle driving, the apparatus comprising:
the acquisition module is used for determining an image acquired by the unmanned vehicle at the current moment;
the coding module is used for inputting the image to a coding end of a pre-trained decision model so as to obtain a road condition characteristic vector corresponding to the current moment;
and the strategy determining module is used for inputting the road condition characteristic vector corresponding to each historical moment obtained by the coding end, the road condition characteristic vector corresponding to the current moment and the motion strategy of the unmanned vehicle at the current moment into the decoding end of the decision model so as to obtain the motion strategy of the unmanned vehicle at the next moment.
7. The apparatus according to claim 6, wherein the encoding end includes a Convolutional Neural Network (CNN) and a first long-short term memory network (LSTM), and the encoding module inputs the image at the current time to the CNN to obtain an image feature vector, and inputs the obtained image feature vector and a traffic feature vector corresponding to a previous time of the current time to the first LSTM to obtain the traffic feature vector corresponding to the current time.
8. The apparatus of claim 6, wherein the decoding end comprises an attention layer and a second long-short term memory network (LSTM), and the policy determining module determines the attention matrix of the attention layer according to the motion policy of the unmanned vehicle at the current time and the position of the unmanned vehicle at the current time, inputs the road condition feature vector corresponding to each historical time and the road condition feature vector corresponding to the current time into the attention layer to obtain an attention-weighted road condition feature vector, and inputs the attention-weighted road condition feature vector into the second LSTM to obtain the motion policy of the unmanned vehicle at the next time.
9. The apparatus according to claim 8, wherein the policy determining module determines a traffic characteristic matrix according to the traffic characteristic vectors corresponding to the historical moments and the current moment, obtains an attention-weighted traffic characteristic matrix according to the traffic characteristic matrix and the attention matrix, and determines the attention-weighted traffic characteristic vector according to the attention-weighted traffic characteristic matrix by a maximum pooling method.
10. The apparatus of claim 8, wherein the decoding side further comprises: the strategy determination module determines at least one planned path according to the target position of the unmanned vehicle in running and the position of the unmanned vehicle at the current moment, acquires coordinates of designated points in each planned path, determines a path feature matrix according to the acquired coordinates, determines an attention weighted path feature matrix according to the path feature matrix and an attention matrix of the road constraint layer, determines an attention weighted road condition feature vector according to the attention weighted path feature matrix through a maximum pooling method, and inputs the attention weighted road condition feature vector and the attention weighted road condition feature vector into the second LSTM to obtain the motion strategy of the unmanned vehicle at the next moment.
11. A computer-readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of any of the preceding claims 1-5.
12. An unmanned vehicle comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements the method of any of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910741637.9A CN110488821B (en) | 2019-08-12 | 2019-08-12 | Method and device for determining unmanned vehicle motion strategy |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910741637.9A CN110488821B (en) | 2019-08-12 | 2019-08-12 | Method and device for determining unmanned vehicle motion strategy |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110488821A CN110488821A (en) | 2019-11-22 |
| CN110488821B true CN110488821B (en) | 2020-12-29 |
Family
ID=68550611
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910741637.9A Active CN110488821B (en) | 2019-08-12 | 2019-08-12 | Method and device for determining unmanned vehicle motion strategy |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110488821B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111046981B (en) * | 2020-03-17 | 2020-07-03 | 北京三快在线科技有限公司 | Training method and device for unmanned vehicle control model |
| CN111414852A (en) * | 2020-03-19 | 2020-07-14 | 驭势科技(南京)有限公司 | Image prediction and vehicle behavior planning method, device and system and storage medium |
| CN111152796B (en) * | 2020-04-07 | 2020-08-07 | 北京三快在线科技有限公司 | Vehicle motion state prediction method and device |
| CN111366150B (en) * | 2020-04-20 | 2022-03-18 | Oppo广东移动通信有限公司 | Running direction detection method and device, electronic equipment and storage medium |
| CN111552294B (en) * | 2020-05-14 | 2024-03-26 | 山东师范大学 | Outdoor robot path finding simulation system and method based on time dependence |
| CN113359744B (en) * | 2021-06-21 | 2022-03-01 | 暨南大学 | Robot obstacle avoidance system based on safety reinforcement learning and visual sensor |
| CN116168362A (en) * | 2023-02-27 | 2023-05-26 | 小米汽车科技有限公司 | Pre-training method and device for vehicle perception model, electronic equipment and vehicle |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109816027A (en) * | 2019-01-29 | 2019-05-28 | 北京三快在线科技有限公司 | Training method, device and the unmanned equipment of unmanned decision model |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107092254B (en) * | 2017-04-27 | 2019-11-29 | 北京航空航天大学 | A kind of design method of the Household floor-sweeping machine device people based on depth enhancing study |
| CN107292229A (en) * | 2017-05-08 | 2017-10-24 | 北京三快在线科技有限公司 | A kind of image-recognizing method and device |
| CN109409497B (en) * | 2017-08-15 | 2021-03-16 | 阿里巴巴（中国）有限公司 | Road condition prediction method and device |
| CN107563332A (en) * | 2017-09-05 | 2018-01-09 | 百度在线网络技术（北京）有限公司 | For the method and apparatus for the driving behavior for determining unmanned vehicle |
| US10611371B2 (en) * | 2017-09-14 | 2020-04-07 | Toyota Motor Engineering & Manufacturing North America, Inc. | System and method for vehicle lane change prediction using structural recurrent neural networks |
| US10866588B2 (en) * | 2017-10-16 | 2020-12-15 | Toyota Research Institute, Inc. | System and method for leveraging end-to-end driving models for improving driving task modules |
| CN108319909B (en) * | 2018-01-29 | 2021-11-30 | 清华大学 | Driving behavior analysis method and system |
| CN109697458A (en) * | 2018-11-27 | 2019-04-30 | 深圳前海达闼云端智能科技有限公司 | Control equipment mobile method, apparatus, storage medium and electronic equipment |
| CN109976153B (en) * | 2019-03-01 | 2021-03-26 | 北京三快在线科技有限公司 | Method and device for controlling unmanned equipment and model training and electronic equipment |
| CN110083163A (en) * | 2019-05-20 | 2019-08-02 | 三亚学院 | A kind of 5G C-V2X bus or train route cloud cooperation perceptive method and system for autonomous driving vehicle |
-
2019
- 2019-08-12 CN CN201910741637.9A patent/CN110488821B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109816027A (en) * | 2019-01-29 | 2019-05-28 | 北京三快在线科技有限公司 | Training method, device and the unmanned equipment of unmanned decision model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110488821A (en) | 2019-11-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110488821B (en) | Method and device for determining unmanned vehicle motion strategy | |
| CN111190427B (en) | Method and device for planning track | |
| CN110989636B (en) | Method and device for predicting track of obstacle | |
| CN111114543B (en) | Trajectory prediction method and device | |
| CN112364997B (en) | Method and device for predicting track of obstacle | |
| CN110929431B (en) | Training method and device for vehicle driving decision model | |
| CN111208838B (en) | Control method and device of unmanned equipment | |
| CN110262486B (en) | Unmanned equipment motion control method and device | |
| CN112015847B (en) | Obstacle trajectory prediction method and device, storage medium and electronic equipment | |
| CN111508258B (en) | Positioning method and device | |
| CN111076739B (en) | Path planning method and device | |
| CN111079721A (en) | Method and device for predicting track of obstacle | |
| CN111238523B (en) | Method and device for predicting motion trail | |
| CN111288971B (en) | Visual positioning method and device | |
| CN112033421B (en) | Method and device for detecting lane in electronic map | |
| CN110942181A (en) | Method and device for predicting obstacle track | |
| CN112629550B (en) | Method and device for predicting obstacle track and model training | |
| CN112677993A (en) | Model training method and device | |
| CN113968243B (en) | Obstacle track prediction method, device, equipment and storage medium | |
| CN110530398B (en) | Method and device for detecting precision of electronic map | |
| CN114332808A (en) | Method and device for predicting steering intention | |
| CN111123957B (en) | Method and device for planning track | |
| CN110895406B (en) | Method and device for testing unmanned equipment based on interferent track planning | |
| CN113340311B (en) | Path planning method and device for unmanned equipment | |
| CN112987754B (en) | Unmanned equipment control method and device, storage medium and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |