CN112712595A - Method and device for generating simulation environment - Google Patents
Method and device for generating simulation environment Download PDFInfo
- Publication number
- CN112712595A CN112712595A CN202110069350.3A CN202110069350A CN112712595A CN 112712595 A CN112712595 A CN 112712595A CN 202110069350 A CN202110069350 A CN 202110069350A CN 112712595 A CN112712595 A CN 112712595A
- Authority
- CN
- China
- Prior art keywords
- dimensional
- map
- dimensional simulation
- determining
- data
- 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.)
- Pending
Links
- 238000004088 simulation Methods 0.000 title claims abstract description 228
- 238000000034 method Methods 0.000 title claims abstract description 53
- 230000008447 perception Effects 0.000 claims abstract description 93
- 238000012549 training Methods 0.000 claims abstract description 71
- 238000009826 distribution Methods 0.000 claims abstract description 57
- 238000003860 storage Methods 0.000 claims description 17
- 238000004590 computer program Methods 0.000 claims description 16
- 230000007613 environmental effect Effects 0.000 claims description 13
- 238000005457 optimization Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 abstract description 8
- 238000010586 diagram Methods 0.000 description 19
- 230000008569 process Effects 0.000 description 13
- 230000006870 function Effects 0.000 description 10
- 230000006872 improvement Effects 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 7
- 238000012545 processing Methods 0.000 description 6
- 230000003287 optical effect Effects 0.000 description 3
- 238000013256 Gubra-Amylin NASH model Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 229920001296 polysiloxane Polymers 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The specification discloses a method and a device for generating a simulation environment, wherein a perception blind area in a two-dimensional simulation map is determined according to the attribute of an obstacle in the three-dimensional simulation environment and the pose of unmanned equipment in the three-dimensional simulation environment, and then two-dimensional simulation map data containing the perception blind area are input into a noise simulation model trained in advance to obtain the two-dimensional noise simulation map conforming to preset noise distribution. The two-dimensional noise simulation map contains the knowledge of the sensor blind area and the fitted noise, so that the two-dimensional noise simulation map is more consistent with the two-dimensional map generated based on the sensing data under the real condition, a better environment is provided for training a path planning model, and the model training efficiency and effect are improved.
Description
Technical Field
The present disclosure relates to the field of computer technologies, and in particular, to a method and an apparatus for generating a simulation environment.
Background
At present, the unmanned technology is mature, and in the unmanned technology, path planning based on sensed data is a very important ring. Generally, in order to improve efficiency, a path planning model is usually trained in a simulation environment when being trained, and therefore it is important to determine whether the simulation environment is real enough.
In the prior art, a simulation environment adopted for path planning is usually a three-dimensional simulation environment, and a two-dimensional simulation map including boundaries of obstacles is extracted, and a path planning model is trained by using the simulation map.
However, the two-dimensional simulation map extracted from the three-dimensional simulation environment has the problem of' view angle of the god, so that the path planning model acquires too much accurate information, the simulation environment is greatly different from the actual environment, and the model training effect is poor.
Disclosure of Invention
The embodiments of the present specification provide a method and an apparatus for generating a simulation environment, so as to partially solve the problems in the prior art.
The embodiment of the specification adopts the following technical scheme:
the present specification provides a method for generating a simulation environment, including:
acquiring three-dimensional simulation environment data, wherein the three-dimensional simulation environment data at least comprises attributes of each obstacle, and the attributes at least comprise a pose and a shape;
determining two-dimensional simulation map data corresponding to the three-dimensional simulation environment data, wherein the two-dimensional simulation map data comprise boundary information of each obstacle;
when a path planning model is trained according to the two-dimensional simulation map data, determining a perception blind area in the two-dimensional simulation map according to the pose of unmanned equipment in the three-dimensional simulation environment and the attribute of each obstacle in the three-dimensional simulation environment data;
and determining a first perception map according to the two-dimensional simulation map data and the perception blind area, inputting a pre-trained noise simulation model, determining a two-dimensional simulation map which accords with preset noise distribution as a second perception map, and planning a path by the path planning model based on the second perception map.
Optionally, determining a perception blind area in the two-dimensional simulation map according to the pose of the unmanned equipment in the three-dimensional simulation environment and the attributes of the obstacles in the three-dimensional simulation environment data, specifically including:
determining the pose of a sensor of the unmanned device according to the pose of the unmanned device in the three-dimensional simulation environment;
determining the pose and the shape of each obstacle according to the attribute of each obstacle in the three-dimensional simulation environment data;
and determining a perception blind area of the sensor in the two-dimensional simulation map according to the pose of the sensor, the pose and the shape of each obstacle and a preset physical shielding rule.
Optionally, determining a perception blind area of the sensor in the two-dimensional simulation map according to the pose of the sensor, the pose of each obstacle, and the shape of each obstacle and a preset physical occlusion rule, specifically including:
inputting the pose of the sensor, the attribute of each obstacle and the three-dimensional simulation environment data as input, and inputting a pre-trained blind area determination model to obtain a perception blind area of the sensor in the two-dimensional simulation map;
the blind area determination model is obtained by training through the following method:
determining a plurality of three-dimensional simulation environment data containing obstacles and sensors as training samples;
for each training sample, determining an environmental point cloud sensed by the sensor according to the pose of the sensor and the three-dimensional model of each obstacle in the three-dimensional simulation environmental data of the training sample;
determining a perception blind area as a mark of the training sample according to the projection of the environmental point cloud on the map and a two-dimensional simulation map corresponding to the three-dimensional simulation environment of the training sample;
and training the blind area determination model according to the determined training samples and the labels of the training samples.
Optionally, the noise simulation model is composed of a first network and a second network, the first network is respectively composed of an encoder-decoder, and the second network is composed of a generation sub-network and a discrimination sub-network;
the noise simulation model was trained using the following method:
acquiring original sensing data collected from an actual environment;
determining a two-dimensional perception map according to the original perception data;
inputting the two-dimensional perception map into a coding layer of the first network to obtain coded data;
inputting the coded data into a decoding layer of the first network, and taking an obtained decoding result as a hidden vector;
inputting the hidden vector into a generation sub-network of the second network to obtain a generated two-dimensional perception map containing noise;
inputting the two-dimensional perception map containing the noise into an authentication sub-network of the second network to obtain an output authentication result of the authentication network layer;
and determining loss according to at least one of the regular loss term of the coded data distribution, the distribution loss term of the hidden variable and the loss term of the identification result, and training the noise simulation model by taking the minimum loss as an optimization target.
Optionally, determining the regular loss term of the encoded data distribution specifically includes:
and determining information entropy as a regular loss term of the distribution of the coded data according to the prior distribution of the coded data.
Optionally, determining a distribution loss term of the hidden variable specifically includes:
for each layer of the identification subnetwork, determining a first identification result of the two-dimensional perception map output by the layer after the two-dimensional perception map is input into the identification subnetwork;
determining a second identification result of the two-dimensional perception map containing the noise output by the layer according to the two-dimensional perception map containing the noise generated after the generation sub-network restores the hidden variable;
determining a distribution loss item of the hidden variable according to the distribution difference between the first identification result and the second identification result output by each layer;
and the distribution of the first identification result and the second identification result conforms to a preset normal distribution graph.
Optionally, determining the loss term of the authentication result specifically includes:
and determining a loss term of the identification result according to the identification result of the two-dimensional perception map output by the identification subnetwork, the identification result of the two-dimensional perception map containing noise output by the identification subnetwork and the identification result of the hidden variable output by the identification subnetwork.
The present specification provides an apparatus for generating a simulation environment, including:
the system comprises an acquisition module, a display module and a control module, wherein the acquisition module acquires three-dimensional simulation environment data, the three-dimensional simulation environment data at least comprises attributes of each obstacle, and the attributes at least comprise a pose and a shape;
the determining module is used for determining two-dimensional simulation map data corresponding to the three-dimensional simulation environment data, wherein the two-dimensional simulation map data comprise boundary information of each obstacle;
the dead zone generating module is used for determining a perception dead zone in the two-dimensional simulation map according to the pose of the unmanned equipment in the three-dimensional simulation environment and the attribute of each obstacle in the three-dimensional simulation environment data when a path planning model is trained according to the two-dimensional simulation map data;
and the noise generation module is used for determining a first perception map according to the two-dimensional simulation map data and the perception blind area, inputting a pre-trained noise simulation model, determining a two-dimensional simulation map which accords with preset noise distribution as a second perception map, and planning a path by the path planning model based on the second perception map.
The present specification provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the above-described method of generating a simulation environment.
The present specification provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the method for generating a simulation environment.
The embodiment of the specification adopts at least one technical scheme which can achieve the following beneficial effects:
when a simulation environment for planning a path is generated, a perception blind area in a two-dimensional simulation map is determined according to the attribute of an obstacle in the three-dimensional simulation environment and the pose of unmanned equipment in the three-dimensional simulation environment, and then two-dimensional simulation map data containing the perception blind area are input into a noise simulation model trained in advance to obtain the two-dimensional noise simulation map conforming to preset noise distribution. The two-dimensional noise simulation map contains the knowledge of the sensor blind area and the fitted noise, so that the two-dimensional noise simulation map is more consistent with the two-dimensional map generated based on the sensing data under the real condition, a better environment is provided for training a path planning model, and the model training efficiency and effect are improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the specification and are incorporated in and constitute a part of this specification, illustrate embodiments of the specification and together with the description serve to explain the specification and not to limit the specification in a non-limiting sense. In the drawings:
FIG. 1 is a schematic flow chart of generating a simulation environment according to an embodiment of the present disclosure;
FIG. 2 is a schematic illustration of a three-dimensional simulation environment provided herein;
FIG. 3 is a schematic illustration of two-dimensional simulated map data provided herein;
FIGS. 4 a-4 c are schematic diagrams of the dead zones of perception provided herein;
FIG. 5 is a schematic flow chart of a training blind area determination model provided in an embodiment of the present disclosure;
FIG. 6 is a schematic structural diagram of an apparatus for generating a simulation environment according to an embodiment of the present disclosure;
fig. 7 is a schematic structural diagram of an electronic device provided in an embodiment of this specification.
Detailed Description
In order to make the objects, technical solutions and advantages of the present disclosure more clear, 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 is to be understood that the embodiments described are only a few embodiments of the present disclosure, and not all embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present specification without any creative effort belong to the protection scope of the present specification.
The technical solutions provided by the embodiments of the present description are described in detail below with reference to the accompanying drawings.
Fig. 1 is a schematic flowchart of generating a simulation environment according to an embodiment of the present disclosure, including:
s100: acquiring three-dimensional simulation environment data, wherein the three-dimensional simulation environment data at least comprises attributes of each obstacle, and the attributes at least comprise a pose and a shape.
In one or more embodiments of the present description, since the simulation environment is mainly used for training models offline, that is, training models in a non-real environment, the simulation environment is generally generated by a server according to needs. Similarly, in this specification, the process of generating the simulation environment may also be performed by the server. Of course, the server is a single device or a system composed of multiple devices, and this specification is not limited thereto.
Specifically, the server may first determine three-dimensional simulation environment data generated by modeling or the like, and of course, the three-dimensional simulation environment data at least needs to include data of an obstacle. In addition, since the simulation environment generated by the present specification can be used for training the path planning model, the three-dimensional simulation environment data may further include road information, as shown in fig. 2.
Fig. 2 is a schematic diagram of a three-dimensional simulation environment provided in the present specification, which is seen to include a road and several obstacles on the road. The obstacles may be static obstacles, such as street lamps, signal lamps, barricades, etc., or dynamic obstacles, such as vehicles, pedestrians, etc.
In addition, since the sensing blind area of the sensor is related to the pose of the sensor and the position and shape of the obstacle, the three-dimensional simulation environment data may further include the attribute of each obstacle, which includes at least the pose and shape of the target object. So that subsequently, based on the three-dimensional simulated environmental data, a dead zone of perception that should occur when the sensor is at a different location can be determined.
Further, in this specification, the three-dimensional simulation environment data is used for modeling in a three-dimensional space, resulting in a three-dimensional simulation environment, i.e., the three-dimensional simulation environment illustrated in fig. 2. Therefore, in order to make the generated simulation environment more realistic, the three-dimensional simulation environment data can be acquired by collecting the actual scene. For example, point clouds of several different types of roads are collected in advance, point clouds of several different types of obstacles are collected, and a three-dimensional model of the roads and the obstacles is generated based on the collected point clouds. Then, when the three-dimensional simulation environment data is needed subsequently, a plurality of three-dimensional environment data can be obtained by combining various three-dimensional models in a random combination or manual setting mode.
Of course, the three-dimensional simulation environment data may also be created in a three-dimensional space, and how to obtain the specific three-dimensional simulation environment data may be set according to needs, which is not limited in this specification.
S102: and determining two-dimensional simulation map data corresponding to the three-dimensional simulation environment data, wherein the two-dimensional simulation map data comprises boundary information of each obstacle.
In one or more embodiments of the present disclosure, since the path planning does not require three-dimensional data, but only requires planar two-dimensional map data, the path planning is usually performed based on the two-dimensional data when the path planning is performed. Thus, in order to conform to the actual application scenario of the path planning model, in one or more embodiments of the present specification, the server may further determine corresponding two-dimensional simulation map data after acquiring the three-dimensional simulation environment data. Since the three-dimensional simulation environment data includes the attribute of the obstacle, the two-dimensional simulation map data includes boundary information of each obstacle when the two-dimensional simulation map data is obtained.
Specifically, the server may perform modeling in a three-dimensional space according to the three-dimensional simulation environment data, and then determine the projection of each obstacle in the three-dimensional space on the ground, to obtain two-dimensional simulation map data including the boundary of each obstacle, as shown in fig. 3.
Fig. 3 is a schematic diagram of two-dimensional simulated map data provided in the present specification, in which a road boundary and a boundary of each obstacle are visible.
At this time, the server still obtains a map having an "upper ground view angle", but at this time, since the position of the virtual unmanned vehicle entering the simulation environment is not determined, information such as a blind sensing area of the sensor cannot be determined.
S104: and when a path planning model is trained according to the two-dimensional simulation map data, determining a perception blind area in the two-dimensional simulation map according to the pose of the unmanned equipment in the three-dimensional simulation environment and the attributes of all obstacles in the three-dimensional simulation environment data.
In one or more embodiments of the present specification, when a path planning model needs to be trained according to the two-dimensional simulation map data, the server may determine a blind sensing area of a sensor of the unmanned aerial vehicle in the two-dimensional simulation map according to a pose of the unmanned aerial vehicle in the three-dimensional simulation environment and attributes of obstacles in the three-dimensional simulation environment data.
In the prior art, when a path planning model is trained, an initial position of a virtual unmanned device and a destination of the virtual unmanned device are set, and the unmanned device performs data planning by using global information of a two-dimensional simulation map as data sensed by a sensor.
Therefore, generally, when path planning is performed, only a two-dimensional simulation map is needed, but in the embodiment provided in this specification, since a sensing blind area of the sensor needs to be simulated, and the sensing blind area described in step S100 is determined by attributes of the obstacle and a pose of the sensor, when path planning is performed in this specification, it is first necessary to determine and also determine a pose of the unmanned aerial vehicle in a three-dimensional simulation environment corresponding to the two-dimensional simulation map. When the path planning training is started, the pose is determined according to the set initial position and orientation of the unmanned vehicle, and when the subsequent unmanned equipment runs according to the planned path, the pose of the unmanned equipment needs to be determined again at different moments.
Simply speaking, as the unmanned device continuously travels along the planned path, the perceived obstacles are not identical, and the generated blind areas are not identical, and it is necessary to re-determine the perceived blind area at each different time according to the pose of the unmanned device at that time.
Secondly, in the three-dimensional simulation environment, the pose of the sensor of the unmanned equipment is determined according to the pose of the unmanned equipment in the three-dimensional simulation environment.
And finally, determining a perception blind area of the sensor in the two-dimensional simulation map according to the pose of the sensor, the pose and the shape of each obstacle and a preset physical shielding rule.
Specifically, the physical occlusion rule may be set empirically, for example, assuming that the height of a general sensor is 2m, the area and the shape of the blind area are determined according to the height of the obstacle and the vertical width in the sensor collecting direction, wherein the larger the difference between the height of the obstacle and the height of the sensor is, the larger the area of the blind area is, and the larger the vertical width of the obstacle in the sensor collecting direction is, the larger the area of the blind area is. The shape of the blind area may be preset, such as an inverted trapezoid, a rectangle, or the like.
Fig. 4a is a schematic diagram of the blind sensing areas provided in the present specification, and fig. 4a is determined based on the two-dimensional simulation map of fig. 3, in which the poses of the sensors are visible and the respective blind sensing areas are determined.
In addition, in this specification, the blind sensing area may be determined according to a preset sensing range, in addition to being generated by occlusion. That is, the boundary of the obstacle sensed by the sensor is determined according to the preset radius by taking the sensor as the center. Of course, the direction sensed by the sensor can be preset, and the sensing blind area can be further enlarged. Of course, the sensing blind area generated by the occlusion and the sensing blind area generated by the sensing range can be used in combination to determine a more precise sensing blind area.
Fig. 4b is a schematic diagram of the sensing blind area provided in this specification, and it can be seen that the boundary of the obstacle within the preset radius with the sensor as the center is determined, and the rest of the map is used as the sensing blind area. Fig. 4c further increases the limit on the perceived orientation. Of course, the above-mentioned multiple ways of determining the blind perception regions may be used in combination, that is, according to multiple physical occlusion rules, a comprehensive blind perception region is determined.
S106: inputting a pre-trained noise simulation model according to the two-dimensional simulation map data and the perception blind area, and determining a two-dimensional noise simulation map which accords with preset noise distribution, wherein the path planning model carries out path planning based on the two-dimensional noise simulation map.
In this specification, since in an actual scene, data sensed by the sensor always has errors and interference, that is, noise, the server may further learn the noise to add the calculated noise to the blind sensing area obtained in step S104, so that the generated simulated environment data is more accurate.
Typically, the sources of noise effects are multiple, including, for example, sensor errors (e.g., one or more sensors used by an unmanned device such as a camera, radar, speedometer, accelerometer, gyroscope, etc. to sense the environment, etc.), system errors, external parameter effects (e.g., the profile of the unmanned device, sensor orientation, external lighting, etc.), vehicle jitter (e.g., jitter of the unmanned device resulting from effects of speed, operating power, stability, etc.).
Specifically, in the present specification, the noise simulation model may be composed of a first network composed of an encoder-decoder, and a second network composed of a generation sub-network-discrimination sub-network. In particular, the first network may be a Variational Automatic Encoder (VAE), and the second network may be a Generative Adaptive Network (GAN).
In this specification, the noise simulation model is trained in the following manner:
first, raw perceptual data acquired from an actual environment is acquired.
In this specification, in order that the trained noise simulation model may output noise similar to that generated by actual acquisition, the server may acquire raw sensing data acquired by a sensor from an actual environment, wherein the sensor may be of the same type as a sensor employed by a subsequent unmanned device.
Since the noise of different types of sensors may have distribution differences, in order to make the noise simulation model more effective, the server may also train a noise simulation model corresponding to each type of sensor. For convenience of description, the noise simulation model is only described as an example of training, and the noise simulation models corresponding to multiple types of sensors may be trained according to the same process.
Secondly, determining a two-dimensional perception map according to the original perception data.
The server may determine a two-dimensional perception map corresponding to the original perception data as a training sample in the same manner as in step S102.
And thirdly, inputting the two-dimensional perception map into the coding layer of the first network to obtain coded data.
Then, the encoded data is input to a decoding layer of the first network, and an obtained decoding result is used as a hidden vector.
The two steps are processes of inputting a two-dimensional perception map as a training sample into an encoder and a decoder of a VAE model to obtain an output result, but in the present specification, the VAE model and a GAN model are combined together, so the training process is also a joint training, and the output result of the decoder is not the output of a conventional VAE model, but is data of a GAN model generation network generation perception map.
And then, inputting the hidden vector into a generation sub-network of the second network to obtain a generated two-dimensional perception map containing noise.
And then, inputting the two-dimensional perception map containing the noise into an authentication sub-network of the second network to obtain an output authentication result of the authentication network layer.
The two steps are processes of inputting the hidden variables output by the VAE model decoder into the generation sub-network and the identification sub-network respectively to obtain identification results.
Finally, determining loss according to at least one of the regular loss term of the coded data distribution, the distribution loss term of the hidden variable and the loss term of the identification result, and training the noise simulation model by taking the minimum loss as an optimization target
In this specification, since the joint training is performed, the server may determine the loss according to at least one of a regular loss term of the encoded data distribution, a distribution loss term of the hidden variable, and a loss term of the discrimination result. The loss formula is specifically expressed as:
wherein L ispriorRepresenting the number of codesAccording to the regular loss term of the distribution,
distribution loss term, L, representing hidden variablesGANA loss term representing the result of the authentication.
Wherein L ispriorDenoted L as regular term in the conventional VAE modelprior＝DKL(q (z | x) | p (z)) wherein DKLIndicating KL divergence, x indicating the input two-dimensional perceptual map, z to enc (x) q (z | x), i.e., the result of the encoder encoding the two-dimensional perceptual map, and therefore LpriorIs the distribution of the coded data obtained by coding the two-dimensional perception map according to the coding layer, which is taken as the posterior distribution, and the information entropy (i.e. KL divergence) determined according to the prior distribution of the coded data.
In particular, the method comprises the following steps of,
wherein, p (Dis)l(x) Lz) constraint condition that indicates that the distribution conforms to the normal distribution diagram
Therein, Disl(x) And
respectively representing the l-th layer of the identifying subnetwork, the identifying result of the two-dimensional perception map and the identifying result of the hidden variable, Disl(x)∈[0，1]That is to say, normal scoreThe mean and the difference of the layout are determined according to the two discrimination results, respectively. That is to say, for each layer of the identifying subnetwork, after the two-dimensional perception map is input into the identifying subnetwork, a first identifying result of the two-dimensional perception map output by the layer is determined, a second identifying result of the two-dimensional perception map output by the layer and containing noise is determined according to a two-dimensional perception map containing noise generated after the hidden variable is restored by the generating subnetwork, and a distribution loss term of the hidden variable is determined according to a distribution difference between the first identifying result and the second identifying result output by each layer.
LGANThat is, the loss of the conventional GAN network, can be represented by the formula LGAN＝log(Disl(x))+log(1-Disl(Gen (z)) representation, where Gen (z) is a two-dimensional perceptual map comprising noise, Dis, that generates the output of a subnetworkl(x) In order to identify the discrimination result of the sub-network to the two-dimensional perception map.
Further, in this specification, to increase the learning effect of the loss of the GAN network on the VAE model, the LGANVAE loss terms can also be added, and the formula can be LGAN＝log(Disl(x))+log(1-Disl(Gen(z)))+log(1-Disl(Dec(Enc(x))))，Disl(Dec (enc (x))) represents the result of the discrimination of the hidden variables output by the decoder. That is, the server determines the loss term of the discrimination result according to the discrimination result of the two-dimensional perception map output by the discrimination sub-network, the discrimination result of the two-dimensional perception map containing noise output by the discrimination sub-network, and the discrimination result of the hidden variable output by the discrimination sub-network.
Through the process of generating the simulation environment shown in fig. 1, a perception blind area in the two-dimensional simulation map is determined according to the attribute of the obstacle in the three-dimensional simulation environment and the pose of the unmanned equipment in the three-dimensional simulation environment, and then the two-dimensional simulation map data including the perception blind area is input into a pre-trained noise simulation model, so that the two-dimensional noise simulation map conforming to the preset noise distribution is obtained. The two-dimensional noise simulation map contains the knowledge of the sensor blind area and the fitted noise, so that the two-dimensional noise simulation map is more consistent with the two-dimensional map generated based on the sensing data under the real condition, a better environment is provided for training a path planning model, and the model training efficiency and effect are improved.
Further, in this specification, the server may also determine the perception blind area of the sensor in the two-dimensional simulation map through a pre-trained blind area determination model. Specifically, the server can input the pose of the sensor, the pose of each obstacle and the shape of each obstacle as input, and input the pre-trained dead zone determination model to obtain a two-dimensional simulation map containing a perception dead zone output by the dead zone determination model.
Wherein, the blind area determination model is obtained by training according to the following method, as shown in fig. 5:
s200: and determining a plurality of three-dimensional simulation environment data containing the obstacles and the sensors as training samples.
The server can obtain a plurality of three-dimensional simulation environment data containing obstacles and sensors as training samples by adding unmanned equipment at different positions of the three-dimensional simulation environment data.
S202: and aiming at each training sample, determining the environmental point cloud sensed by the sensor according to the pose of the sensor and the three-dimensional model of each obstacle in the three-dimensional simulation environmental data of the training sample.
In this specification, after the server determines the training samples, it needs to determine labels of the training samples, so the server can perform data modeling for each training sample according to three-dimensional simulation environment data to obtain a three-dimensional space including a three-dimensional model of each obstacle and a sensor pose, and then determine an environmental point cloud sensed by the sensor according to the pose of the sensor and the three-dimensional model of each obstacle.
S204, determining a perception blind area as a mark of the training sample according to the projection of the environmental point cloud on the map and the two-dimensional simulation map corresponding to the three-dimensional simulation environment of the training sample.
In this specification, for each training sample, the server may determine a projection of the environmental point cloud on the ground as a boundary of an obstacle in the two-dimensional simulation map according to determining the environmental point cloud perceived by the sensor. And then determining a perception blind area as a mark of the training sample according to a two-dimensional simulation map corresponding to the three-dimensional simulation environment of the training sample and the determined boundary of the perceivable obstacle.
S206: and training the blind area determination model according to the determined training samples and the labels of the training samples.
In this specification, the aforementioned steps S200 to S206 correspond to a stage of generating training samples, and after the obtained training samples are generated, the server can train the blind area determination model according to the training samples and labels thereof. The blind area determination model may be a neural network model, and the specification does not limit the specific model.
Based on the same idea, the present specification further provides a corresponding apparatus, a storage medium, and an electronic device.
Fig. 6 is a schematic structural diagram of a simulation environment generation apparatus provided in an embodiment of the present specification, where the apparatus includes:
the acquiring module 300 acquires three-dimensional simulation environment data, wherein the three-dimensional simulation environment data at least comprises attributes of each obstacle, and the attributes at least comprise a pose and a shape;
a determining module 302, configured to determine two-dimensional simulation map data corresponding to the three-dimensional simulation environment data, where the two-dimensional simulation map data includes boundary information of each obstacle;
a blind area generation module 304, which determines a perception blind area in the two-dimensional simulation map according to the pose of the unmanned equipment in the three-dimensional simulation environment and the attribute of each obstacle in the three-dimensional simulation environment data when a path planning model is trained according to the two-dimensional simulation map data;
the noise generation module 306 determines a first perception map according to the two-dimensional simulation map data and the perception blind area, inputs a pre-trained noise simulation model, determines a two-dimensional simulation map conforming to preset noise distribution as a second perception map, and performs path planning based on the second perception map by the path planning model.
Optionally, the blind area generating module 304 determines a pose of a sensor of the unmanned device according to the pose of the unmanned device in the three-dimensional simulation environment, determines a pose and a shape of each obstacle according to an attribute of each obstacle in the three-dimensional simulation environment data, and determines a sensing blind area of the sensor in the two-dimensional simulation map according to a preset physical occlusion rule according to the pose of the sensor, the pose of each obstacle, and the shape of each obstacle.
Optionally, the blind area generating module 304 inputs the pose of the sensor, the attribute of each obstacle, and the three-dimensional simulation environment data as inputs to a pre-trained blind area determination model, so as to obtain the sensing blind area of the sensor in the two-dimensional simulation map.
The device further comprises:
the training module 308 determines a plurality of three-dimensional simulation environment data including obstacles and sensors as training samples, determines, for each training sample, an environment point cloud sensed by the sensor in the three-dimensional simulation environment data of the training sample according to the pose of the sensor and the three-dimensional model of each obstacle, determines a sensing blind area as a label of the training sample according to the projection of the environment point cloud on a map and a two-dimensional simulation map corresponding to the three-dimensional simulation environment of the training sample, and trains the blind area determination model according to the determined labels of each training sample and each training sample.
Optionally, the noise simulation model is composed of a first network and a second network, the first network is respectively composed of an encoder-decoder, the second network is composed of a generation sub-network and a discrimination sub-network,
the training module 308 acquires original sensing data collected from an actual environment, determines a two-dimensional sensing map according to the original sensing data, inputs the two-dimensional sensing map into a coding layer of the first network to obtain coded data, inputs the coded data into a decoding layer of the first network, uses an obtained decoding result as a hidden vector, inputs the hidden vector into a generation sub-network of the second network to obtain a generated two-dimensional sensing map containing noise, inputs the two-dimensional sensing map containing noise into a discrimination sub-network of the second network to obtain an output discrimination result of the discrimination network layer, determines a loss according to at least one of a loss term of the coded data distribution, a distribution loss term of the hidden variable and a loss term of the discrimination result, and takes a loss minimum as an optimization target, and training the noise simulation model.
Optionally, the training module 308 determines an information entropy as a regular loss term of the encoded data distribution according to the distribution of the encoded data obtained by encoding the two-dimensional sensing map by the encoding layer, which is used as a posterior distribution, and according to the prior distribution of the encoded data.
Optionally, the training module 308 determines, for each layer of the identification subnetwork, a first identification result of the two-dimensional perception map output by the layer after the two-dimensional perception map is input into the identification subnetwork, determines, according to the two-dimensional perception map including noise generated after the generation subnetwork restores the hidden variable, a second identification result of the two-dimensional perception map including noise output by the layer, and determines a distribution loss term of the hidden variable according to a distribution difference between the first identification result and the second identification result output by each layer, where distributions of the first identification result and the second identification result both conform to a preset normal distribution diagram
Optionally, the training module 308 determines a loss term of the discrimination result according to the discrimination result of the two-dimensional perception map output by the discrimination sub-network, the discrimination result of the two-dimensional perception map including noise output by the discrimination sub-network, and the discrimination result of the hidden variable output by the discrimination sub-network.
The present specification also provides a computer readable storage medium storing a computer program which, when executed by a processor, is operable to perform the method of generating a simulation environment provided above.
Based on the method for generating a simulation environment provided above, an embodiment of the present specification further provides a schematic structural diagram of the electronic device shown in fig. 7. As shown in fig. 7, at the hardware level, the drone includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, although it may also include hardware required for other services. The processor reads the corresponding computer program from the nonvolatile memory into the memory and then runs the computer program to realize the method for generating the simulation environment.
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 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.
The description has been presented with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the description. 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 (10)
1. A method of generating a simulation environment, comprising:
acquiring three-dimensional simulation environment data, wherein the three-dimensional simulation environment data at least comprises attributes of each obstacle, and the attributes at least comprise a pose and a shape;
determining two-dimensional simulation map data corresponding to the three-dimensional simulation environment data, wherein the two-dimensional simulation map data comprise boundary information of each obstacle;
when a path planning model is trained according to the two-dimensional simulation map data, determining a perception blind area in the two-dimensional simulation map according to the pose of unmanned equipment in the three-dimensional simulation environment and the attribute of each obstacle in the three-dimensional simulation environment data;
inputting a pre-trained noise simulation model according to the two-dimensional simulation map data and the perception blind area, and determining a two-dimensional noise simulation map which accords with preset noise distribution, wherein the path planning model carries out path planning based on the two-dimensional noise simulation map.
2. The method of claim 1, wherein determining a perceived blind area in the two-dimensional simulation map according to the pose of the unmanned aerial vehicle in the three-dimensional simulation environment and the attributes of the obstacles in the three-dimensional simulation environment data specifically comprises:
determining the pose of a sensor of the unmanned device according to the pose of the unmanned device in the three-dimensional simulation environment;
determining the pose and the shape of each obstacle according to the attribute of each obstacle in the three-dimensional simulation environment data;
and determining a perception blind area of the sensor in the two-dimensional simulation map according to the pose of the sensor, the pose and the shape of each obstacle and a preset physical shielding rule.
3. The method of claim 2, wherein determining a blind sensing area of the sensor in the two-dimensional simulation map according to the pose of the sensor, the pose of each obstacle, and the shape of each obstacle and according to a preset physical occlusion rule specifically comprises:
inputting the pose of the sensor, the attribute of each obstacle and the three-dimensional simulation environment data as input, and inputting a pre-trained blind area determination model to obtain a perception blind area of the sensor in the two-dimensional simulation map;
the blind area determination model is obtained by training through the following method:
determining a plurality of three-dimensional simulation environment data containing obstacles and sensors as training samples;
for each training sample, determining an environmental point cloud sensed by the sensor according to the pose of the sensor and the three-dimensional model of each obstacle in the three-dimensional simulation environmental data of the training sample;
determining a perception blind area as a mark of the training sample according to the projection of the environmental point cloud on the map and a two-dimensional simulation map corresponding to the three-dimensional simulation environment of the training sample;
and training the blind area determination model according to the determined training samples and the labels of the training samples.
4. The method of claim 1, wherein the noise simulation model is composed of a first network and a second network, the first network being respectively composed of an encoder-decoder, the second network being composed of a generator-discriminator subnetwork;
the noise simulation model was trained using the following method:
acquiring original sensing data collected from an actual environment;
determining a two-dimensional perception map according to the original perception data;
inputting the two-dimensional perception map into a coding layer of the first network to obtain coded data;
inputting the coded data into a decoding layer of the first network, and taking an obtained decoding result as a hidden vector;
inputting the hidden vector into a generation sub-network of the second network to obtain a generated two-dimensional perception map containing noise;
inputting the two-dimensional perception map containing the noise into an authentication sub-network of the second network to obtain an output authentication result of the authentication network layer;
and determining loss according to at least one of the regular loss term of the coded data distribution, the distribution loss term of the hidden variable and the loss term of the identification result, and training the noise simulation model by taking the minimum loss as an optimization target.
5. The method according to claim 4, wherein determining the canonical loss term for the encoded data distribution specifically comprises:
and determining information entropy as a regular loss term of the distribution of the coded data according to the prior distribution of the coded data.
6. The method of claim 4, wherein determining the distribution loss term for the hidden variable specifically comprises:
for each layer of the identification subnetwork, determining a first identification result of the two-dimensional perception map output by the layer after the two-dimensional perception map is input into the identification subnetwork;
determining a second identification result of the two-dimensional perception map containing the noise output by the layer according to the two-dimensional perception map containing the noise generated after the generation sub-network restores the hidden variable;
determining a distribution loss item of the hidden variable according to the distribution difference between the first identification result and the second identification result output by each layer;
and the distribution of the first identification result and the second identification result conforms to a preset normal distribution graph.
7. The method of claim 4, wherein determining the loss term for the authentication result comprises:
and determining a loss term of the identification result according to the identification result of the two-dimensional perception map output by the identification subnetwork, the identification result of the two-dimensional perception map containing noise output by the identification subnetwork and the identification result of the hidden variable output by the identification subnetwork.
8. An apparatus for generating a simulation environment, comprising:
the system comprises an acquisition module, a display module and a control module, wherein the acquisition module acquires three-dimensional simulation environment data, the three-dimensional simulation environment data at least comprises attributes of each obstacle, and the attributes at least comprise a pose and a shape;
the determining module is used for determining two-dimensional simulation map data corresponding to the three-dimensional simulation environment data, wherein the two-dimensional simulation map data comprise boundary information of each obstacle;
the dead zone generating module is used for determining a perception dead zone in the two-dimensional simulation map according to the pose of the unmanned equipment in the three-dimensional simulation environment and the attribute of each obstacle in the three-dimensional simulation environment data when a path planning model is trained according to the two-dimensional simulation map data;
and the noise generation module is used for determining a first perception map according to the two-dimensional simulation map data and the perception blind area, inputting a pre-trained noise simulation model, determining a two-dimensional simulation map which accords with preset noise distribution as a second perception map, and planning a path by the path planning model based on the second perception map.
9. 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 1-7.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any of claims 1-7 when executing the program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110069350.3A CN112712595A (en) | 2021-01-19 | 2021-01-19 | Method and device for generating simulation environment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110069350.3A CN112712595A (en) | 2021-01-19 | 2021-01-19 | Method and device for generating simulation environment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112712595A true CN112712595A (en) | 2021-04-27 |
Family
ID=75549373
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110069350.3A Pending CN112712595A (en) | 2021-01-19 | 2021-01-19 | Method and device for generating simulation environment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112712595A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117807782A (en) * | 2023-12-29 | 2024-04-02 | 南京仁高隆软件科技有限公司 | Method for realizing three-dimensional simulation model |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414098A (en) * | 2019-07-12 | 2019-11-05 | 北京三快在线科技有限公司 | A kind of generation method and device of Simulation Test Environment |
| US20200033866A1 (en) * | 2018-07-24 | 2020-01-30 | Pony.ai, Inc. | Generative adversarial network enriched driving simulation |
| CN111161424A (en) * | 2019-12-30 | 2020-05-15 | 浙江欣奕华智能科技有限公司 | Three-dimensional map determination method and determination device |
| CN111578932A (en) * | 2020-05-28 | 2020-08-25 | 长沙中联重科环境产业有限公司 | Image establishing method, device, medium and equipment based on multi-line laser radar |
| CN111753424A (en) * | 2020-06-26 | 2020-10-09 | 北京百度网讯科技有限公司 | Simulation method, simulation device, electronic equipment and storage medium |
| CN112102151A (en) * | 2020-07-27 | 2020-12-18 | 广州视源电子科技股份有限公司 | Grid map generation method and device, mobile intelligent device and storage medium |
-
2021
- 2021-01-19 CN CN202110069350.3A patent/CN112712595A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200033866A1 (en) * | 2018-07-24 | 2020-01-30 | Pony.ai, Inc. | Generative adversarial network enriched driving simulation |
| CN110414098A (en) * | 2019-07-12 | 2019-11-05 | 北京三快在线科技有限公司 | A kind of generation method and device of Simulation Test Environment |
| CN111161424A (en) * | 2019-12-30 | 2020-05-15 | 浙江欣奕华智能科技有限公司 | Three-dimensional map determination method and determination device |
| CN111578932A (en) * | 2020-05-28 | 2020-08-25 | 长沙中联重科环境产业有限公司 | Image establishing method, device, medium and equipment based on multi-line laser radar |
| CN111753424A (en) * | 2020-06-26 | 2020-10-09 | 北京百度网讯科技有限公司 | Simulation method, simulation device, electronic equipment and storage medium |
| CN112102151A (en) * | 2020-07-27 | 2020-12-18 | 广州视源电子科技股份有限公司 | Grid map generation method and device, mobile intelligent device and storage medium |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117807782A (en) * | 2023-12-29 | 2024-04-02 | 南京仁高隆软件科技有限公司 | Method for realizing three-dimensional simulation model |
| CN117807782B (en) * | 2023-12-29 | 2024-06-07 | 南京仁高隆软件科技有限公司 | Method for realizing three-dimensional simulation model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111190427B (en) | Method and device for planning track | |
| CN112766468B (en) | Trajectory prediction method and device, storage medium and electronic equipment | |
| CN111311709B (en) | Method and device for generating high-precision map | |
| CN111508258B (en) | Positioning method and device | |
| CN111076739B (en) | Path planning method and device | |
| CN111208838B (en) | Control method and device of unmanned equipment | |
| CN111062372B (en) | Method and device for predicting obstacle track | |
| CN111238450B (en) | Visual positioning method and device | |
| CN112163475A (en) | Method and device for determining lane line direction | |
| CN110660103B (en) | Unmanned vehicle positioning method and device | |
| CN112327864A (en) | Control method and control device of unmanned equipment | |
| CN111797711A (en) | Model training method and device | |
| CN112649012A (en) | Trajectory planning method, equipment, medium and unmanned equipment | |
| CN110530398B (en) | Method and device for detecting precision of electronic map | |
| CN113674424B (en) | Method and device for drawing electronic map | |
| CN112712595A (en) | Method and device for generating simulation environment | |
| CN114283148A (en) | Road element extraction method and device, storage medium and electronic equipment | |
| CN112818968A (en) | Target object classification method and device | |
| CN113033527A (en) | Scene recognition method and device, storage medium and unmanned equipment | |
| CN114332201A (en) | Model training and target detection method and device | |
| CN112684720A (en) | Simulation test method and device | |
| CN114545940A (en) | Unmanned equipment control method and device and electronic equipment | |
| CN114187355A (en) | Image calibration method and device | |
| CN112859883A (en) | Control method and control device of unmanned equipment | |
| Subramanyam | Chartopolis-A Self Driving Car Test Bed |
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 |