CN109376371A - A kind of Pedestrian Movement Simulation Studies method and system - Google Patents
A kind of Pedestrian Movement Simulation Studies method and system Download PDFInfo
- Publication number
- CN109376371A CN109376371A CN201810982158.1A CN201810982158A CN109376371A CN 109376371 A CN109376371 A CN 109376371A CN 201810982158 A CN201810982158 A CN 201810982158A CN 109376371 A CN109376371 A CN 109376371A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- variable
- movement
- sample
- model
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of Pedestrian Movement Simulation Studies method and systems, it include: to pass through actual measurement, obtain the displacement variable and information Perception variable of sample pedestrian, displacement variable includes the current kinetic speed and historical movement speed of destination locations, sample pedestrian, and information Perception variable includes the sensing region range of sample pedestrian, number in sensing region;Using displacement variable and information Perception variable as training data, training machine learning model obtains pedestrian movement's model;Initial motion state and the target position of pedestrian to be emulated are obtained, pedestrian movement's model predicts the motor imagination of pedestrian to be emulated, the simulation result as pedestrian to be emulated according to the displacement variable and information Perception variable of pedestrian to be emulated.Thus the present invention can get true, natural Pedestrian Movement Simulation Studies result.
Description
Technical field
The present invention relates to groups to emulate field, and the Pedestrian Movement Simulation Studies method of in particular to a kind of data-driven and is
System.
Background technique
The considerations of for human cost, time cost, convenience, controllability, safety and other various objective condition,
The crowd in real world is emulated using the crowd movement that computer generates, is a huge demand, in video display
Animation, Entertainment, supplemental training and scheme evaluation, building space design and the fields such as safety rehearsal, augmented reality have extensively
General application.Crowd movement's emulation mode of existing Most models driving is due to based on more simplifying it is assumed that therefore generating
Avoid-obstacle behavior it is excessively stiff, do not meet the complexity and diversity of pedestrian's avoid-obstacle behavior in real world.Meanwhile existing base
It learns motor pattern from instance data in instance method, is generated using path segment existing in instance database
Simulation result considers that the factor is excessively single, and modeling excessively simplifies, the avoid-obstacle behavior of generation when facing complex scene not enough from
So.
Present invention aim to address the model rule of the above-mentioned prior art excessively complexity, Consideration are excessively single, real
The bad disadvantage of existing effect, proposes a kind of Pedestrian Movement Simulation Studies method of novel data-driven, realizing has avoidance true to nature
The pedestrian simulation motion modeling of behavior.
Summary of the invention
In order to solve the above-mentioned technical problem, it is an object of that present invention to provide a kind of Pedestrian Movement Simulation Studies sides of data-driven
Method, realizing has the pedestrian simulation motion modeling of avoid-obstacle behavior true to nature.
Specifically, the invention discloses a kind of Pedestrian Movement Simulation Studies method, including:
Step 1 passes through actual measurement, the displacement variable and information Perception variable of acquisition sample pedestrian, displacement change
Amount includes the current kinetic speed and historical movement speed of destination locations, sample pedestrian, which includes sample
Number in the sensing region range of pedestrian, the sensing region;
Step 2, using the displacement variable and the information Perception variable as training data, training machine learning model,
Obtain pedestrian movement's model;
Step 3, the initial motion state for obtaining pedestrian to be emulated and target position, pedestrian movement's model is according to this to imitative
The displacement variable and information Perception variable of true pedestrian, predicts the motor imagination of the pedestrian to be emulated, waits emulating row as this
The simulation result of people.
The Pedestrian Movement Simulation Studies method, wherein the displacement variable further include: the Euclidean distance of current pedestrian and target,
The current targeted direction of pedestrian, current pedestrian's directional velocity and the deviation for arriving target direction.
The Pedestrian Movement Simulation Studies method, wherein the information Perception variable includes: to incite somebody to action using sample pedestrian as the center of circle in step 1
The sensing region range is divided into multiple fan-shaped subregions, counts number and the speed of subregion one skilled in the art point in the subregion
Cloth.
The Pedestrian Movement Simulation Studies method, the detailed process that wherein step 2 is trained are as follows: sometime put with sample pedestrian
Displacement variable and information Perception variable as input, the displacement variable of this pedestrian of latter time point sample as target,
Training obtains pedestrian movement's model.
The Pedestrian Movement Simulation Studies method, wherein the machine learning model is decision-tree model in step 2.
The invention also discloses a kind of Pedestrian Movement Simulation Studies system, including:
Measurement module, for obtaining the displacement variable and information Perception variable of sample pedestrian, the displacement variable
Current kinetic speed and historical movement speed including destination locations, sample pedestrian, which includes sample row
Number in the sensing region range of people, the sensing region;
Training module, for using the displacement variable and the information Perception variable as training data, training machine
Model is practised, pedestrian movement's model is obtained;
Emulation module, for obtaining initial motion state and the target position of pedestrian to be emulated, pedestrian movement's model root
According to the displacement variable and information Perception variable of the pedestrian to be emulated, the motor imagination of the pedestrian to be emulated is predicted, as this
The simulation result of pedestrian to be emulated.
The Pedestrian Movement Simulation Studies system, wherein the displacement variable further include: the Euclidean distance of current pedestrian and target,
The current targeted direction of pedestrian, current pedestrian's directional velocity and the deviation for arriving target direction.
The Pedestrian Movement Simulation Studies system, wherein in measurement module the information Perception variable include: using sample pedestrian as the center of circle,
By the sensing region, range is divided into multiple fan-shaped subregions, counts the speed of number and subregion one skilled in the art in the subregion
Distribution.
The Pedestrian Movement Simulation Studies system, the detailed process that wherein training module is trained are as follows: sometime with sample pedestrian
The displacement variable and information Perception variable of point are as input, the displacement variable of this pedestrian of latter time point sample is as mesh
Mark, training obtain pedestrian movement's model.
The Pedestrian Movement Simulation Studies system, wherein the machine learning model is decision-tree model in training module.
The beneficial effects of the present invention are:
1) existing method (model driven method or Case-based Reasoning method) is solved when simulating group movement, local avoidance row
Not conform to the actual conditions, the rough problem of motion profile.
2) method proposed by the present invention produces truer, more natural pedestrian movement in emulation experiment compared with other methods
Simulation result.
3) proposed by the present invention method is simple, and the complexity of algorithm is lower, to hardware close friend, without high hardware
Cost.
Detailed description of the invention
Fig. 1 is the object effects schematic diagram of pedestrian's displacement variable of the present invention;
Fig. 2 is the sensing region schematic diagram of pedestrian's displacement variable of the present invention;
Fig. 3 is flow chart of the present invention.
Specific embodiment
Present system includes three parts: (1) perceiving from pedestrian's displacement state and pedestrian to other pedestrians of surrounding
Angle, to influence pedestrian movement the factor model;(2) determined based on structure of these impact factors to instance data
Then justice studies the method for learning motor pattern by machine learning model;(3) according to above-mentioned motor pattern, pedestrian's fortune is constructed
The calculation method of dynamic emulation calculates the dynamic process of pedestrian movement by given pedestrian's initial motion state and target position, real
Existing Pedestrian Movement Simulation Studies process.Shown in specific step is as follows:
1 feature selection process:
A) consider single pedestrian's displacement variable, comprising: the influence of target (destination), displacement speed shadow
It rings, the influence of history inertia motion.
B) consider that neighbor information perceives variable, comprising: number, Perception Area in preset sensing region size, sensing region
The VELOCITY DISTRIBUTION of domain one skilled in the art.
2 evacuation decision learning processes:
According to the feature that said extracted comes out, to construct instance database.Some movement current institute face of pedestrian in crowd
It is considered as input the case where facing;When facing this situation, which is output in the corresponding motor behavior that next step is taken.
Instance data is learnt from instance data as one machine learning model of sample training, trained model expression again
Motor pattern;
3 simulation calculation processes:
The first step needs given initial motion state (including position coordinates, velocity magnitude and directional velocity) and target position
It sets, is then based on the good displacement variable of processed offline stage definitions and neighbor information perception variable, constructs input condition, then
The trained model obtained by the processed offline stage obtains the prediction to next step movement, imitative by calculating final generation
Genuine result.
Specifically, the invention discloses a kind of Pedestrian Movement Simulation Studies method, including:
Step 1 passes through actual measurement, the displacement variable and information Perception variable of acquisition sample pedestrian, displacement change
Amount includes the current kinetic speed and historical movement speed of destination locations, sample pedestrian, which includes sample
Number in the sensing region range of pedestrian, the sensing region;
Step 2, using the displacement variable and the information Perception variable as training data, training machine learning model,
Obtain pedestrian movement's model;
Step 3, the initial motion state for obtaining pedestrian to be emulated and target position, pedestrian movement's model is according to this to imitative
The displacement variable and information Perception variable of true pedestrian, predicts the motor imagination of the pedestrian to be emulated, waits emulating row as this
The simulation result of people.
The Pedestrian Movement Simulation Studies method, wherein the displacement variable further include: the Euclidean distance of current pedestrian and target,
The current targeted direction of pedestrian, current pedestrian's directional velocity and the deviation for arriving target direction.
The Pedestrian Movement Simulation Studies method, wherein the information Perception variable includes: to incite somebody to action using sample pedestrian as the center of circle in step 1
The sensing region range is divided into multiple fan-shaped subregions, counts number and the speed of subregion one skilled in the art point in the subregion
Cloth.
The Pedestrian Movement Simulation Studies method, the detailed process that wherein step 2 is trained are as follows: sometime put with sample pedestrian
Displacement variable and information Perception variable as input, the displacement variable of this pedestrian of latter time point sample as target,
Training obtains pedestrian movement's model.
To allow features described above and effect of the invention that can illustrate more clearly understandable, special embodiment below, and cooperate
Bright book attached drawing is described in detail below.
Whole flow process is as shown in Figure 3, comprising:
Feature selection process:
Single pedestrian's displacement variable is considered first, comprising:
The influence of target: as shown in Figure 1, consider the Euclidean distance of current pedestrian and target, targeted direction, speed
The angle in direction and the deviation to target direction;
The influence of displacement speed: consider pedestrian itself velocity magnitude, directional velocity;
History inertia motion: historical speed size, historical speed direction.Use referring both to as history speed for multiple sampled points
Size, historical speed direction, such as the velocity magnitude and direction average value of preceding 10 history samples point are spent, quantization means history is carried out
Motional inertia.
Then consider that neighbor information perceives variable, comprising:
The size of sensing region: consider 360 ° around current pedestrian of region, and be divided into 9 fan-shaped sons for 360 °
Region (here with " 9 " signal, can choose arbitrary subregion quantity).The fan-shaped angle of fan-shaped subregion before body is designed
It is smaller and smaller closer to front, to be because the observation of people be symmetrically, so on the basis of itself direction, sector
The division in region is symmetrical.The fan-shaped angle of fan-shaped subregion behind is designed bigger, and closer to dead astern
Bigger, dead astern only exists the region (as shown in Figure 2) of a maximum angle;
Number in region: other pedestrian's numbers in statistics each subregion.
The VELOCITY DISTRIBUTION of region one skilled in the art: consider average speed situation, the speed apart from nearest pedestrian of region one skilled in the art
Spend situation, the motion conditions of fastest pedestrian.
Avoid decision learning process:
The process for learning motor pattern (Movement Patterns) from true crowd movement's data, can be regarded as
By machine learning algorithm model, to learn the process of the data rule lain in instance database (Examples), according to
The feature that said extracted comes out constructs instance database.Instance data is specifically defined, and is shown to pedestrian movement's row in crowd
For a kind of understanding of essence.A movement instance data herein, is defined as being made of two parts: first part, in crowd movement
In scene, some in crowd moves the case where pedestrian is currently faced, it is considered as a kind of input condition (Input
Situation).Second part, when facing this situation, the corresponding motor behavior which is taken in next step will
Be considered as a kind of output response (OutputAction).Again (here using instance data as one decision-tree model of sample training
By taking decision tree as an example, other machine learning models be can be used), trained model indicate from instance data study to
Motor pattern.
Simulation calculation process:
The Pedestrian Movement Simulation Studies operation phase, main purpose be in the processed offline stage from true crowd movement's data middle school
The motor pattern practised, applied to the crowd movement generated in virtual scene.Firstly the need of given intelligent body (certain of current research
A pedestrian's object, Agent) initial motion state and target position, be then based on processed offline stage definitions it is good itself fortune
Dynamic variable and neighbor information perceive variable, construct the first part of instance data, i.e. the input item of these intelligent bodies at a certain moment
Part, then the trained model obtained by the processed offline stage, prediction obtain the output response moved in next step, by calculating
The final result for generating emulation.It should be noted that pedestrian to be emulated can be to be multiple, because wait emulate starting point and mesh between pedestrian
Ground it is different, be possible to meet during pedestrian's row to be emulated to destination, that is, appear in the sensing region of emulation pedestrian
In range.
The following are system embodiment corresponding with above method embodiment, present embodiment can be mutual with above embodiment
Cooperation is implemented.The relevant technical details mentioned in above embodiment are still effective in the present embodiment, in order to reduce repetition,
Which is not described herein again.Correspondingly, the relevant technical details mentioned in present embodiment are also applicable in above embodiment.
The invention also discloses a kind of Pedestrian Movement Simulation Studies system, including:
Measurement module, for obtaining the displacement variable and information Perception variable of sample pedestrian, the displacement variable
Current kinetic speed and historical movement speed including destination locations, sample pedestrian, which includes sample row
Number in the sensing region range of people, the sensing region;
Training module, for using the displacement variable and the information Perception variable as training data, training machine
Model is practised, pedestrian movement's model is obtained;
Emulation module, for obtaining initial motion state and the target position of pedestrian to be emulated, pedestrian movement's model root
According to the displacement variable and information Perception variable of the pedestrian to be emulated, the motor imagination of the pedestrian to be emulated is predicted, as this
The simulation result of pedestrian to be emulated.
The Pedestrian Movement Simulation Studies system, wherein the displacement variable further include: the Euclidean distance of current pedestrian and target,
The current targeted direction of pedestrian, current pedestrian's directional velocity and the deviation for arriving target direction.
The Pedestrian Movement Simulation Studies system, wherein in measurement module the information Perception variable include: using sample pedestrian as the center of circle,
By the sensing region, range is divided into multiple fan-shaped subregions, counts the speed of number and subregion one skilled in the art in the subregion
Distribution.
The Pedestrian Movement Simulation Studies system, the detailed process that wherein training module is trained are as follows: sometime with sample pedestrian
The displacement variable and information Perception variable of point are as input, the displacement variable of this pedestrian of latter time point sample is as mesh
Mark, training obtain pedestrian movement's model.
The Pedestrian Movement Simulation Studies system, wherein the machine learning model is decision-tree model in training module.
Claims (10)
1. a kind of Pedestrian Movement Simulation Studies method characterized by comprising
Step 1 passes through actual measurement, the displacement variable and information Perception variable of acquisition sample pedestrian, the displacement variable packet
The current kinetic speed and historical movement speed of destination locations, sample pedestrian are included, which includes sample pedestrian
Sensing region range, number in the sensing region;
Step 2, using the displacement variable and the information Perception variable as training data, training machine learning model obtains
Pedestrian movement's model;
Step 3, the initial motion state for obtaining pedestrian to be emulated and target position, pedestrian movement's model is according to the row to be emulated
The displacement variable and information Perception variable of people, predicts the motor imagination of the pedestrian to be emulated, as the pedestrian's to be emulated
Simulation result.
2. Pedestrian Movement Simulation Studies method as described in claim 1, which is characterized in that the displacement variable further include: current
The Euclidean distance of pedestrian and target, the targeted direction of current pedestrian, current pedestrian's directional velocity and inclined to target direction
From.
3. Pedestrian Movement Simulation Studies method as described in claim 1, which is characterized in that the information Perception variable includes: in step 1
Using sample pedestrian as the center of circle, by the sensing region, range is divided into multiple fan-shaped subregions, counts number in the subregion and is somebody's turn to do
The VELOCITY DISTRIBUTION of subregion one skilled in the art.
4. Pedestrian Movement Simulation Studies method as described in claim 1, which is characterized in that the detailed process of the step 2 training are as follows: with
Displacement variable and information Perception variable that sample pedestrian sometime puts as input, this pedestrian of latter time point sample from
Body kinematic variables obtain pedestrian movement's model as target, training.
5. Pedestrian Movement Simulation Studies method as described in claim 1, which is characterized in that the machine learning model is certainly in step 2
Plan tree-model.
6. a kind of Pedestrian Movement Simulation Studies system characterized by comprising
Measurement module, for obtaining the displacement variable and information Perception variable of sample pedestrian, which includes
Destination locations, sample pedestrian current kinetic speed and historical movement speed, which includes sample pedestrian
Number in sensing region range, the sensing region;
Training module, for using the displacement variable and the information Perception variable as training data, training machine to learn mould
Type obtains pedestrian movement's model;
Emulation module, for obtaining initial motion state and the target position of pedestrian to be emulated, pedestrian movement's model is according to this
The displacement variable and information Perception variable of pedestrian to be emulated, predicts the motor imagination of the pedestrian to be emulated, waits imitating as this
The simulation result of true pedestrian.
7. Pedestrian Movement Simulation Studies system as claimed in claim 6, which is characterized in that the displacement variable further include: current
The Euclidean distance of pedestrian and target, the targeted direction of current pedestrian, current pedestrian's directional velocity and inclined to target direction
From.
8. Pedestrian Movement Simulation Studies system as claimed in claim 6, which is characterized in that the information Perception variable packet in measurement module
Include: using sample pedestrian as the center of circle, by the sensing region, range is divided into multiple fan-shaped subregions, count in the subregion number and
The VELOCITY DISTRIBUTION of subregion one skilled in the art.
9. Pedestrian Movement Simulation Studies system as claimed in claim 6, which is characterized in that the detailed process of training module training
Are as follows: the displacement variable and information Perception variable sometime put using sample pedestrian as input, latter time point sample current row
The displacement variable of people obtains pedestrian movement's model as target, training.
10. Pedestrian Movement Simulation Studies system as claimed in claim 6, which is characterized in that the machine learning model in training module
For decision-tree model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810982158.1A CN109376371A (en) | 2018-08-27 | 2018-08-27 | A kind of Pedestrian Movement Simulation Studies method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810982158.1A CN109376371A (en) | 2018-08-27 | 2018-08-27 | A kind of Pedestrian Movement Simulation Studies method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109376371A true CN109376371A (en) | 2019-02-22 |
Family
ID=65404747
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810982158.1A Pending CN109376371A (en) | 2018-08-27 | 2018-08-27 | A kind of Pedestrian Movement Simulation Studies method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109376371A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110335300A (en) * | 2019-05-14 | 2019-10-15 | 广东康云科技有限公司 | Scene dynamics analogy method, system and storage medium based on video fusion |
CN111671405A (en) * | 2020-05-29 | 2020-09-18 | 昭苏县西域马业有限责任公司 | Saddle with health detection device, health detection system and method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101075352A (en) * | 2007-06-29 | 2007-11-21 | 中国科学院计算技术研究所 | Laminated barrier-avoiding method for dynamic body |
CN101216951A (en) * | 2007-12-27 | 2008-07-09 | 电子科技大学 | Intelligent group motion simulation method in virtual scenes |
BR102014026998A2 (en) * | 2014-10-29 | 2016-07-12 | José Henrique Porto Silveira | process for obtaining environmental education indicators by perception |
CN107480821A (en) * | 2017-08-14 | 2017-12-15 | 山东师范大学 | The multi-Agent cooperation crowd evacuation emulation method and device of instance-based learning |
CN108388752A (en) * | 2018-03-22 | 2018-08-10 | 中国科学院计算技术研究所 | One kind of groups emulation mode |
CN108428243A (en) * | 2018-03-07 | 2018-08-21 | 北京航空航天大学 | A kind of pedestrian movement's speed predicting method based on artificial neural network |
-
2018
- 2018-08-27 CN CN201810982158.1A patent/CN109376371A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101075352A (en) * | 2007-06-29 | 2007-11-21 | 中国科学院计算技术研究所 | Laminated barrier-avoiding method for dynamic body |
CN101216951A (en) * | 2007-12-27 | 2008-07-09 | 电子科技大学 | Intelligent group motion simulation method in virtual scenes |
BR102014026998A2 (en) * | 2014-10-29 | 2016-07-12 | José Henrique Porto Silveira | process for obtaining environmental education indicators by perception |
CN107480821A (en) * | 2017-08-14 | 2017-12-15 | 山东师范大学 | The multi-Agent cooperation crowd evacuation emulation method and device of instance-based learning |
CN108428243A (en) * | 2018-03-07 | 2018-08-21 | 北京航空航天大学 | A kind of pedestrian movement's speed predicting method based on artificial neural network |
CN108388752A (en) * | 2018-03-22 | 2018-08-10 | 中国科学院计算技术研究所 | One kind of groups emulation mode |
Non-Patent Citations (3)
Title |
---|
曹爱春 等: "Agent-CA的体育场馆人群疏散模型", 《计算机工程与应用》 * |
田景文 等著: "《人工神经网络算法研究及应用》", 31 July 2006, 北京理工大学出版社 * |
许汉成: "基于复杂网络控制及多智能体***的人群行为仿真研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110335300A (en) * | 2019-05-14 | 2019-10-15 | 广东康云科技有限公司 | Scene dynamics analogy method, system and storage medium based on video fusion |
CN111671405A (en) * | 2020-05-29 | 2020-09-18 | 昭苏县西域马业有限责任公司 | Saddle with health detection device, health detection system and method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Tai et al. | Socially compliant navigation through raw depth inputs with generative adversarial imitation learning | |
CN105056528B (en) | A kind of moving method and device of virtual role | |
CN109543285B (en) | Crowd evacuation simulation method and system integrating data driving and reinforcement learning | |
CN103679264B (en) | Crowd evacuation paths planning method based on artificial fish-swarm algorithm | |
Choi et al. | Multi-focus attention network for efficient deep reinforcement learning | |
CN113255936B (en) | Deep reinforcement learning strategy protection defense method and device based on imitation learning and attention mechanism | |
Wang et al. | Probabilistic modeling of human movements for intention inference | |
CN106022239A (en) | Multi-target tracking method based on recurrent neural network | |
CN111461437B (en) | Data-driven crowd motion simulation method based on generation of countermeasure network | |
CN110210320A (en) | The unmarked Attitude estimation method of multiple target based on depth convolutional neural networks | |
JP5905481B2 (en) | Determination method and determination apparatus | |
CN109986560A (en) | A kind of mechanical arm self-adapting grasping method towards multiple target type | |
CN105468801A (en) | Simulation method and system for crowd evacuation in public place | |
Zieliński et al. | 3D robotic navigation using a vision-based deep reinforcement learning model | |
CN112508164B (en) | End-to-end automatic driving model pre-training method based on asynchronous supervised learning | |
CN109376371A (en) | A kind of Pedestrian Movement Simulation Studies method and system | |
CN110956684B (en) | Crowd movement evacuation simulation method and system based on residual error network | |
Liu et al. | Active object detection based on a novel deep Q-learning network and long-term learning strategy for the service robot | |
Kumra et al. | Learning robotic manipulation tasks via task progress based Gaussian reward and loss adjusted exploration | |
CN109731338A (en) | Artificial intelligence training method and device, storage medium and electronic device in game | |
Price et al. | GA directed self-organized search and attack UAV swarms | |
Antonelo et al. | Mobile robot control in the road sign problem using reservoir computing networks | |
Yang et al. | Mapless navigation for UAVs via reinforcement learning from demonstrations | |
Contardo et al. | Learning states representations in pomdp | |
Ritz et al. | Towards ecosystem management from greedy reinforcement learning in a predator-prey setting |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190222 |