CN112857370A - Robot map-free navigation method based on time sequence information modeling - Google Patents
Robot map-free navigation method based on time sequence information modeling Download PDFInfo
- Publication number
- CN112857370A CN112857370A CN202110018866.5A CN202110018866A CN112857370A CN 112857370 A CN112857370 A CN 112857370A CN 202110018866 A CN202110018866 A CN 202110018866A CN 112857370 A CN112857370 A CN 112857370A
- Authority
- CN
- China
- Prior art keywords
- mobile robot
- navigation
- robot
- target position
- 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
- 238000000034 method Methods 0.000 title claims abstract description 49
- 238000013528 artificial neural network Methods 0.000 claims abstract description 19
- 238000005516 engineering process Methods 0.000 claims abstract description 8
- 125000004122 cyclic group Chemical group 0.000 claims abstract description 5
- 238000013507 mapping Methods 0.000 claims abstract description 5
- 238000004088 simulation Methods 0.000 claims description 18
- 238000012549 training Methods 0.000 claims description 12
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 5
- 230000033001 locomotion Effects 0.000 description 5
- 230000000306 recurrent effect Effects 0.000 description 5
- 230000007613 environmental effect Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000002787 reinforcement Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 240000007594 Oryza sativa Species 0.000 description 1
- 235000007164 Oryza sativa Nutrition 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 235000009566 rice Nutrition 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/20—Instruments for performing navigational calculations
- G01C21/206—Instruments for performing navigational calculations specially adapted for indoor navigation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/005—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 with correlation of navigation data from several sources, e.g. map or contour matching
Abstract
The invention discloses a robot map-free navigation method based on time sequence information modeling, which comprises the following steps: 1) constructing a mobile robot navigation model based on a cyclic neural network, wherein the robot navigation model contains a mapping relation from the mobile robot to a mobile robot execution speed instruction according to sensor information on the mobile robot and target position information in a scene; the mobile robot reaches a navigation target position according to a speed instruction output by the robot navigation model; 2) acquiring or constructing a navigation data set as supervision data to train the mobile robot navigation model; 3) sensor data are obtained through a laser radar carried by the mobile robot, target position information in a scene is obtained through a positioning technology, then an execution speed instruction of the mobile robot is calculated through a trained mobile robot navigation model, and the mobile robot is controlled to reach a target position through the execution speed instruction. The invention can enable the mobile robot to obtain the obstacle avoidance navigation capability.
Description
Technical Field
The invention belongs to the field of information science, relates to a navigation method, and particularly relates to a robot map-free navigation method based on time sequence information modeling.
Background
To date, robots have created great value in the fields of industrial manufacturing, home service, interplanetary exploration, military reconnaissance, and the like. Compared with a fixed structure robot commonly used in the industrial field, the mobile robot has the important characteristics of mobility and flexibility, and in order to realize the mobility of the mobile robot, autonomous navigation is one of indispensable technologies. Autonomous navigation means that a mobile robot senses external environment information through sensors such as radars, sonars and cameras, and completes an autonomous movement process of reaching a target point without collision in an obstacle environment by combining self state information. The mobile robot has an efficient and reliable navigation function, can be better applied to the aspects of industry, service, military and the like, and brings higher value to the society.
The robot navigation technique can be divided into two cases: facing a known environment and facing an unknown environment. In the case of a navigation map in which the robot has an environment, which is referred to as a known environment, a map-based navigation method may be used. The navigation map used by the robot is generally constructed by slam (simultaneous Localization and mapping), and then the robot path is planned by using a path planning algorithm to realize navigation. Fig. 1 is an example of map-based robot navigation.
The map-based navigation robot method needs to take a long time to construct an environment map, and cannot adapt to an unfamiliar environment. In the case of an unknown environment, the robot navigation is also called map-less navigation, and commonly used traditional algorithms include a dynamic window algorithm, a D-ary algorithm, a vector histogram algorithm and the like. In addition, with the development of deep learning, a learning-based method is gradually becoming the mainstream of the research of a map-free navigation method, an unsupervised learning method can be used, the robot navigation process is considered to have markov property, the navigation is modeled based on reinforcement learning, and the robot learns how to make appropriate action according to the current state, and the learning of the navigation strategy is completed through random exploration. However, the unsupervised method based on reinforcement learning has the disadvantages of long learning time, low data utilization rate, difficulty in model convergence and incapability of guaranteeing the navigation effect. For the problem, some learners solve the navigation problem by using a method based on supervised learning, and build a neural network model to fit navigation data after acquiring navigation supervision data.
The mobile robot acquires external environment information through a sensor, and in the work of predecessors, the sensor information is usually directly processed by a fully-connected neural network, or a convolutional neural network is used for extracting sensor data characteristics. However, these methods do not fully utilize the information provided by the data time sequence characteristics, and cannot predict time-varying objects (such as dynamic obstacles) in a dynamic environment, thereby causing a significant reduction in navigation performance in a complex dynamic scene.
Disclosure of Invention
The invention aims to provide a robot map-free navigation method, and the mobile robot map-free navigation method based on time sequence information modeling obtains a navigation model with better performance by constructing a large-scale navigation data set and aiming at the time sequence information modeling in navigation data. Under any environment, target position information is obtained through the existing positioning technology, the robot avoids obstacles in the environment to reach the target position, and the robot can be used in scenes such as automatic person searching of shopping carts of robots in shopping malls, automatic person searching of service robots in hospitals and the like.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a robot map-free navigation method based on time sequence information modeling comprises the following steps:
constructing a mobile robot navigation model based on a cyclic neural network, wherein the robot navigation model contains a mapping relation from the mobile robot to a mobile robot execution speed instruction according to sensor information on the mobile robot and target position information in a scene; the mobile robot reaches a navigation target position according to a speed instruction output by the robot navigation model;
acquiring or constructing a navigation data set as supervision data to train the mobile robot navigation model;
sensor data are obtained through a laser radar carried by the mobile robot, target position information in a scene is obtained through a positioning technology, then an execution speed instruction of the mobile robot is calculated through a trained mobile robot navigation model, and the mobile robot is controlled to reach a target position through the execution speed instruction.
Further, the method for constructing the navigation data set comprises the following steps: building different simulation environments according to various real indoor environment layouts; then in a simulation environment, a mobile robot simulation model pre-constructs an environment map, then randomly sets a target point, plans a navigation path and navigates by using a path planning algorithm, and records the state information of the mobile robot in the navigation process; the collected mobile robot state information is then used as a data set for navigation model training.
Further, the mobile robot state information includes 360-dimensional radar information, 2-dimensional velocity information, and 2-dimensional target position information.
Furthermore, under a Gazebo simulation environment, different simulation environments are built according to various real indoor environment layouts.
Further, the mobile robot simulation model previously uses the SLAM technology to construct an environment map.
Further, the mobile robot navigation model comprises an LSTM neural network and a CNN; the CNN is used for extracting the spatial characteristics of the current moment from the laser data collected by the mobile robot and inputting the spatial characteristics into the LSTM neural network, and acquiring the relative target position in a coordinate system taking the mobile robot as the center at the current moment according to the laser data collected by the mobile robot; and then the LSTM neural network predicts the execution speed of the mobile robot at the current moment according to the moving speed of the mobile robot at the last moment, the relative target position information at the current moment and the spatial characteristics at the current moment.
Further, the relative target position is represented by polar coordinates including distance and angle.
Further, the speed of the mobile robot includes a linear speed and an angular speed.
Further, the method for training the mobile robot navigation model comprises the following steps: and training the mobile robot navigation model by using the navigation data set as supervision data, wherein the loss function uses a mean square error loss function, and the model training is realized by a gradient descent method.
The map-free navigation method of the mobile robot comprises the following steps:
and constructing a mobile robot navigation model based on a cyclic neural network, wherein the model contains a mapping relation from the sensor information and the target position information of the mobile robot to an execution speed instruction, and the robot reaches a navigation target position according to the speed instruction output by the model.
Navigation models are trained using SLAM-based navigation methods to build a navigation data set as supervisory data. The method comprises the steps of obtaining sensor data through a laser radar carried by the robot, obtaining target position information through a positioning technology, calculating an execution speed instruction of the mobile robot through a navigation model, and controlling the mobile robot to reach a target position through an execution control instruction.
Compared with the prior art, the invention has the following positive effects:
the invention explicitly utilizes the time sequence characteristics of the sensor signals and improves the navigation performance of the robot in a complex environment. The navigation success rate and the navigation efficiency of the robot are improved compared with those of the traditional SLAM method, and specific data and analysis are shown in the following table 1.
The method can enable the mobile robot to obtain the obstacle avoidance navigation capability under the condition of obtaining the environment target information.
Drawings
FIG. 1 is a schematic view of a mobile robot navigation;
FIG. 2 is a schematic diagram of a recurrent neural network model.
Detailed Description
The invention provides a map-free navigation method for realizing the optimal collision-free action of a mobile robot in an unknown environment. The invention provides an end-to-end navigation model facing to a target and facing to a robot platform through learning demonstration. The model can learn a complex strategy: the robot selects a moving mode according to the environment information, wherein the moving mode comprises an original 2D laser ranging result and a target position. The environmental information measured by the sensor has a time series characteristic: for example, from two-dimensional laser ranging findings of the current time and the last time, it can be determined whether the robot is approaching or departing from the surrounding environmental obstacle; according to the target position information of the current time and the previous time, whether the current movement of the robot is close to or far away from the target can be determined. From time series information of the environmental state of the navigation process, we learn the navigation strategy from the presentation data using a recurrent neural network. Long and Short Term Memory (LSTM) neural networks are one type of recurrent neural networks, and navigation strategies using this neural network architecture can be used to plan movements based on past circumstances and movement states. In order to verify the validity of the navigation model of the work, compared with the latest method, a quantitative and accurate evaluation method is provided.
The invention models the time sequence information of the mobile robot in the motion process through the recurrent neural network so as to learn the navigation capability. The following describes the implementation process, and in conjunction with the accompanying drawings, specifically describes the method for learning navigation based on the recurrent neural network proposed in the present invention.
(1) Data acquisition: in the technical scheme adopted by the invention, navigation data sets are relied on, and as no open-source navigation data set exists at present, a data set of the user needs to be constructed. Under the Gazebo simulation environment, different simulation environments are built according to various real indoor environment layouts. In a simulation environment, a mobile robot simulation model uses SLAM technology to construct an environment map in advance, then target points are randomly set, and a navigation path is planned and navigated by using a path planning algorithm. And recording the state information of the mobile robot in the navigation process, including 360-dimensional radar information, 2-dimensional speed information and 2-dimensional target position information, and setting the sampling interval to be 0.1 second. The whole data set comprises 8 different environments, 5-7 random target points are set in each environment, and finally more than 1000 pieces of sampling data are supplied to a subsequent scheme for training.
(2) Constructing a model: our work has focused on the modeling of timing information during navigation. To model the complex association between the robot and the obstacle, we use the LSTM neural network, which has successfully demonstrated its high performance in sequence modeling and prediction tasks in recent years. As shown in fig. 2, the input is given by the 2D laser rangefinder, relative target position in the coordinate system of the robot center and the last minute speed of the robot. The laser range measurement has a size of 360 degrees, the relative target position is represented by polar coordinates including distance and angle, and the robot speed at the last moment includes linear speed and angular speed. In order to extract the spatial features of the laser data, the 360-dimensional laser data is processed using the CNN; the dimension reduction is then performed through the full link layer FC1, which is a good way to model the spatial characteristics of the laser data. Then, the processed laser information, relative to the target position, three information channels of the robot velocity at one moment are used as input and are fused by the two-layer stacked LSTM, and finally the execution speed of the mobile robot is predicted through the full connection layer FC 2.
(3) Model training: and training a cyclic neural network navigation model by using the acquired navigation data as supervision data, wherein the loss function uses a mean square error loss function, and the model training is realized by a gradient descent method.
In order to verify the effectiveness of the method, simulation experiments are carried out on a Gazebo simulation platform, and the experimental results are shown in Table 1:
table 1 is an experimental result table
| Navigation success rate (%) | Average time length (seconds) | Average mileage (rice) | |
| Method for producing a composite material | 89.1 | 26.8 | 5.7 |
| Navigation method based on SLAM | 25.3 | 25.3 | 5.3 |
From table 1, it can be seen that the method provided by the present invention can improve the navigation success rate and the navigation efficiency of the mobile robot.
The above embodiments are only intended to illustrate the technical solution of the present invention, but not to limit it, and a person skilled in the art can modify the technical solution of the present invention or substitute it with an equivalent, and the protection scope of the present invention is subject to the claims.
Claims (9)
1. A robot map-free navigation method based on time sequence information modeling comprises the following steps:
1) constructing a mobile robot navigation model based on a cyclic neural network, wherein the robot navigation model contains a mapping relation from the mobile robot to a mobile robot execution speed instruction according to sensor information on the mobile robot and target position information in a scene; the mobile robot reaches a navigation target position according to a speed instruction output by the robot navigation model;
2) acquiring or constructing a navigation data set as supervision data to train the mobile robot navigation model;
3) sensor data are obtained through a laser radar carried by the mobile robot, target position information in a scene is obtained through a positioning technology, then an execution speed instruction of the mobile robot is calculated through a trained mobile robot navigation model, and the mobile robot is controlled to reach a target position through the execution speed instruction.
2. The method of claim 1, wherein the navigation data set is constructed by: building different simulation environments according to various real indoor environment layouts; then in a simulation environment, a mobile robot simulation model pre-constructs an environment map, then randomly sets a target point, plans a navigation path and navigates by using a path planning algorithm, and records the state information of the mobile robot in the navigation process; the collected mobile robot state information is then used as a data set for navigation model training.
3. The method of claim 2, wherein the mobile robot state information includes 360-dimensional radar information, 2-dimensional velocity information, 2-dimensional target position information.
4. The method of claim 2, wherein in a Gazebo simulation environment, different simulation environments are built according to a plurality of real indoor environment layouts.
5. The method of claim 2, wherein the mobile robot simulation model previously constructed the environment map using SLAM techniques.
6. The method of claim 1, wherein the mobile robot navigation model comprises an LSTM neural network and a CNN; the CNN is used for extracting the spatial characteristics of the current moment from the laser data collected by the mobile robot and inputting the spatial characteristics into the LSTM neural network, and acquiring the relative target position in a coordinate system taking the mobile robot as the center at the current moment according to the laser data collected by the mobile robot; and then the LSTM neural network predicts the execution speed of the mobile robot at the current moment according to the moving speed of the mobile robot at the last moment, the relative target position information at the current moment and the spatial characteristics at the current moment.
7. The method of claim 6, wherein the relative target position is represented by polar coordinates including distance and angle.
8. The method of claim 6, wherein the speed of the mobile robot comprises a linear speed and an angular speed.
9. The method of claim 1, wherein the method of training the mobile robot navigation model comprises: and training the mobile robot navigation model by using the navigation data set as supervision data, wherein the loss function uses a mean square error loss function, and the model training is realized by a gradient descent method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110018866.5A CN112857370A (en) | 2021-01-07 | 2021-01-07 | Robot map-free navigation method based on time sequence information modeling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110018866.5A CN112857370A (en) | 2021-01-07 | 2021-01-07 | Robot map-free navigation method based on time sequence information modeling |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112857370A true CN112857370A (en) | 2021-05-28 |
Family
ID=76004909
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110018866.5A Pending CN112857370A (en) | 2021-01-07 | 2021-01-07 | Robot map-free navigation method based on time sequence information modeling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112857370A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113486871A (en) * | 2021-09-07 | 2021-10-08 | 中国人民解放军国防科技大学 | Unmanned vehicle local autonomous control method, device and equipment based on depth map |
| CN113485382A (en) * | 2021-08-26 | 2021-10-08 | 苏州大学 | Mobile robot autonomous navigation method and system for man-machine natural interaction |
| CN113959446A (en) * | 2021-10-20 | 2022-01-21 | 苏州大学 | Robot autonomous logistics transportation navigation method based on neural network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106950969A (en) * | 2017-04-28 | 2017-07-14 | 深圳市唯特视科技有限公司 | It is a kind of based on the mobile robot continuous control method without map movement planner |
| US20170219359A1 (en) * | 2015-12-21 | 2017-08-03 | Invensense, Inc. | Method and system for estimating uncertainty for offline map information aided enhanced portable navigation |
| CN111141300A (en) * | 2019-12-18 | 2020-05-12 | 南京理工大学 | Intelligent mobile platform map-free autonomous navigation method based on deep reinforcement learning |
-
2021
- 2021-01-07 CN CN202110018866.5A patent/CN112857370A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170219359A1 (en) * | 2015-12-21 | 2017-08-03 | Invensense, Inc. | Method and system for estimating uncertainty for offline map information aided enhanced portable navigation |
| CN106950969A (en) * | 2017-04-28 | 2017-07-14 | 深圳市唯特视科技有限公司 | It is a kind of based on the mobile robot continuous control method without map movement planner |
| CN111141300A (en) * | 2019-12-18 | 2020-05-12 | 南京理工大学 | Intelligent mobile platform map-free autonomous navigation method based on deep reinforcement learning |
Non-Patent Citations (2)
| Title |
|---|
| MARK PFEIFFER 等，: ""From perception to decision: A data-driven approach to end-to-end motion planning for autonomous ground robots"", 《2017 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA)》 * |
| 马留龙，: ""基于强化学习的无地图导航策略研究"", 《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113485382A (en) * | 2021-08-26 | 2021-10-08 | 苏州大学 | Mobile robot autonomous navigation method and system for man-machine natural interaction |
| CN113486871A (en) * | 2021-09-07 | 2021-10-08 | 中国人民解放军国防科技大学 | Unmanned vehicle local autonomous control method, device and equipment based on depth map |
| CN113959446A (en) * | 2021-10-20 | 2022-01-21 | 苏州大学 | Robot autonomous logistics transportation navigation method based on neural network |
| CN113959446B (en) * | 2021-10-20 | 2024-01-23 | 苏州大学 | Autonomous logistics transportation navigation method for robot based on neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sun et al. | Motion planning for mobile robots—Focusing on deep reinforcement learning: A systematic review | |
| Fan et al. | Crowdmove: Autonomous mapless navigation in crowded scenarios | |
| Cheng et al. | Topological indoor localization and navigation for autonomous mobile robot | |
| CN112857370A (en) | Robot map-free navigation method based on time sequence information modeling | |
| Cao et al. | Target search control of AUV in underwater environment with deep reinforcement learning | |
| Zghair et al. | A one decade survey of autonomous mobile robot systems | |
| CN105629970A (en) | Robot positioning obstacle-avoiding method based on supersonic wave | |
| Gao et al. | Multi-mobile robot autonomous navigation system for intelligent logistics | |
| Chen et al. | Robot navigation with map-based deep reinforcement learning | |
| CN108320051B (en) | Mobile robot dynamic collision avoidance planning method based on GRU network model | |
| Kim et al. | Motion planning by reinforcement learning for an unmanned aerial vehicle in virtual open space with static obstacles | |
| Li et al. | A behavior-based mobile robot navigation method with deep reinforcement learning | |
| Guan et al. | Robot formation control based on internet of things technology platform | |
| Sundram et al. | Development of a miniature robot for multi-robot occupancy grid mapping | |
| Karpov et al. | Multi-robot exploration and mapping based on the subdefinite models | |
| Wang et al. | Research on autonomous planning method based on improved quantum Particle Swarm Optimization for Autonomous Underwater Vehicle | |
| Short et al. | Abio-inspiredalgorithminimage-based pathplanning and localization using visual features and maps | |
| CN114200920A (en) | Path planning method, device and control system | |
| Baranzadeh | A decentralized control algorithm for target search by a multi-robot team | |
| Zeng et al. | Robot navigation in crowd based on dual social attention deep reinforcement learning | |
| CN115790600A (en) | Algorithm of task coverage strategy for long-term inspection of robot in large range based on digital twin | |
| Li et al. | Vision-based obstacle avoidance algorithm for mobile robot | |
| Zhang et al. | 2D map building and path planning based on LiDAR | |
| Li et al. | An enhanced direction calibration based on reinforcement learning for indoor localization system | |
| Chen et al. | Deep reinforcement learning-based robot exploration for constructing map of unknown environment |
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: 20210528 |