CN116858253A - Lightweight predictive navigation method and system suitable for indoor environment - Google Patents

Lightweight predictive navigation method and system suitable for indoor environment Download PDF

Info

Publication number
CN116858253A
CN116858253A CN202310965884.3A CN202310965884A CN116858253A CN 116858253 A CN116858253 A CN 116858253A CN 202310965884 A CN202310965884 A CN 202310965884A CN 116858253 A CN116858253 A CN 116858253A
Authority
CN
China
Prior art keywords
dynamic
future
static
constraint
environment
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
Application number
CN202310965884.3A
Other languages
Chinese (zh)
Inventor
郑南宁
张晓彤
符嘉玮
史佳敏
陈仕韬
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ningbo Shun'an Artificial Intelligence Research Institute
Xian Jiaotong University
Original Assignee
Ningbo Shun'an Artificial Intelligence Research Institute
Xian Jiaotong University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Ningbo Shun'an Artificial Intelligence Research Institute, Xian Jiaotong University filed Critical Ningbo Shun'an Artificial Intelligence Research Institute
Priority to CN202310965884.3A priority Critical patent/CN116858253A/en
Publication of CN116858253A publication Critical patent/CN116858253A/en
Pending legal-status Critical Current

Links

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
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • 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
    • G06N3/0455—Auto-encoder networks; Encoder-decoder 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Automation & Control Theory (AREA)
  • Probability & Statistics with Applications (AREA)
  • Algebra (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application discloses a lightweight predictive navigation method and a system suitable for indoor environment, wherein environmental prediction combines obtained perception information and motion information, and can generate future environmental prediction and mapping which are expressed as an occupied raster pattern sequence; the motion planning takes a grid sequence with time attribute as a search space, solves and formulates a track as a nonlinear optimization problem, and solves the nonlinear optimization problem to obtain the track; under a unified environment representation form, acquiring perception data and vehicle motion data (speed and coordinates) to partially decouple a dynamic environment, obtaining a future position by the static environment through coordinate transformation and mapping, predicting a grid map of a plurality of steps in the future by extracting scene motion characteristic information, then directly utilizing a future grid, simultaneously processing dynamic and static obstacles, and improving the efficiency and effectiveness of track optimization; the method provided by the application can improve the real-time performance and the safety, and simultaneously reduce the calculation requirement.

Description

Lightweight predictive navigation method and system suitable for indoor environment
Technical Field
The application belongs to the technical field of intelligent navigation, and particularly relates to a lightweight predictive navigation method and system suitable for an indoor environment.
Background
To accomplish a particular task, the robot requires accurate and safe navigation to reach the destination without collision. This navigation is achieved by implementing a navigation algorithm. In a dynamic environment, the algorithm comprises two main components: and (5) predicting and planning. The prediction component provides information regarding the local obstacle distribution, and the planning component generates a safe and seamless trajectory based on the prediction information to accommodate dynamic changes. Although researchers have investigated the separate implementation of these two functions, they have found insufficient performance in lightweight, unstructured indoor dynamic environments. Due to the crowded space, the obstacle has high density and small volume, and lacks a unified representation form, so that the information of the obstacle is lost. In addition, the limited space limits the intelligent body, and the poor real-time performance of the intelligent body can not timely avoid maneuver, so that collision is caused.
Most existing methods have high requirements on a sensor system, are long in time consumption, and are difficult to meet the requirements on real-time performance of intelligent agent planning. In addition, these methods employ object tracking to obtain future trajectories of dynamic obstacles, which are then used as inputs to the planning module. However, many methods rely on large-scale models, resulting in poor real-time performance. This can be attributed to the limitations of robots in terms of cost and size, making it challenging to implement predictive methods that require excessive resources. Furthermore, the lack of a uniform representation between predictions and plans results in loss of basic information during the conversion of multiple input and output formats. Furthermore, planning algorithms often lack perspective on future behavior of the obstacle, which may lead to the planning algorithm becoming dilemma, resulting in track deviation or discontinuities. Most existing prediction methods rely heavily on multi-line lidar, which is expensive and generates a large-scale point cloud, adding additional computational burden to indoor agent navigation. Traditional planning algorithms require high quality prediction information, but are less robust, resulting in limited practical applicability.
Disclosure of Invention
In order to solve the problems in the prior art, the lightweight predictive navigation method suitable for indoor environments adopts a layered prediction-planning method, obtains sensing data and motion data (speed and coordinates) of a vehicle per se under a unified environment representation form, partially decouples a dynamic environment, obtains a future position through coordinate transformation and mapping by the static environment, predicts a grid map of a plurality of steps in the future by extracting scene motion characteristic information, then directly utilizes a future grid, and simultaneously processes dynamic and static obstacles, thereby improving the efficiency and the effectiveness of track optimization.
In order to achieve the above purpose, the technical scheme adopted by the application is as follows: a lightweight predictive navigation method applicable to indoor environments comprises the following steps:
the method comprises the steps of performing coordinate conversion of sensing measurement values on pose information, speed information and environmental characterization of an intelligent agent obtained by a multi-sensor system, and projecting environmental variables to a coordinate system where the intelligent agent is located in the future;
taking into consideration the self-walking of the dynamic barrier and the occupation of the static barrier in the environment of the environment variable under the coordinate system of the future intelligent agent, capturing the interaction between the dynamic barrier and the static barrier, decoupling the dynamic component and the static component, and extracting the potential variable characteristics of the future probability occupation grid map;
constructing an OGM occupation raster pattern prediction generation module by adopting a variation automatic encoder, taking potential variable characteristics of a future probability occupation raster pattern as input, and outputting a predicted occupation raster pattern sequence by a decoder;
sequentially using each raster pattern as a feasible space for sampling candidate trajectories;
generating candidate tracks in the feasible space, and acquiring initial tracks meeting kinematic constraint and barrier constraint;
a smoother and safer trajectory is generated based on the initial trajectory using a graph optimization method.
Further, the method for projecting the environmental variable to the coordinate system where the future intelligent agent is located by performing coordinate conversion of the sensing measurement value on the pose information, the speed information and the environmental characterization of the intelligent agent obtained by the multi-sensor system comprises the following steps:
firstly, calculating the current obstacle occupation position according to the measured value of the laser radar, and acquiring the obstacle occupation position under the coordinate system of the current intelligent agent;
deducing the pose of the intelligent agent at the next moment according to the conversion relation between the current pose and the next moment of the intelligent agent and the current pose information of the intelligent agent; and projecting the current obstacle distribution to a coordinate system of the pose of the intelligent agent after the set time length, and obtaining the projection of the environment variable under the coordinate system of the intelligent agent in the future.
Further, taking into account the occupation of the dynamic obstacle by itself and the static obstacle in the environment, capturing the interaction between the dynamic obstacle and the static obstacle, and decoupling the dynamic component and the static component, extracting the potential variable features of the future probability occupation grid map specifically comprises the following steps:
feeding the whole grid map information into a convolution long-short time memory network module, wherein the convolution long-short time memory network module takes a one-dimensional occupied grid graph as input, and generates dynamic grid features by using a convolution kernel with the size of 3;
predicting a local map of the static obstacle in the future by adopting a Bayesian generation method, and filtering out dynamic information to obtain local map information of the static obstacle;
and taking the extracted dynamic grid characteristics and the local map information of the static obstacle as the input of the OWM occupation grid map prediction generation module.
Further, the OWM occupation raster image prediction generation module comprises an encoder and a decoder, wherein the encoder is composed of two layers of convolution layers and two layers of residual layers, and parameters of the two layers of residual layers are consistent; the decoder consists of two residual layers and two deconvolution layers, the residual layer structure is identical to that in the encoder.
Further, when each raster pattern is sequentially used as a feasible space for sampling candidate tracks, specifically, based on a spreading sequence of occupied raster patterns, the occupied space in a prediction sequence is derived through a position transformation relation, in a set time period, an intelligent body uniformly moves, the center of mass is a circle center, the feasible space representation of each frame is obtained through an area which can be reached by the intelligent body, and states in a feasible interval observe strict speed interval constraint and collision constraint.
Further, generating candidate tracks in the feasible space, selecting frames in the feasible space by adopting a random sampling method when acquiring initial tracks meeting kinematic constraint and barrier constraint, independently sampling raster pattern sequences one by one, wherein the sampled space represents an unoccupied area in each pattern, and acquiring a search space on the basis of meeting the kinematic constraint of an intelligent agent and ensuring accessibility during sampling; when sampling is carried out, if sampling points between adjacent frames can be connected under the condition of no collision, judging whether the connected edges meet the dynamic constraint and the kinematic constraint of an intelligent agent, and if so, connecting the sampling points together to form the edges; traversing the node list to identify the lowest cost node and initiating a backtracking process to obtain an initial trajectory that satisfies the kinematic and obstacle constraints.
Further, generating a smoother, safer trajectory based on the initial trajectory using a graph optimization method includes:
creating smoother and safer trajectories using the graph optimization method is formulated as a nonlinear optimization problem:
wherein L ({ P) t }|{m t:t+β }) represents optimization objective, { m t:t+β "represents a raster pattern sequence, C ego Represents the outline of the bicycle, C obs Represents the outline of the obstacle and,represents the maximum and minimum speeds achievable, ω min ,ω {max} Represents the maximum and minimum angular velocities achievable, |a v |≤a {max} Indicating that the linear acceleration is less than the maximum linear acceleration, indicating that the angular acceleration is less than the maximum angular acceleration;
converting the hard constraint into punishment in the objective function, and treating the hard constraint as a soft constraint; and (3) performing iterative optimization until the optimized track meets kinematic constraint and barrier constraint or reaches the iterative upper limit, and calculating to obtain a final planning path.
Based on the technical conception of the method, the application also provides a lightweight predictive navigation system suitable for the indoor environment, which comprises a pose prediction module, a dynamic and static feature extraction module, an occupied grid pattern sequence prediction module, a feasible space acquisition module, an initial track acquisition module and a track optimization module;
the pose prediction module is used for carrying out coordinate conversion on sensing measurement values on pose information, speed information and environment characterization of the intelligent agent obtained by the multi-sensor system, and projecting the environment variables to a coordinate system where the intelligent agent is located in the future;
the dynamic and static feature extraction module is used for taking into consideration the self-walking of the dynamic obstacle and the occupation of the static obstacle in the environment of the environment variable under the coordinate system of the future intelligent agent, capturing the interaction between the dynamic component and the static component, decoupling the dynamic component and the static component, and extracting the potential variable feature of the future probability occupation grid map;
the occupied raster pattern sequence prediction module adopts a variation automatic encoder to construct an OWM occupied raster pattern prediction generation module, potential variable characteristics of future probability occupied raster patterns are used as input, and a decoder outputs a predicted occupied raster pattern sequence;
the feasible space acquisition module is used for sequentially using each grid graph as a feasible space of the sampling candidate track;
the initial track acquisition module is used for generating candidate tracks in the feasible space and acquiring initial tracks meeting kinematic constraint and obstacle constraint;
the track optimization module adopts a graph optimization method to generate smoother and safer tracks based on the initial tracks.
In addition, a computer device is provided, which includes a processor and a memory, the memory is used for storing a computer executable program, the processor reads the computer executable program from the memory and executes the computer executable program, and when the processor executes the program, the lightweight predictive navigation method applicable to the indoor environment can be realized.
The application also provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and when the computer program is executed by a processor, the lightweight predictive navigation method applicable to indoor environment can be realized.
Compared with the prior art, the application has at least the following beneficial effects:
the application carries out environment prediction based on deep learning, is beneficial to self-making data sets and is suitable for different indoor environments. As the collected data of the environment is increased, a better prediction effect can be obtained; the frame is light, fast and universal, can be well applied to indoor mobile robots and unmanned vehicles, has low resource consumption and high instantaneity, and solves the problem of single-line laser prediction; the motion of the intelligent body is considered in the prediction work, and the motion information is necessary, so that the dynamic and static barriers can be distinguished, and priori information can be provided; the dynamic barrier is represented by the occupied state of the time sequence grid, the dynamic barrier and the static barrier can be processed simultaneously by planning, and the track can safely pass under the condition of multiple barriers.
Drawings
Fig. 1 is a technical framework of the present application.
Fig. 2 is a diagram of a prediction part framework.
Fig. 3 is a diagram of a network structure of a variant automatic encoder for prediction.
Fig. 4 is a plan part frame diagram.
Detailed Description
Exemplary examples of the application are set forth in detail below, with reference to the drawings and detailed description, wherein various details of embodiments of the application are included to facilitate understanding. It is to be understood that these examples are for the purpose of illustrating the application only and are not to be construed as limiting the scope of the application, since modifications to the application, which are various equivalent to those skilled in the art, will fall within the scope of the application as defined in the appended claims after reading the application.
Real-time navigation is an important component of the ability of an agent to safely drive in a complex dynamic environment, including prediction and motion planning. However, most of the existing methods require high quality sensor systems and huge computational consumption, which runs counter to real-time requirements. Furthermore, these methods lack a unified representation framework for prediction and planning, resulting in loss of accuracy in the representation conversion process. The application provides a prediction planning framework based on an occupied grid graph. The prediction module combines the perception information and the motion information to process dynamic and static obstacles and generate an occupancy grid pattern sequence. The planning module samples, selects and optimizes the candidate trajectories. Experimental results in multiple scenarios show that the proposed framework can improve real-time performance and security while reducing computational requirements compared to conventional approaches.
Fig. 1 is a schematic flow chart of a predictive navigation method based on deep learning, which is applicable to complex dynamic environments and has universality for indoor and outdoor environments. The object of the present application is to generate a series of future states containing pose and velocity information for an intelligent agent from measurements obtained from sensors. The application takes as input the characterization of the current environment by the sensor and the expected pose of the next time step, including initial environment information and potential motion variables, the expected output being a viable motion trajectory. The application is divided into two interrelated parts: environmental prediction and motion planning. The environmental prediction combines the obtained perception information with the motion information, and can generate future environmental prediction and mapping, which are expressed as an occupied grid graph sequence; and the motion planning takes a grid sequence with time attribute as a search space, solves the track into a nonlinear optimization problem, and solves the nonlinear optimization problem to obtain the track. The details of the application are as follows:
step 1: in the actual running process of the intelligent agent, the local coordinate system where the collected laser radar data is located is changed continuously along with the movement of the intelligent agent, and the prediction of the occupied grid graph is influenced. The motion compensation is divided into two steps, namely, the posture prediction and the coordinate transformation of the intelligent body. In the pose prediction process, the sensing measurement value coordinate conversion is carried out on the pose information, the speed information and the environment characterization of the intelligent agent obtained by the multi-sensor system, and the environment variable is projected to the coordinate system where the intelligent agent is located in the future, and the specific details are as follows:
the application can be illustrated using any sensor that can construct a probability grid pattern, such as a low cost sensor like a single line laser. Let the measurement of lidar be m= [ d ] t ,k t *δ]Wherein d is t Is the distance measured by the laser radar at the moment t, k t The range order of the laser radar is shown, and delta is the resolution of the laser radar and is shown by an angle.
(1) Firstly, calculating the current position occupied by the obstacle according to the measured value of the laser radar, wherein the position occupied by the obstacle is as follows under the coordinate system of the current intelligent agent:
[x,y]＝d t [cos(k t *δ),sin(k t *δ)] T
the linear velocity V at the moment t of the intelligent body can be obtained according to the inertial measurement unit t Angular velocity omega t Linear acceleration a vt Angular acceleration a ωt 。
(2) According to the current pose information [ x ] of the intelligent agent given by the positioning module ego ,y ego ,θ 1 ]The application presumes that the intelligent body performs uniform linear motion within the time interval delta t of 0.1s, and can infer the posture [ x 'of the intelligent body at the next moment' ego ,y′ ego ,θ 2 ]. Wherein x, y and theta respectively represent the abscissa and the ordinate of the position of the intelligent agent and the direction.
The conversion relationship between the current and next pose of the agent can be expressed as follows:
the current obstacle distribution is projected into a coordinate system of the pose of the intelligent agent after the duration of deltat, and the corresponding coordinate [ x ] of the obstacle occupying point can be obtained ′ ,y ′ ]The calculation method is as follows:
wherein [ x ] ′ ,y ′ ]Corresponding coordinates of points occupied by the obstacle can be obtained in a coordinate system of the pose of the intelligent body after the duration of deltat is passed, [ x ] ego ,y ego ]Is the position of the intelligent agent under the current global coordinate system, theta 1 Is oriented, [ x ] ′ ego ,y ′ ego ]Is the position of the intelligent agent in the global coordinate system after the duration of deltat, theta 2 Is oriented; and then obtaining the projection of the environment variable under the coordinate system of the future intelligent agent.
Step 2: and extracting dynamic and static characteristic information. The process of extracting dynamic-static feature information simultaneously considers the self-motion of the dynamic obstacle and the occupation of the static obstacle in the environment, implicitly captures the interaction between the two, and partly decouples the dynamic and static components to extract the potential variable features of the future probability occupation grid map.
(1) To predict dynamic grid information, the entire grid map information is fed into a convolved long and short time memory network module. The convolution long-short-time memory network module takes a one-dimensional occupancy grid graph as an input and generates a 32-dimensional dynamic grid feature by using a convolution kernel with the size of 3.
(2) Regarding the static obstacle information, predicting a local map of the static obstacle in the future by adopting a Bayesian generation method, and filtering out the dynamic information to obtain the local map information of the static obstacle. This approach allows for successful decoupling and feature extraction of dynamic and static information.
(3) The extracted dynamic grid features and the local map information of the static obstacle are input to an occupancy grid map prediction generation module.
Step 3: according to the application, an OGM occupation raster image prediction generation module is constructed by adopting a Variation Automatic Encoder (VAE), and the module obtains pre-trained network parameters through model training by the extracted dynamic raster features and a local map, predicts raster image sequences at future time and characterizes environment distribution. Reference is made to fig. 2 and 3.
(1) An encoder: the encoder section is composed of two convolutional layers and two residual layers. The first layer convolves features M with input dimensions of 32, output dimensions of 64, convolution kernel size of 4, step size of 2, and padding value of 1. The second layer convolves features with input dimensions of 64, output dimensions of 128, convolution kernel size of 4, step size of 2, and padding value of 1. The parameters of the two layers of residual layers are consistent, each residual layer inputs a feature X with 128 dimensions, the feature X is subjected to two-layer convolution treatment to obtain X2, and the output of the residual layers is X+X2; the first layer convolves input 128 dimensions, output 64 dimensions, convolution kernel size 3, step size 1. The second layer convolves input 64 dimensions, output 128 dimensions, convolution kernel size 1, step size 1. The output of the residual layer is then encoded to μ and σ via two structurally identical convolution blocks (input dimension 128, output dimension 2).
(2) Reconstructing μ and σ (dimension 16×16×2) into 512-dimensional vectors, and then generating distributions q (z|x): N (z|mu, z_var) and p (Z): N (z| 0,I), sampling the distribution q (z|x) to obtain the hidden vector Z in the variable automatic encoder.
(3) A decoder: the decoder section generates a feature of 128 dimensions by deconvolution of the normal distributed samples Z (input dimension 2, output dimension 128), and then inputs the feature to the VAE decoder. The decoder consists of two residual layers and two deconvolution layers, the residual layer structure is consistent with the structure in the encoder, 128-dimensional characteristics are input, and 128-dimensional characteristics are output. Then, through the first convolution layer, the characteristics of dimension 64 are output, the convolution kernel size is 4, and the step size is 2. Then the second convolution layer is input, the feature Q with the dimension of 64 is output, the convolution kernel size is 4, and the step length is 2.
(4) Q is input into an output convolution layer (the convolution kernel size is 3, the step size is 1), a predicted occupied raster pattern sequence with the dimension of 1 is output, and the predicted occupied raster pattern sequence is delivered to a motion planning module for processing.
Based on the network structure, the application performs model training by applying the whole training set, and mixes the loss function based on KL divergence and cross entropy:
n is the number of samples, C is the number of classifications, y ij Indicating whether the real label of the ith sample is the jth class (if 1, if not 0), p ij Representing the prediction probability that the ith sample belongs to the jth class. The application performs 60 rounds of training, each round traversing the entire training set. The gradient descent method selects Adam, the batch_size is set to 64, and the learning rate is set to be continuously reduced along with the increase of the training round number. After 60 rounds of training, the loss function gradually decreases and tends to stabilize.
Step 4: the present application sequentially uses each raster pattern as a viable space for sampling candidate trajectories after obtaining a predicted sequence of occupied raster patterns. A feasible spatial sequence is established that incorporates the temporal information into the local coordinate system of the agent. The present application defines the planning problem as an optimization problem, which can be solved by obtaining an initial solution and iterating the optimization, reference can be made to fig. 4.
To simplify the calculation, the obstacle is inflated with the agent radius as the inflation radius, and the intelligent agent is converted into particles for the purpose of calculating the collision and determining the trajectory of the agent.
By obtaining the extended sequence of the occupied raster pattern, the occupied space in the predicted sequence can be derived through a position transformation relation, wherein the position transformation relation is shown as the following formula:
the agent can be considered to be moving uniformly over a short time interval, with the centroid being the center of the circle, [ V ] max Δt,V min Δt]As a radius is the area that the agent can reach. Thus, the feasible spatial representation of each frame may be represented as follows: the states within the feasible interval must adhere to the strict speed interval constraints and collision constraints, and the final sampled trajectory must also remain within the feasible space. V (V) max And V min Respectively representing the maximum and minimum speeds of the agent,characterizing the spatial range that can be reached for Δt time driving at maximum speed, +.>Characterizing the spatial range, C, that can be reached by driving at minimum speed for Δt time obs Representing the space not occupied by the obstacle.
Step 5: after the feasible space is obtained, candidate track generation is performed in the feasible space. The application adopts a random sampling method to select frames in a feasible space. The application samples the expanded raster pattern sequence of step 4 individually one by one, with the sampled space representing the unoccupied area in each map. When sampling a subsequent map, the accessibility of the unoccupied grid is evaluated to ensure accessibility, and in order to satisfy the kinematic constraint of the agent, the search space is described as follows:
if the sampling points between adjacent frames can be connected without any collision, and the construction condition is satisfied: the connected edges do not fall outside the feasible space, and the dynamic constraint and the kinematic constraint are met between the two points, so that sampling points between adjacent frames meeting the requirements are connected together to form the edges. If the sample points do not meet these construction conditions, they are discarded, then the sample point list is traversed to identify the lowest cost node, and a backtracking process is initiated to obtain the initial trajectory as the initial solution.
Step 6: after the initial trajectory satisfying both the kinematic constraint and the obstacle constraint is obtained, a smoother, safer trajectory is generated by using the graph optimization method. The state transition of the agent at any given time satisfies the following conditions:
the evaluation functions to be optimized include lateral and longitudinal distance deviations from the reference trajectory, direction deviations, linear velocity deviations and angular velocity deviations. Optimizing evaluation index L ({ P) t -two parts): the first part represents the reference trajectory { P } obtained with the global path ref Deviation of the current path, while the second part reflects the smoothness of the current path: l ({ P) t })＝L offset ({P t }；{P ref })+L smooth ({P t }). Wherein L ({ P) t Cost of the current path, L }) represents offset ({P t }；{P ref }) representation and reference trajectory { P } ref Offset of L smooth ({P t -j) represents the smoothness of the track.
Dynamic constraints can be expressed to ensure that the acceleration and velocity of the agent obeys certain constraints, and finally, the path planning problem is formulated as a nonlinear optimization problem:
wherein L ({ P) t }|{m t:t+β }) represents optimization objective, { m t:t+β "represents a raster pattern sequence, C ego Represents the outline of the bicycle, C obs Represents the outline of the obstacle and,represents the maximum and minimum speeds achievable, ω min ,ω {max} Represents the maximum and minimum angular velocities achievable, |a v |≤a {max} Indicating that the linear acceleration is less than the maximum linear acceleration, indicating that the angular acceleration is less than the maximum angular acceleration.
Considering the challenges associated with solving the non-convex constraint problem, a hard constraint is treated as a soft constraint by converting it into a penalty within the objective function. If the optimized track can not meet the kinematic constraint and the obstacle constraint, repeating the optimizing process of the step to further optimize until reaching the upper limit of iteration, ending the solving process and outputting the final planning path.
The application verifies on the simulation platform based on Gazebo software, and constructs the indoor environment simulation platform comprising a corridor, a table and a chair by using the Gazebo software to construct the simulation environment for corresponding to various indoor environments. Each round of test adopts a random starting point and a random movement speed, the starting positions are randomly distributed on barriers outside 0.5m around the intelligent body, the number of the barriers and the maximum and minimum speeds can be set, when the distance between the barriers and the intelligent body is smaller than 0.05m, the collision is considered to happen, and if the collision happens, the current intelligent body test round is considered to be finished. If the end point is normally reached, the current intelligent agent testing round is regarded as ending.
Based on the above test requirements, the test parameters adopted by the application are shown in the following table:
the final test results are shown in the following table:
based on the conception of the method, the application also provides a lightweight predictive navigation system suitable for indoor environment, which comprises a pose prediction module, a dynamic and static feature extraction module, an occupied grid pattern sequence prediction module, a feasible space acquisition module, an initial track acquisition module and a track optimization module;
the pose prediction module is used for carrying out coordinate conversion on sensing measurement values on pose information, speed information and environment characterization of the intelligent agent obtained by the multi-sensor system, and projecting the environment variables to a coordinate system where the intelligent agent is located in the future;
the dynamic and static feature extraction module is used for taking into consideration the self-walking of the dynamic obstacle and the occupation of the static obstacle in the environment of the environment variable under the coordinate system of the future intelligent agent, capturing the interaction between the dynamic component and the static component, decoupling the dynamic component and the static component, and extracting the potential variable feature of the future probability occupation grid map;
the occupied raster pattern sequence prediction module adopts a variation automatic encoder to construct an OWM occupied raster pattern prediction generation module, potential variable characteristics of future probability occupied raster patterns are used as input, and a decoder outputs a predicted occupied raster pattern sequence;
the feasible space acquisition module is used for sequentially using each grid graph as a feasible space of the sampling candidate track;
the initial track acquisition module is used for generating candidate tracks in the feasible space and acquiring initial tracks meeting kinematic constraint and obstacle constraint;
the track optimization module adopts a graph optimization method to generate smoother and safer tracks based on the initial tracks.
On the other hand, the application also provides a computer readable storage medium, and the computer readable storage medium stores a computer program, and when the computer program is executed by a processor, the lightweight predictive navigation method applicable to the indoor environment can be realized.
The computer device may be a notebook computer, desktop computer, workstation or vehicle computer.
For the processor of the present application, it may be a Central Processing Unit (CPU), a Graphics Processor (GPU), a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), or an off-the-shelf programmable gate array (FPGA).
The memory can be an internal memory unit of a notebook computer, a desktop computer, a workstation or a vehicle-mounted computer, such as a memory and a hard disk; external storage units such as removable hard disks, flash memory cards may also be used.
The application also provides a computer device, which comprises a processor and a memory, wherein the memory is used for storing a computer executable program, the processor reads the computer executable program from the memory and executes the computer executable program, and the lightweight predictive navigation method applicable to the indoor environment can be realized when the processor executes the computer executable program.
Computer readable storage media may include computer storage media and communication media. Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. The computer readable storage medium may include: read Only Memory (ROM), random access Memory (RAM, random Access Memory), solid state disk (SSD, solid State Drives), or optical disk, etc. The random access memory may include resistive random access memory (ReRAM, resistance Random Access Memory) and dynamic random access memory (DRAM, dynamic Random Access Memory), among others.
In summary, the lightweight predictive navigation method suitable for indoor environments adopts a layered prediction-planning method, obtains sensing data and motion data (speed and coordinates) of a vehicle per se under a unified environment representation form, partially decouples dynamic and static environments, obtains future positions through coordinate transformation and mapping of the static environments, predicts grid maps of a plurality of steps in the future by extracting scene motion characteristic information, directly utilizes future grids, and simultaneously processes dynamic and static obstacles, thereby improving the efficiency and effectiveness of track optimization.
The method can carry out long-time movement planning of the intelligent agent under a unified characterization form, and is a difficult task. Grid maps are a compact, common method of environmental characterization, with similar characterization as in top view, suitable for unstructured environments. The grid map can represent the feasible space required by planning, has abstract and high reusability, and can be directly utilized by planning. The benefit of occupying the mesh map is that it is possible to represent objects of arbitrary shape and to estimate the motion of the object without explicit data correlation. The method of the application uses probability occupancy trellis diagram as a comprehensive representation to prevent information loss during conversion. The application organizes the predicted occupancy grid graph sequence in time order to generate trajectories, thereby mitigating the dependence of the planning process on prior prediction information. Regarding the aspect of prediction, the challenges of poor real-time performance and inaccurate prediction result are solved by sensing through the single-line laser radar and combining with motion information, and in addition, the calculation speed is increased by adopting a light structure. With respect to the planning problem, a feasible trajectory is generated based on the occupancy raster pattern sequence while taking into account future distributions of obstacles, thereby solving the short-term planning problem. Furthermore, the obtained trajectory is optimized to ensure smoothness and reduce the discontinuity of the trajectory.
It should be noted that, the above description is only for illustrating the specific embodiments of the present application, but the scope of the present application is not limited thereto, and those skilled in the art should understand that, based on the technical solution of the present application, modifications or variations made according to the technical solution of the present application and the inventive concept thereof should be covered in the scope of the present application.

Claims (10)

1. The lightweight predictive navigation method suitable for the indoor environment is characterized by comprising the following steps of:
the method comprises the steps of performing coordinate conversion of sensing measurement values on pose information, speed information and environmental characterization of an intelligent agent obtained by a multi-sensor system, and projecting environmental variables to a coordinate system where the intelligent agent is located in the future;
taking into consideration the self-walking of the dynamic barrier and the occupation of the static barrier in the environment of the environment variable under the coordinate system of the future intelligent agent, capturing the interaction between the dynamic barrier and the static barrier, decoupling the dynamic component and the static component, and extracting the potential variable characteristics of the future probability occupation grid map;
constructing an OGM occupation raster pattern prediction generation module by adopting a variation automatic encoder, taking potential variable characteristics of a future probability occupation raster pattern as input, and outputting a predicted occupation raster pattern sequence by a decoder;
sequentially using each raster pattern as a feasible space for sampling candidate trajectories;
generating candidate tracks in the feasible space, and acquiring initial tracks meeting kinematic constraint and barrier constraint;
a smoother and safer trajectory is generated based on the initial trajectory using a graph optimization method.
2. The lightweight predictive navigation method for an applicable indoor environment according to claim 1, wherein projecting the environmental variable into a coordinate system in which the future agent is located by performing coordinate transformation of the sensed measurement values of the agent pose information, the velocity information, and the environmental characterization obtained by the multi-sensor system comprises:
firstly, calculating the current obstacle occupation position according to the measured value of the laser radar, and acquiring the obstacle occupation position under the coordinate system of the current intelligent agent;
deducing the pose of the intelligent agent at the next moment according to the conversion relation between the current pose and the next moment of the intelligent agent and the current pose information of the intelligent agent; and projecting the current obstacle distribution to a coordinate system of the pose of the intelligent agent after the set time length, and obtaining the projection of the environment variable under the coordinate system of the intelligent agent in the future.
3. The lightweight predictive navigation method for an applicable indoor environment according to claim 1, wherein taking into account the occupancy of the environment by the self-moving and static obstacles of the dynamic obstacle, capturing the interaction between the two, and decoupling the dynamic and static components, extracting the potential variable features of the future probability occupancy trellis diagram comprises the steps of:
feeding the whole grid map information into a convolution long-short time memory network module, wherein the convolution long-short time memory network module takes a one-dimensional occupied grid graph as input, and generates dynamic grid features by using a convolution kernel with the size of 3;
predicting a local map of the static obstacle in the future by adopting a Bayesian generation method, and filtering out dynamic information to obtain local map information of the static obstacle;
and taking the extracted dynamic grid characteristics and the local map information of the static obstacle as the input of the OWM occupation grid map prediction generation module.
4. The lightweight predictive navigation method applicable to indoor environments according to claim 1, wherein the OGM occupying raster image prediction generation module comprises an encoder and a decoder, the encoder is composed of two convolution layers and two residual layers, and parameters of the two residual layers are consistent; the decoder consists of two residual layers and two deconvolution layers, the residual layer structure is identical to that in the encoder.
5. The lightweight predictive navigation method for an applicable indoor environment according to claim 1, wherein when each raster pattern is sequentially used as a feasible space for sampling candidate trajectories, specifically, based on an extended sequence of occupied raster patterns, the occupied space in the predicted sequence is derived through a position transformation relationship, in a set period of time, an agent uniformly moves, a centroid is a center of a circle, a feasible space representation of each frame is obtained through an area that the agent can reach, and states in the feasible interval obey a strict speed interval constraint and a collision constraint.
6. The lightweight predictive navigation method applicable to indoor environments according to claim 1, wherein candidate tracks are generated in the feasible space, frames are selected in the feasible space by adopting a random sampling method when initial tracks meeting kinematic constraint and barrier constraint are obtained, grid graph sequences are sampled individually and one by one, the sampled space represents an unoccupied area in each graph, and a search space is obtained on the basis of meeting the kinematic constraint of an intelligent agent and ensuring accessibility during sampling; when sampling is carried out, if sampling points between adjacent frames can be connected under the condition of no collision, judging whether the connected edges meet the dynamic constraint and the kinematic constraint of an intelligent agent, and if so, connecting the sampling points together to form the edges; traversing the node list to identify the lowest cost node and initiating a backtracking process to obtain an initial trajectory that satisfies the kinematic and obstacle constraints.
7. The lightweight predictive navigation method for an applicable indoor environment of claim 1, wherein generating a smoother, safer trajectory based on the initial trajectory using a graph optimization method comprises:
creating smoother and safer trajectories using the graph optimization method is formulated as a nonlinear optimization problem:
wherein L ({ P) t }|{m t:t+β }) represents optimization objective, { m t:t+β "represents a raster pattern sequence, C ego Represents the outline of the bicycle, C obs Represents the outline of the obstacle and,represents the maximum and minimum speeds achievable, ω min ,ω {max} Represents the maximum and minimum angular velocities achievable, |a v |≤a {max} Indicating that the linear acceleration is less than the maximum linear acceleration, indicating that the angular acceleration is less than the maximum angular acceleration;
converting the hard constraint into punishment in the objective function, and treating the hard constraint as a soft constraint; and (3) performing iterative optimization until the optimized track meets kinematic constraint and barrier constraint or reaches the iterative upper limit, and calculating to obtain a final planning path.
8. The lightweight predictive navigation system suitable for the indoor environment is characterized by comprising a pose prediction module, a dynamic and static feature extraction module, an occupied grid pattern sequence prediction module, a feasible space acquisition module, an initial track acquisition module and a track optimization module;
the pose prediction module is used for carrying out coordinate conversion on sensing measurement values on pose information, speed information and environment characterization of the intelligent agent obtained by the multi-sensor system, and projecting the environment variables to a coordinate system where the intelligent agent is located in the future;
the dynamic and static feature extraction module is used for taking into account the occupation of the dynamic barrier and the static barrier in the environment by the self-walking of the environment variable under the coordinate system of the future intelligent agent, capturing the interaction between the dynamic component and the static component, decoupling the dynamic component and the static component, and extracting the potential variable feature of the future probability occupation grid map;
the occupied raster pattern sequence prediction module adopts a variation automatic encoder to construct an OWM occupied raster pattern prediction generation module, potential variable characteristics of future probability occupied raster patterns are used as input, and a decoder outputs a predicted occupied raster pattern sequence;
the feasible space acquisition module is used for sequentially using each grid graph as a feasible space of the sampling candidate track;
the initial track acquisition module is used for generating candidate tracks in the feasible space and acquiring initial tracks meeting kinematic constraint and obstacle constraint;
the track optimization module adopts a graph optimization method to generate smoother and safer tracks based on the initial tracks.
9. A computer device comprising a processor and a memory, the memory storing a computer executable program, the processor reading the computer executable program from the memory and executing the program, the processor executing the program to implement the lightweight predictive navigation method of any one of claims 1-7 for use in an indoor environment.
10. A computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, the computer program, when executed by a processor, is capable of implementing the lightweight predictive navigation method of any one of claims 1-7, which is applicable to indoor environments.
CN202310965884.3A 2023-08-02 2023-08-02 Lightweight predictive navigation method and system suitable for indoor environment Pending CN116858253A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310965884.3A CN116858253A (en) 2023-08-02 2023-08-02 Lightweight predictive navigation method and system suitable for indoor environment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310965884.3A CN116858253A (en) 2023-08-02 2023-08-02 Lightweight predictive navigation method and system suitable for indoor environment

Publications (1)

Publication Number Publication Date
CN116858253A true CN116858253A (en) 2023-10-10

Family

ID=88227022

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310965884.3A Pending CN116858253A (en) 2023-08-02 2023-08-02 Lightweight predictive navigation method and system suitable for indoor environment

Country Status (1)

Country Link
CN (1) CN116858253A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117349545A (en) * 2023-12-04 2024-01-05 中国电子科技集团公司第五十四研究所 Target space-time distribution prediction method based on environment constraint grid

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117349545A (en) * 2023-12-04 2024-01-05 中国电子科技集团公司第五十四研究所 Target space-time distribution prediction method based on environment constraint grid

Similar Documents

Publication Publication Date Title
US11970161B2 (en) Apparatus, method and article to facilitate motion planning of an autonomous vehicle in an environment having dynamic objects
CN109782763B (en) Mobile robot path planning method in dynamic environment
JP7086111B2 (en) Feature extraction method based on deep learning used for LIDAR positioning of autonomous vehicles
CN111142557B (en) Unmanned aerial vehicle path planning method and system, computer equipment and readable storage medium
US20220057803A1 (en) Apparatus, method and article to facilitate motion planning in an environment having dynamic objects
CN112015847B (en) Obstacle trajectory prediction method and device, storage medium and electronic equipment
KR102695522B1 (en) Method and device to train image recognition model and to recognize image
CN107610235B (en) Mobile platform navigation method and device based on deep learning
CN116858253A (en) Lightweight predictive navigation method and system suitable for indoor environment
CN113885567B (en) Collaborative path planning method for multiple unmanned aerial vehicles based on conflict search
CN113515131B (en) Mobile robot obstacle avoidance method and system based on condition variation automatic encoder
CN116295415A (en) Map-free maze navigation method and system based on pulse neural network reinforcement learning
CN116628448A (en) Sensor management method based on deep reinforcement learning in extended target
CN117232522A (en) Robot crowd navigation method and system based on space-time interaction diagram and dangerous area
Srinivasan et al. Beyond RMSE: Do machine-learned models of road user interaction produce human-like behavior?
CN114326810A (en) Obstacle avoidance method of unmanned aerial vehicle in complex dynamic environment
CN116448134B (en) Vehicle path planning method and device based on risk field and uncertain analysis
Doellinger et al. Environment-aware multi-target tracking of pedestrians
CN117197767A (en) Vehicle track prediction method, device, electronic equipment and storage medium
JP2021196632A (en) Prediction device, prediction method, program and vehicle control system
Neloy et al. Alpha-N-V2: Shortest path finder automated delivery robot with obstacle detection and avoiding system
WO2022099526A1 (en) Method for training lane change prediction regression model, and lane change predicton method and apparatus
CN115903773A (en) Mobile object control device, mobile object, learning device and method, and storage medium
EP4339051A1 (en) Driving strategy determination method and apparatus, device, and vehicle
JP2023066619A (en) Control device, control method and control program

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