Control method for positioning and navigation of intelligent mobile robot

Abstract

The invention relates to a control method for positioning and navigation of an intelligent mobile robot. The intelligent mobile robot positioning and navigation control method comprises the steps that a first path is obtained, and the first path is formed based on global map data; obtaining multi-mode real-time observation data; jointly forming a multi-modal constraint function based on the multi-modal real-time observation data; acquiring a real-time pose of the intelligent mobile robot according to the multi-modal constraint function; and generating a second path based on the multi-modal real-time observation data and the real-time pose in combination with the target deep reinforcement learning model. The control method for positioning and navigation of the intelligent mobile robot has the advantages that the overall navigation precision and robustness of the intelligent mobile robot can be improved, the method is suitable for complex dynamic scenes, and local dynamic path planning can be effectively carried out.

Description

Technical Field

[0001] The present invention relates to the technical field of mobile robots, and in particular to a control method for positioning and navigating an intelligent mobile robot. Background Art

[0002] With the development of science and technology, in order to reduce the intensity and workload of human labor, the field of intelligent mobile robots has attracted much attention. Mobile robot positioning and navigation technology has always received widespread attention due to its comprehensiveness and practicality. It provides a basis for intelligent mobile robots to complete various complex tasks, and also enables robots to be widely used in various scenarios such as transportation and cleaning, and active in production and life.

[0003] However, existing mobile robot positioning and navigation technologies mostly focus on specific types of sensors, specific modules or specific scenarios. They not only lack the ability to analyze and process multimodal sensor data, resulting in the inability to identify their specific location in the map; they also lack the ability to perform real-time intelligent local planning and even interact with dynamic objects, making it difficult for mobile robots to move to the target area safely and efficiently. Summary of the invention

[0004] Based on this, the purpose of this application is to provide a control method for positioning and navigation of an intelligent mobile robot, so that the intelligent mobile robot can move to a target area safely and efficiently.

[0005] In a first aspect, the present application provides a method for controlling positioning and navigation of an intelligent mobile robot, comprising:

[0006] A first path is obtained, where the first path is formed based on global map data, and the first path is used to characterize a global moving trajectory of the intelligent mobile robot and an initial position of the intelligent mobile robot.

[0007] Multimodal real-time observation data is acquired, wherein the multimodal real-time observation data includes landmark information, and the landmark information is used to characterize the real-time behavior control instructions of the intelligent mobile robot.

[0008] A multimodal constraint function is formed based on the multimodal real-time observation data.

[0009] The real-time position and pose of the intelligent mobile robot are obtained according to the multimodal constraint function.

[0010] Based on multimodal real-time observation data and real-time posture, combined with the target deep reinforcement learning model, a second path is generated, which is used to characterize the local moving trajectory of the intelligent mobile robot and the real-time posture of the intelligent mobile robot.

[0011] Further, the multi-modal real-time observation data includes inertial navigation information and landmark information;

[0012] A multimodal constraint function is formed based on multimodal real-time observation data, including the following steps:

[0013] Acquire a first constraint function according to the landmark information and the first path;

[0014] Acquire a second constraint function through an extended Kalman filter algorithm according to the inertial navigation information;

[0015] A multimodal constraint function is constructed based on the first constraint function and the second constraint function.

[0016] Furthermore, the real-time position and posture of the intelligent mobile robot is obtained according to the multimodal constraint function, including the steps of:

[0017] Obtaining the state space distribution of the intelligent mobile robot according to the multimodal constraint function;

[0018] According to the state space distribution, weighted particles are collected and formed into a first particle set, which is used to characterize the real-time position and posture of the intelligent mobile robot;

[0019] After weight normalization processing is performed on the first particle set, resampling and KLD sampling are performed alternately to generate a second particle set, and the second particle set meets the preset KLD sampling termination judgment criterion;

[0020] The real-time pose is obtained based on the state information and weight information of all particles in the second particle set.

[0021] Further, based on the multimodal real-time observation data and the real-time posture, combined with the target deep reinforcement learning model, a second path is generated, including the steps of:

[0022] Generate local map information based on real-time observation data and real-time posture;

[0023] Based on the local map information, real-time behavior control instructions are obtained, and the real-time behavior control instructions are used to control the intelligent mobile robot to realize path planning corresponding to the real-time posture;

[0024] According to the real-time behavior control instructions, the second path is obtained in combination with the target deep reinforcement learning model. The second path is used to characterize the local moving trajectory of the intelligent mobile robot and the real-time position and posture of the intelligent mobile robot.

[0025] Further, obtaining real-time behavior control instructions based on local map information includes the following steps:

[0026] Inputting the local map information into the target first machine learning model for classification processing to obtain a first behavior control instruction, the first behavior control instruction including a complex environment behavior control instruction, the complex environment behavior control instruction corresponding to a navigation scenario that cannot be processed by the target first machine learning model;

[0027] Updating the real-time behavior control instruction to the first behavior control instruction;

[0028] In response to obtaining the complex environment behavior control instruction, sending the local map information to the target second machine learning model, the target second machine learning model is used to generate a second behavior control instruction corresponding to the navigation scene that the target first machine learning model cannot handle;

[0029] In response to obtaining the second behavior control instruction generated by the target second machine learning model, the real-time behavior control instruction is updated to the second behavior control instruction.

[0030] Further, the landmark information includes movable object information;

[0031] The target deep reinforcement learning model includes the optimal action strategy function determined by the reward function and the value function. The expression of the optimal action strategy function is as follows:

[0032]

[0033] In the formula, a represents the optional action of the intelligent mobile robot, a t represents the optimal action of the intelligent mobile robot at time t, s t represents the state of the intelligent mobile robot at time t, represents the state of the intelligent mobile robot and the landmark at time t, represents the state of the intelligent mobile robot and the landmark after time t, is the optimal action strategy function, R(s t , a) is the reward function term used to evaluate the real-time state, γ is the discount factor, is the value function term used to evaluate future states.

[0034] Further, based on the multimodal real-time observation data and combined with the target deep reinforcement learning model, a second path is generated, including the steps of:

[0035] Based on multimodal real-time observation data, the expected spatial model of movable objects is established through the two-dimensional asymmetric Gaussian formula. The expression of the two-dimensional asymmetric Gaussian formula is as follows:

[0036]

[0037] In the formula, μ x is the mean value in the x direction, μ y is the mean in the y direction, σ x is the standard deviation in the x direction, σ y is the standard deviation in the y direction;

[0038] Based on the expected space model of the movable object, the state space of the intelligent mobile robot and the landmark is constructed, and the state space of the intelligent mobile robot and the landmark includes the state of the intelligent mobile robot and the landmark at time t and the state of the intelligent mobile robot and the landmark after time t;

[0039] Based on the state space and multimodal real-time observation data, combined with the target deep reinforcement learning model, the second path is generated.

[0040] In a second aspect, the present application also provides a control device for positioning and navigation of an intelligent mobile robot, comprising:

[0041] A first path acquisition module, used to acquire a first path, the first path is formed based on global map data, and the first path is used to represent the global moving trajectory of the intelligent mobile robot and the initial position of the intelligent mobile robot;

[0042] A real-time observation module is used to obtain multi-modal real-time observation data, the multi-modal real-time observation data includes landmark information, and the landmark information is used to represent the real-time behavior control instructions of the intelligent mobile robot;

[0043] A multimodal data processing module, used to jointly form a multimodal constraint function based on multimodal real-time observation data;

[0044] A real-time positioning module is used to obtain the real-time position and posture of the intelligent mobile robot according to the multimodal constraint function;

[0045] The second path acquisition module is used to generate a second path based on multimodal real-time observation data and real-time posture, combined with a target deep reinforcement learning model. The second path is used to characterize the local movement trajectory of the intelligent mobile robot and the real-time posture of the intelligent mobile robot.

[0046] In a third aspect, the present application further provides a computer device, including:

[0047] one or more processors;

[0048] A storage device for storing one or more programs;

[0049] When one or more programs are executed by one or more processors, the one or more processors implement the steps of the above-mentioned intelligent mobile robot positioning and navigation control method.

[0050] In a fourth aspect, a computer-readable storage medium stores a computer program thereon, characterized in that when the program is executed by a processor, the steps of the above-mentioned intelligent mobile robot positioning and navigation control method are implemented.

[0051] The intelligent mobile robot positioning and navigation control method provided in the present application can provide high-precision navigation results by acquiring and processing multimodal real-time observation data, and acquiring the real-time position and posture of the intelligent mobile robot based on a multimodal constraint function formed by the multimodal real-time observation data. The method of fusing multimodal real-time observation data can effectively deal with errors, faults and other problems that cannot be handled by single-mode positioning and navigation methods, thereby improving the overall navigation accuracy and robustness of the intelligent mobile robot.

[0052] In addition, the control method for positioning and navigation of the intelligent mobile robot provided in this application, combined with the target deep reinforcement learning model, generates a second path for characterizing the local movement trajectory and real-time posture of the intelligent mobile robot, which can effectively perform local dynamic path planning for the robot and improve the effect and efficiency of path planning, and is particularly suitable for complex dynamic scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1 A schematic diagram of an implementation environment for a method for controlling positioning and navigation of an intelligent mobile robot provided by an embodiment of the present application;

[0055] Figure 2 A flowchart of a method for controlling positioning and navigation of an intelligent mobile robot provided by one embodiment of the present application;

[0056] Figure 3 A schematic diagram of a process for obtaining the real-time position and posture of an intelligent mobile robot provided in one embodiment of the present application;

[0057] Figure 4 A schematic diagram of a process flow of local path planning for an intelligent mobile robot provided in one embodiment of the present application;

[0058] Figure 5 A schematic diagram of a process flow of local path planning for an intelligent mobile robot provided in one embodiment of the present application;

[0059] Figure 6 A schematic diagram of the architecture of an intelligent mobile robot provided for one embodiment of the present application;

[0060] Figure 7 A schematic diagram of an implementation environment for a method for controlling positioning and navigation of an intelligent mobile robot provided by an embodiment of the present application;

[0061] Figure 8 A schematic diagram of the structure of a control device for positioning and navigating an intelligent mobile robot provided in one embodiment of the present application. DETAILED DESCRIPTION

[0062] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0063] It should be clear that the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the embodiments of the present application.

[0064] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0065] When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0066] First, a brief introduction is given to the terms involved in the embodiments of the present application. The following related technologies can be combined arbitrarily with the technical solutions of the embodiments of the present application as optional solutions, and they all belong to the protection scope of the embodiments of the present application. The embodiments of the present application include at least part of the following contents.

[0067] Landmarks refer to points or objects with significant features in the environment, which can be used as references for navigation and positioning. In the fields of robot navigation, autonomous vehicles, aerospace, etc., landmarks are used to help determine location, plan paths or build maps.

[0068] Deep Reinforcement Learning (DRL) is a technology that combines deep learning and reinforcement learning to enable intelligent agents to learn optimal strategies through interaction with the environment in complex environments.

[0069] Multimodal constraint functions, in robot navigation, usually refer to using data from multiple sensors to provide more accurate and reliable environmental information, thereby helping the robot to perform positioning and path planning.

[0070] Next, in conjunction with the above explanations, the implementation environment of the embodiments of the present application is described. Figure 1 The implementation environment includes an intelligent mobile robot 101 and a landmark 102.

[0071] The intelligent mobile robot 101 may be an independent physical intelligent mobile robot, a robot cluster composed of multiple physical intelligent mobile robots, or a virtual intelligent mobile robot in a virtual environment.

[0072] Illustratively, the intelligent mobile robot 101 can achieve the function of reaching a target position via a safe and efficient path in a complex environment by executing relevant behavior control instructions.

[0073] Landmark 102 may be a natural landmark or an artificial landmark. It may be a building such as a house, a bridge, a tower, or a road sign such as a traffic sign or a preset positioning image. Schematically, landmark 102 includes an object or geographical feature on the ground that has obvious features and is easy to identify and locate.

[0074] Schematically, the intelligent mobile robot 101 can plan the most optimized moving route according to the distribution and location of the landmarks 102 to avoid obstacles or dangerous areas.

[0075] Please refer to Figure 1 and Figure 2 , which shows a method for controlling positioning and navigation of an intelligent mobile robot provided by an exemplary embodiment of the present application, and takes the method applied to the intelligent mobile robot 101 in the above implementation environment as an example for explanation. In the embodiment of the present application, the method includes the following steps:

[0076] Step 201: Obtain a first path.

[0077] Specifically, the intelligent mobile robot 101 can obtain a first path through a communication network. The first path can be formed based on global map data. The first path is used to characterize the global movement trajectory of the intelligent mobile robot and the initial position of the intelligent mobile robot.

[0078] Schematically, the intelligent mobile robot 101 can obtain a first path generated by a server via a communication network through a communication module. The first path can be generated by the server through a corresponding first path generation application based on preset global map data. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server.

[0079] Step 202: Acquire multi-modal real-time observation data.

[0080] Specifically, the intelligent mobile robot 101 can obtain multimodal real-time observation data through the multimodal sensor system carried thereon. The multimodal real-time observation data includes landmark information used to characterize relevant characteristics of the landmark 102. The intelligent mobile robot 101 can obtain real-time behavior control instructions based on the landmark information, thereby optimizing the movement path and behavior pattern of the intelligent mobile robot 101.

[0081] Exemplarily, the multimodal real-time observation data may include the heading angle θ of the intelligent mobile robot 101 h , the intelligent mobile robot 101 can measure its heading angle through the electronic compass observation model, and the observation model expression of the electronic compass is as follows:

[0082]

[0083] In the formula, θ h is the heading angle of the intelligent mobile robot 101, v x is the speed of the intelligent mobile robot 101 in the X-axis direction, v y is the speed of the intelligent mobile robot 101 in the Y-axis direction, and λ is the measurement error of the electronic compass.

[0084] Exemplarily, the multimodal real-time observation data may include the angular velocity ω of the intelligent mobile robot 101 h , the intelligent mobile robot 101 can measure its angular velocity through a gyroscope, and the observation model expression of the gyroscope is as follows:

[0085]

[0086] In the formula, ω h is the angular velocity of the intelligent mobile robot 101, v x (t) is the speed-time function of the intelligent mobile robot 101 in the X-axis direction, v y (t) is the speed-time function of the intelligent mobile robot 101 in the Y-axis direction, σ is the drift error of the gyroscope, is the angle derivative.

[0087] Exemplarily, the multimodal real-time observation data may include the distance s of the intelligent mobile robot 101. h , the intelligent mobile robot 101 can measure its distance s by using an odometer h , the observation model expression of the odometer is as follows:

[0088]

[0089] In the formula, s h is the angular velocity distance of the intelligent mobile robot 101, v x is the speed of the intelligent mobile robot 101 in the X-axis direction, v y is the speed of the intelligent mobile robot 101 in the Y-axis direction, μ is the measurement error of the odometer, is the rate integral term.

[0090] Step 203: forming a multimodal constraint function based on the multimodal real-time observation data.

[0091] Specifically, the intelligent mobile robot 101 jointly forms a multimodal constraint function based on the acquired multimodal real-time observation data.

[0092] Schematically, possible expressions of multimodal constraint functions are as follows:

[0093]

[0094] Where W n is the real-time pose vector, T is the time integral operator, F[W n ] is the multi-modal constraint function, H[W n ] is the multimodal observation matrix item, V n is the observation error term, G(v x , v y ) is the visual modality item, is the visual modality error. Real-time pose vector W n The possible expressions are as follows:

[0095]

[0096] In the formula, s x is the distance in the X direction, v x is the velocity in the X direction, a x is the acceleration in the X direction; s y is the distance in the Y direction, v y is the velocity in the Y direction, a y is the acceleration in the Y direction, and θ is the heading angle.

[0097] Step 204: acquiring the real-time position and posture of the intelligent mobile robot according to the multimodal constraint function.

[0098] Specifically, the intelligent mobile robot 101 obtains the real-time position and posture of the intelligent mobile robot 101 according to the multimodal constraint function.

[0099] Schematically, if the data error in a certain mode is greater than a preset threshold, the intelligent mobile robot 101 can implement fault-tolerant control according to a preset strategy.

[0100] For example, if the data error in a certain modality is greater than a preset threshold, the intelligent mobile robot 101 can abandon the real-time observation data corresponding to the modality with a large data error according to a preset strategy, and obtain the real-time posture of the intelligent mobile robot 101 through the real-time observation data of the remaining modalities.

[0101] Step 205: Generate a second path based on the multimodal real-time observation data and the real-time posture in combination with the target deep reinforcement learning model.

[0102] Specifically, the intelligent mobile robot 101 generates a second path based on the real-time posture and multimodal real-time observation data in combination with the target deep reinforcement learning model. The second path is used to characterize the local movement trajectory of the intelligent mobile robot 101 and the real-time posture of the intelligent mobile robot 101.

[0103] Schematically, the intelligent mobile robot 101 can input the preprocessed real-time posture and multi-modal real-time observation data into the target deep reinforcement learning model based on the proximal policy optimization algorithm. The objective function expression of the proximal policy optimization algorithm is as follows:

[0104]

[0105] In the formula, L(s, a, ε old , ε) is the objective function, s is the distance of the intelligent mobile robot, a is the acceleration of the intelligent mobile robot, π ε (a|s) is the new strategy updated during the training process, is the old strategy updated during training, ε is the parameter of the new strategy, and ε old is the parameter of the old strategy, R(a|s) is the advantage function term, and the clip function is a restriction function that can constrain the ratio of the new and old strategies to [1-δ,1+δ].

[0106] In the embodiment of the present application, the real-time position and posture of the intelligent mobile robot 101 is determined by a multimodal constraint function formed based on the acquired multimodal real-time observation data, so as to improve the positioning and navigation accuracy of the intelligent mobile robot 101. In addition, by adopting the method of multimodal real-time observation data fusion, it is possible to effectively deal with the problems of errors and faults that are difficult to handle by the positioning and navigation methods using only single-modal data, and it is possible to increase the redundancy of the intelligent mobile robot positioning and navigation system, thereby achieving a better fault-tolerant control effect, thereby improving the overall navigation accuracy and robustness of the intelligent mobile robot 101.

[0107] In an embodiment of the present application, by utilizing a target deep reinforcement learning model to generate a second path for characterizing the moving trajectory and real-time posture of the intelligent mobile robot 101 in the local positioning and navigation space, the robot can be effectively planned for a local dynamic path, taking into full account the real-time changes in the environment in which the intelligent mobile robot 101 is located and various dynamic factors, such as the position and speed of moving obstacles, the behavior of other intelligent agents in the environment, etc.; the effect and efficiency of path planning can be significantly improved, ensuring that the robot can reach the target position safely and quickly in complex dynamic scenes; it can also enable the intelligent mobile robot 101 to flexibly respond to various emergencies and uncertainties, greatly improving the performance and reliability of the intelligent mobile robot 101 in practical applications, and is particularly suitable for complex dynamic scenes.

[0108] Please refer to Figure 3 , in one embodiment of the present application, the multimodal real-time observation data includes inertial navigation information and landmark information;

[0109] A multimodal constraint function is formed based on multimodal real-time observation data, including the following steps:

[0110] Step 301: Acquire a first constraint function according to landmark information and a first path.

[0111] Specifically, the intelligent mobile robot 101 obtains landmark information from the multimodal real-time observation data, and obtains the first constraint function in combination with the first path.

[0112] As shown, the sensor of the intelligent mobile robot 101 can collect the original image information of the landmark 102, and obtain the positioning information and navigation information carried by the landmark 102 based on the original image information. The positioning information is used to represent the real-time position and posture of the intelligent mobile robot 101 relative to the landmark 102, and the navigation information is used to represent the behavior control instructions that the intelligent mobile robot 101 should take after detecting the landmark 102. Based on the obtained positioning information carried by the landmark 102, the intelligent mobile robot 101 can obtain the first constraint function G(v x , v y ), the first constraint function G(v x , vy )The possible expressions are as follows:

[0113]

[0114] Where D is the distance between the intelligent mobile robot 101 and the target object, f s is the focal length of the visual sensor of the intelligent mobile robot 101, v g is the optical flow speed.

[0115] Optionally, the intelligent mobile robot 101 may obtain the first constraint function G(v x , v y ).

[0116] Step 311: Obtain a second constraint function according to the inertial navigation information by using an extended Kalman filter algorithm.

[0117] Specifically, the moving trajectory of the intelligent mobile robot 101 can be regarded as being in a two-dimensional plane. The intelligent mobile robot 101 obtains inertial navigation information through the inertial navigation system carried by it, and obtains the second constraint function through the extended Kalman filter algorithm.

[0118] For example, the inertial navigation system carried by the intelligent mobile robot 101 may include a gyroscope, an accelerometer, and a magnetometer. The intelligent mobile robot 101 may measure acceleration through the gyroscope. A possible model expression of the gyroscope is as follows:

[0119] ω s =ω g +Δω+∈ ω

[0120] In the formula, ω s is the angular velocity of the intelligent mobile robot 101, ω g is the measurement value of the gyroscope in the inertial navigation system, Δω is the measurement error of the gyroscope, ∈ ω is Gaussian white noise. The possible model expression of the accelerometer is as follows:

[0121] A s =A g +ΔA+∈ A

[0122] In the formula, A s is the acceleration matrix of the intelligent mobile robot 101, A g is the measurement matrix of the accelerometer in the inertial navigation system, ΔA is the measurement error of the accelerometer, ∈ A is Gaussian white noise. The possible model expression of the magnetometer is as follows:

[0123]

[0124] In the formula, B s is the magnetic field vector of the environment where the intelligent mobile robot 101 is located, is the posture transformation matrix, B g is the measurement matrix of the magnetometer in the inertial navigation system, ΔB is the measurement error of the magnetometer, ∈ B is Gaussian white noise.

[0125] Furthermore, the second constraint function is calculated by using an extended Kalman filter algorithm based on the inertial navigation information obtained by the inertial navigation system carried by the intelligent mobile robot 101. A possible model expression of the extended Kalman filter algorithm is as follows:

[0126]

[0127] Where W n ∈R N is the real-time pose vector, is the process noise, F(W n-1 , n-1) is the dynamic function term, G n ∈R M is the real-time observation vector, To measure noise, H(W n , n) is the dynamic measurement function term.

[0128] Step 321: construct a multimodal constraint function based on the first constraint function and the second constraint function.

[0129] Specifically, the intelligent mobile robot 101 constructs a multimodal constraint function based on the first constraint function and the second constraint function.

[0130] Furthermore, a joint check may be performed using the first constraint function and the second constraint function.

[0131] Illustratively, if the deviation of the first constraint function is greater than a preset threshold, it can be verified by the second constraint function; if the deviation of the second constraint function is greater than the preset threshold, it can be verified by the first constraint function.

[0132] Please refer to Figure 3 In one embodiment of the present application, obtaining the real-time position and posture of the intelligent mobile robot according to the multimodal constraint function includes the steps of:

[0133] Step 322, obtaining the state space distribution of the intelligent mobile robot according to the multimodal constraint function.

[0134] Specifically, the intelligent mobile robot 101 can obtain its state space distribution N(w, P) according to the multimodal constraint function, where w represents the mean vector of the state space distribution, and P represents the covariance matrix of the state space distribution.

[0135] Step 323, according to the state space distribution, weighted particles are collected and formed into a first particle set.

[0136] Specifically, the intelligent mobile robot 101 can obtain the initial state space distribution N0 (w0, P0) based on the state space distribution. At this time, the intelligent mobile robot 101 can collect k particles from the initial state space distribution N0 (w0, P0) and form a first particle set. The possible expression of the first particle set is as follows:

[0137]

[0138] In the formula, represents the i-th particle, where i is a positive integer not exceeding k.

[0139] Step 324 , performing weight normalization processing on the first particle set and then performing resampling and KLD sampling alternately to generate a second particle set.

[0140] Specifically, after the resampling process of each particle in the first particle set is completed, the intelligent mobile robot 101 performs KLD sampling, and then compares the number of particles obtained by sampling with the number of particles required by the KLD sampling criterion. If the number of particles does not meet the requirements of the KLD sampling criterion, the next particle will continue to be alternately resampled and KLD sampled; if the number of particles obtained by sampling a certain particle meets the requirements of the KLD sampling criterion, the remaining unsampled particles will be discarded, and the second particle set will be generated. b is the number of particles that meet the requirements of the KLD sampling criteria.

[0141] Schematically, a possible expression of the KLD sampling criterion is as follows:

[0142]

[0143] Where j is the number of resampled particles, ε is the difference parameter of KL distance in KLD sampling, and σ is the chi-square distribution quantile parameter.

[0144] Step 325: Acquire the real-time position and posture based on the state information and weight information of all particles in the second particle set.

[0145] Specifically, the intelligent mobile robot 101 can calculate the expected state of all particles in the second particle set based on the state information and weight information of all particles in the second particle set. The real-time position W of the intelligent mobile robot 101 can be obtained based on the state of all particles in the second particle set. n .

[0146] Please refer to Figure 4In one embodiment of the present application, based on multimodal real-time observation data and real-time posture, combined with the target deep reinforcement learning model, a second path is generated, including the steps of:

[0147] Step 401, generating local map information based on real-time observation data and real-time posture.

[0148] Specifically, the intelligent mobile robot 101 generates local map information based on real-time observation data and real-time posture. The local map information may be expressed as follows:

[0149]

[0150]

[0151] Where W n is the real-time position and posture of the intelligent mobile robot 101, is the landmark real-time pose matrix corresponding to the landmark information, is the real-time pose matrix of a single landmark.

[0152] Step 402: Acquire real-time behavior control instructions based on local map information.

[0153] Specifically, the intelligent mobile robot 101 obtains real-time behavior control instructions based on local map information.

[0154] Illustratively, if the intelligent mobile robot 101 recognizes that the local map information includes a landmark real-time pose matrix representing a preset real-time behavior control instruction Then obtain the corresponding real-time behavior control instructions.

[0155] For example, if the intelligent mobile robot 101 recognizes that the local map information includes a real-time pose matrix of a landmark that is not passable in this direction Get preset steering or stop control instructions.

[0156] For example, if the intelligent mobile robot 101 recognizes that the local map information includes a real-time pose matrix of a landmark that needs to be slowed down and passed Get the preset speed limit control instruction.

[0157] For example, if the intelligent mobile robot 101 recognizes that the local map information includes a real-time pose matrix of a landmark in a complex environment Get preset complex environment control instructions.

[0158] In an illustrative manner, if the intelligent mobile robot 101 recognizes the real-time position W in the local map information n The corresponding positioning component is the same as the real-time pose matrix of a single landmark If the square of the second norm of the corresponding positioning component is less than a preset threshold, a preset distance limit control instruction is obtained.

[0159] Indicatively, if the intelligent mobile robot 101 recognizes that the real-time posture corresponding to the positioning component in the local map information is the same as the real-time posture matrix of the landmark The mean of the square of the second norm of the corresponding positioning component of the real-time pose matrix of each single landmark is less than a preset threshold, and the preset complex environment control instruction is obtained.

[0160] Step 403: Acquire a second path according to the real-time behavior control instruction and the target deep reinforcement learning model.

[0161] Specifically, the mobile robot 101 inputs the local map information into the target deep reinforcement learning model according to the real-time behavior control instruction to obtain the second path.

[0162] Exemplarily, the target deep reinforcement learning model may include an iterable loss function, and the loss function expression after iteration d times is as follows:

[0163]

[0164] In the formula, a is used to characterize the action of the second path, S n is the current state represented by the local map information, S n+1 is the future state represented by the local map information, is the loss function, is the value function term, is the value function parameter after d iterations, Y d is the optimization objective term of the value function, Indicates the state S n , action a and next state S n+1 expectations.

[0165] Please refer to Figure 5 and Figure 6 In one embodiment of the present application, based on multimodal real-time observation data and real-time posture, combined with the target deep reinforcement learning model, a second path is generated, including the steps of:

[0166] Step 501, generating local map information based on real-time observation data and real-time posture.

[0167] Step 502 inputs the local map information into the target first machine learning model for classification processing to obtain the first behavior control index.

[0168] Specifically, the intelligent mobile robot 101 inputs the local map information obtained through its perception system 601 into the first control system 602 constructed based on the target first machine learning model for classification processing, and obtains the first behavior control instructions, which include complex environment behavior control instructions, and the complex environment behavior control instructions correspond to navigation scenarios that the target first machine learning model cannot handle.

[0169] Optionally, the target first machine learning model can be carried on the intelligent mobile robot 101.

[0170] Step 511: Update the real-time behavior control instruction to the first behavior control instruction.

[0171] Specifically, the intelligent mobile robot 101 updates the real-time behavior control instruction to the first behavior control instruction, and controls the execution system 603 to implement corresponding positioning and navigation control based on the real-time behavior control instruction.

[0172] Step 521, in response to obtaining the complex environment behavior control instruction, sending local map information to the target second machine learning model.

[0173] Specifically, in response to obtaining complex environment behavior control instructions, the intelligent mobile robot 101 sends local map information to the target second machine learning model, and the target second machine learning model is used to generate second behavior control instructions corresponding to navigation scenarios that the target first machine learning model cannot handle.

[0174] Optionally, the target second machine learning model can be installed on a server. The server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server.

[0175] Optionally, the target second machine learning model can be installed on a computing terminal. The computing terminal can be a computing device such as a mobile phone, a tablet computer, a portable computer, a desktop computer, an industrial computer or an all-in-one computer.

[0176] Optionally, the target second machine learning model can be carried on the intelligent mobile robot 101.

[0177] Step 522, in response to obtaining the second behavior control instruction generated by the target second machine learning model, the real-time behavior control instruction is updated to the second behavior control instruction.

[0178] Specifically, the intelligent mobile robot 101 updates the real-time behavior control instruction to the second behavior control instruction in response to obtaining the second behavior control instruction generated by the target second machine learning model, and controls the execution system 603 to implement corresponding positioning and navigation control based on the real-time behavior control instruction.

[0179] Step 531: input the real-time behavior control instruction into the target deep reinforcement learning model to obtain the second path.

[0180] In the embodiment of the present application, by setting up a first control system 602 and a second control system 604, the real-time performance and robustness of the intelligent mobile robot 101 are balanced through a multi-level control system, thereby avoiding the defects of poor algorithm real-time performance and long execution response time brought about by the hierarchical architecture of the traditional intelligent mobile robot.

[0181] Please refer to Figure 7 In one embodiment of the present application, a method for controlling positioning and navigation of an intelligent mobile robot includes:

[0182] Landmark information includes movable object information.

[0183] Optionally, the landmarks include static landmarks 702 and movable objects 703. The intelligent mobile robot 701 can construct a landmark real-time pose matrix including motion information based on multimodal real-time observation data. The optional expression of the landmark real-time pose matrix including motion information is as follows:

[0184]

[0185] In the formula, is the landmark real-time pose matrix including motion information, is the position component, For the movement component.

[0186] The target deep reinforcement learning model includes the optimal action strategy function determined by the reward function and the value function. The expression of the optimal action strategy function is as follows:

[0187]

[0188] In the formula, a represents the optional action of the intelligent mobile robot, a t represents the optimal action of the intelligent mobile robot at time t, s t represents the state of the intelligent mobile robot at time t, represents the state of the intelligent mobile robot and the landmark at time t, represents the state of the intelligent mobile robot and the landmark after time t, is the optimal action strategy function, R(s t , a) is the reward function term used to evaluate the real-time state, γ is the discount factor, is the value function term used to evaluate future states.

[0189] Please refer to Figure 7In one embodiment of the present application, based on multimodal real-time observation data and real-time posture, combined with the target deep reinforcement learning model, a second path is generated, including the steps of:

[0190] Based on the multi-modal real-time observation data, the expected spatial model of the movable object 703 is established by a two-dimensional asymmetric Gaussian formula. The expression of the two-dimensional asymmetric Gaussian formula is as follows:

[0191]

[0192] In the formula, μ x is the mean value in the x direction, μ y is the mean in the y direction, σ x is the standard deviation in the x direction, σ y is the standard deviation in the y direction;

[0193] Illustratively, the symmetry axis of the expected spatial model of the movable object 703 may be set to be parallel to the direction of its velocity vector. A possible expression for the symmetry axis of the expected spatial model of the movable object 703 is as follows:

[0194]

[0195] Schematically, the two-dimensional asymmetric Gaussian formula expression in which the x-direction and the y-direction are not independent is as follows:

[0196]

[0197] Where ρ is the correlation coefficient between the x-direction and the y-direction.

[0198] Furthermore, the symmetry axis of the expected spatial model of the movable object 703 may be set to be parallel to the direction of its velocity vector. A possible expression of the symmetry axis of the expected spatial model of the movable object 703 is as follows:

[0199]

[0200] In the formula, is the slope term, and μx is the intercept term.

[0201] The state space of the intelligent mobile robot and the landmark is constructed based on the expected space model of the movable object 703. The state space of the intelligent mobile robot and the landmark includes the state of the intelligent mobile robot and the landmark at time t and the state of the intelligent mobile robot and the landmark after time t.

[0202] Based on the state space and multimodal real-time observation data of the intelligent mobile robot and landmarks, the second path is generated in combination with the target deep reinforcement learning model.

[0203] Please refer to Figure 8The present application also provides a control device 800 for positioning and navigating an intelligent mobile robot, comprising:

[0204] A first path acquisition module 801 is used to acquire a first path, the first path is formed based on global map data, and the first path is used to represent the global moving trajectory of the intelligent mobile robot and the initial position of the intelligent mobile robot;

[0205] A real-time observation module 802 is used to obtain multi-modal real-time observation data, the multi-modal real-time observation data includes landmark information, and the landmark information is used to represent the real-time behavior control instructions of the intelligent mobile robot;

[0206] A multimodal data processing module 803, for jointly forming a multimodal constraint function based on multimodal real-time observation data;

[0207] A real-time positioning module 804 is used to obtain the real-time position and posture of the intelligent mobile robot according to the multimodal constraint function;

[0208] The second path acquisition module 805 is used to generate a second path based on multimodal real-time observation data and real-time posture, combined with a target deep reinforcement learning model. The second path is used to characterize the local movement trajectory of the intelligent mobile robot and the real-time posture of the intelligent mobile robot.

[0209] In an optional embodiment, the multimodal data processing module 803 may also be used to:

[0210] Acquire a first constraint function according to the landmark information and the first path;

[0211] Acquire a second constraint function through an extended Kalman filter algorithm according to the inertial navigation information;

[0212] A multimodal constraint function is constructed based on the first constraint function and the second constraint function.

[0213] In an optional embodiment, the real-time positioning module 804 may also be used to:

[0214] Obtaining the state space distribution of the intelligent mobile robot according to the multimodal constraint function;

[0215] According to the state space distribution, weighted particles are collected and formed into a first particle set, which is used to characterize the real-time position and posture of the intelligent mobile robot;

[0216] After weight normalization processing is performed on the first particle set, resampling and KLD sampling are performed alternately to generate a second particle set, and the second particle set meets the preset KLD sampling termination judgment criterion;

[0217] The real-time pose is obtained based on the state information and weight information of all particles in the second particle set.

[0218] In an optional embodiment, the second path acquisition module 805 may also be used to:

[0219] Generate local map information based on real-time observation data and real-time posture;

[0220] Based on the local map information, real-time behavior control instructions are obtained, and the real-time behavior control instructions are used to control the intelligent mobile robot to realize path planning corresponding to the real-time posture;

[0221] The real-time behavior control instructions are input into the target deep reinforcement learning model to obtain the second path, which is used to characterize the local moving trajectory of the intelligent mobile robot and the real-time position and posture of the intelligent mobile robot.

[0222] In an optional embodiment, the second path acquisition module 805 may also be used to:

[0223] Inputting the local map information into the target first machine learning model for classification processing to obtain a first behavior control instruction, the first behavior control instruction including a complex environment behavior control instruction, the complex environment behavior control instruction corresponding to a navigation scenario that cannot be processed by the target first machine learning model;

[0224] Updating the real-time behavior control instruction to the first behavior control instruction;

[0225] In response to obtaining the complex environment behavior control instruction, sending the local map information to the target second machine learning model, the target second machine learning model is used to generate a second behavior control instruction corresponding to the navigation scene that the target first machine learning model cannot handle;

[0226] In response to obtaining the second behavior control instruction generated by the target second machine learning model, the real-time behavior control instruction is updated to the second behavior control instruction.

[0227] In an optional embodiment, the second path acquisition module 805 may also be used to:

[0228] Based on multimodal real-time observation data, the expected spatial model of movable objects is established through the two-dimensional asymmetric Gaussian formula. The expression of the two-dimensional asymmetric Gaussian formula is as follows:

[0229]

[0230] In the formula, μ x is the mean value in the x direction, μ y is the mean in the y direction, σ x is the standard deviation in the x direction, σ y is the standard deviation in the y direction;

[0231] Based on the expected space model of the movable object, the state space of the intelligent mobile robot and the movable object is constructed, and the state space of the intelligent robot and the movable object includes the state of the intelligent robot and the movable object at time t and the state of the intelligent robot and the movable object after time t;

[0232] Based on the state space and multimodal real-time observation data, combined with the target deep reinforcement learning model, the second path is generated.

[0233] In one embodiment, the present application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a product display method when executing the computer program. The processor is responsible for executing the instructions in the computer program, specifically including the steps of model acquisition, feature recognition extraction, data processing, data combination and model construction. The memory is used to store computer programs, data processing and virtual models, including acquired image data, tracking algorithms, virtual model information, etc.

[0234] In one embodiment, the present application also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of a product display method when the computer program is executed by a processor. The computer-readable storage medium may include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer storage media may include random access memory, read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, flash memory or other solid-state storage technology, digital versatile disk or other optical storage, tape cassettes, magnetic tapes, disk storage or other magnetic storage devices.

[0235] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application.

Claims

1. A method for controlling positioning and navigation of an intelligent mobile robot, characterized in that: The following steps are involved: Acquire a first path, where the first path is formed based on global map data, and the first path is used to represent a global moving trajectory of the intelligent mobile robot and an initial position of the intelligent mobile robot; Acquiring multimodal real-time observation data, wherein the multimodal real-time observation data includes landmark information, and the landmark information is used to represent the real-time behavior control instructions of the intelligent mobile robot; Jointly forming a multimodal constraint function based on the multimodal real-time observation data; Acquiring the real-time position and posture of the intelligent mobile robot according to the multimodal constraint function; Based on the multimodal real-time observation data and the real-time posture, combined with the target deep reinforcement learning model, a second path is generated, and the second path is used to characterize the local movement trajectory of the intelligent mobile robot and the real-time posture of the intelligent mobile robot.

2. The method according to claim 1, characterized in that: The multimodal real-time observation data includes inertial navigation information; The forming of a multimodal constraint function based on the multimodal real-time observation data comprises the following steps: Acquire a first constraint function according to the landmark information and the first path; Acquire a second constraint function according to the inertial navigation information by using an extended Kalman filter algorithm; A multimodal constraint function is constructed based on the first constraint function and the second constraint function.

3. The method according to claim 1, characterized in that The step of obtaining the real-time position and posture of the intelligent mobile robot according to the multimodal constraint function comprises the following steps: Acquiring a state space distribution of the intelligent mobile robot according to the multimodal constraint function; According to the state space distribution, weighted particles are collected and formed into a first particle set, where the first particle set is used to characterize the real-time position and posture of the intelligent mobile robot; After weight normalization processing is performed on the first particle set, resampling and KLD sampling are performed alternately to generate a second particle set, and the second particle set meets a preset KLD sampling termination judgment criterion; The real-time position and posture are obtained based on the state information and weight information of all particles in the second particle set.

4. The method according to claim 1, characterized in that: The generating of a second path based on the multimodal real-time observation data and the real-time posture in combination with the target deep reinforcement learning model comprises the following steps: Generate local map information based on the real-time observation data and the real-time posture; Based on the local map information, obtaining a real-time behavior control instruction, wherein the real-time behavior control instruction is used to control the intelligent mobile robot to implement path planning corresponding to the real-time posture; According to the real-time behavior control instruction, a second path is obtained in combination with a target deep reinforcement learning model. The second path is used to characterize the local movement trajectory of the intelligent mobile robot and the real-time posture of the intelligent mobile robot.

5. The method according to claim 4, characterized in that The step of obtaining the real-time behavior control instruction based on the local map information comprises the following steps: Inputting the local map information into a target first machine learning model for classification processing to obtain a first behavior control instruction, wherein the first behavior control instruction includes a complex environment behavior control instruction corresponding to a navigation scenario that cannot be processed by the target first machine learning model; Updating the real-time behavior control instruction to the first behavior control instruction; In response to acquiring the complex environment behavior control instruction, sending the local map information to a target second machine learning model, the target second machine learning model is used to generate a second behavior control instruction corresponding to a navigation scenario that cannot be processed by the target first machine learning model; In response to obtaining the second behavior control instruction generated by the target second machine learning model, the real-time behavior control instruction is updated to the second behavior control instruction.

6. The method according to any one of claims 1 to 5, characterized in that: The landmark information includes movable object information; The target deep reinforcement learning model includes an optimal action strategy function determined by a reward function and a value function. The expression of the optimal action strategy function is as follows: In the formula, a represents the optional action of the intelligent mobile robot, a t represents the optimal action of the intelligent mobile robot at time t, s t represents the state of the intelligent mobile robot at time t, represents the state of the intelligent mobile robot and the landmark at time t, represents the state of the intelligent mobile robot and the landmark after time t, is the optimal action strategy function, R(s t , a) is the reward function term used to evaluate the real-time state, γ is the discount factor, is the value function term used to evaluate future states.

7. The method according to claim 6, characterized in that The method of generating a second path based on the multimodal real-time observation data and combining the target deep reinforcement learning model comprises the steps of: Based on the multimodal real-time observation data, an expected spatial model of the movable object is established by a two-dimensional asymmetric Gaussian formula, and the two-dimensional asymmetric Gaussian formula is expressed as follows: In the formula, μ x is the mean value in the x direction, μ y is the mean in the y direction, σ x is the standard deviation in the x direction, σ y is the standard deviation in the y direction; Constructing a state space of the intelligent mobile robot and the landmark based on the expected spatial model of the movable object, wherein the state space of the intelligent mobile robot and the landmark includes the state of the intelligent mobile robot and the landmark at time t and the state of the intelligent mobile robot and the landmark after time t; Based on the state space and multimodal real-time observation data, combined with the target deep reinforcement learning model, the second path is generated.

8. A control device for positioning and navigation of an intelligent mobile robot, characterized in that: include: A first path acquisition module, used to acquire a first path, wherein the first path is formed based on global map data, and the first path is used to represent the global moving trajectory of the intelligent mobile robot and the initial position of the intelligent mobile robot; A real-time observation module, used to obtain multi-modal real-time observation data, wherein the multi-modal real-time observation data includes landmark information, and the landmark information is used to represent the real-time behavior control instructions of the intelligent mobile robot; A multimodal data processing module, used to jointly form a multimodal constraint function based on the multimodal real-time observation data; A real-time positioning module, used to obtain the real-time position and posture of the intelligent mobile robot according to the multimodal constraint function; The second path acquisition module is used to generate a second path based on the multimodal real-time observation data and the real-time posture, combined with the target deep reinforcement learning model, wherein the second path is used to characterize the local movement trajectory of the intelligent mobile robot and the real-time posture of the intelligent mobile robot.

9. A computer device, characterized in that: include: one or more processors; A storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the control method for positioning and navigation of the intelligent mobile robot as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the control method for positioning and navigation of an intelligent mobile robot as described in any one of claims 1 to 7 is implemented.

Images