A multi-mobile robot obstacle avoidance control method and system based on radar and visual feature fusion

Abstract

The application discloses a kind of multi-mobile robot obstacle avoidance control method and system based on radar and vision feature fusion, method includes: the multi-modal sensor point cloud data of mobile robot is collected, and space-time fusion and pre-processing are carried out, and the data after pre-processing is carried out obstacle clustering segmentation, obtain obstacle candidate cluster, then calculate discrete obstacle boundary;Smooth distance approximation model of obstacle is constructed, for from the discrete obstacle boundary extracted from multi-modal point cloud data Conversion into continuous differentiable mathematical representation, obtain smooth distance indicator function, then calculate the safety barrier function value of mobile robot relative to obstacle through smooth distance indicator function;Real-time acquisition mobile robot multi-modal point cloud data, calculate safety barrier function value;According to the safety barrier function value, it is judged whether to trigger obstacle avoidance early warning threshold value.The application can realize the collision-free safe movement of multi-mobile robot system, also can accurately maintain predetermined formation shape and reference trajectory.

Description

Technical Field

[0001] This application belongs to the field of collaborative control and environmental perception technology for multi-mobile robot systems, specifically involving a multi-mobile robot obstacle avoidance control method and system based on radar and visual feature fusion. Background Technology

[0002] With the widespread deployment of intelligent robots in warehousing and logistics, agricultural production, medical care, and inspection tasks, distributed collaboration and formation control of Multi-Robot Systems (MRSs) have become key technologies for improving task efficiency and system reliability. In complex and congested environments, each mobile robot not only needs to maintain a predetermined formation shape through local perception and limited communication, but also must detect and avoid various obstacles in the environment in real time to ensure safe movement.

[0003] In the control hierarchy of multi-mobile robot systems, formation control is a widely used autonomous cooperative and navigation control method. However, existing formation control often leads to insufficient safety distances or collision risks when handling complex obstacles. Therefore, a hierarchical perception and control framework has been developed to adjust for safety deviations caused by primary control and ensure accurate obstacle avoidance trajectories and stable formation shapes. Control methods for achieving the above formation control objectives in multi-mobile robot systems can be divided into centralized control methods and distributed control methods. Distributed control methods have received widespread attention due to their greater flexibility and robustness. Since distributed control assigns environmental perception, mathematical modeling, and real-time obstacle avoidance calculation tasks to each mobile robot, the system is extremely sensitive to the differentiability and conservatism of the obstacle model. Irregular obstacle shapes can lead to uncertainty in perception data and low space utilization. Most existing methods do not construct obstacle constraints from the perspective of multimodal point cloud fusion and smooth nonlinear modeling, resulting in implementation irrationality or limitations when dealing with sensor blind zones or unstructured boundaries. Summary of the Invention

[0004] The technical problem to be solved by this invention is: in the process of formation control of multi-mobile robot systems, in complex environments, how to improve the accuracy of formation control by improving the fitting accuracy of irregular obstacle boundaries.

[0005] To address the aforementioned technical problems, this invention provides a multi-mobile robot obstacle avoidance control method based on radar and visual feature fusion, comprising the following steps:

[0006] A method for obstacle avoidance control of multiple mobile robots based on radar and visual feature fusion, characterized by the following steps:

[0007] Step S1: Collect multimodal sensor point cloud data of the mobile robot, perform spatiotemporal fusion on the multimodal sensor point cloud data, preprocess the spatiotemporal fusion global point cloud data, and perform obstacle clustering and segmentation on the preprocessed data to obtain obstacle candidate clusters, and then calculate the discrete obstacle boundaries.

[0008] Step S2: Construct an obstacle smooth distance approximation model to transform the discrete obstacle boundary extracted from multimodal point cloud data into a continuously differentiable mathematical representation, obtain a smooth distance indicator function, calculate the safety barrier function value of the mobile robot relative to the obstacle through the smooth distance indicator function, and generate an obstacle avoidance guidance reference signal.

[0009] Step S3: Collect multimodal point cloud data of the mobile robot in real time, and calculate the safety barrier function value using the obstacle smoothing distance approximation model in step S2;

[0010] Step S4: Determine whether the obstacle avoidance warning threshold is triggered based on the safety barrier function value.

[0011] The aforementioned obstacle avoidance control method for multiple mobile robots based on radar and visual feature fusion includes, in step S1, acquiring multimodal sensor point cloud data of the mobile robot and performing spatiotemporal fusion of the multimodal sensor point cloud data, including:

[0012] The system uses LiDAR and depth camera to simultaneously acquire the original 3D point cloud data of the mobile robot. It then uses a pre-calibrated external transformation matrix to map the heterogeneous point cloud data to a unified local coordinate system of the mobile robot. Finally, it uses data completion to eliminate the near-range perception blind spots or limited field of view that exist with a single sensor.

[0013] The aforementioned obstacle avoidance control method for multi-mobile robots based on radar and visual feature fusion, in step S1, preprocesses the spatiotemporally fused global point cloud data and performs obstacle clustering and segmentation on the preprocessed data to obtain obstacle candidate clusters, including:

[0014] Ground filtering and outlier removal are performed on the spatiotemporally fused global point cloud data. A neighborhood search is performed using a spatial indexing algorithm. The physical adjacency relationship between the current query point and its neighboring search points is determined based on the geometric connectivity of the point cloud data in the local space. Then, Euclidean clustering or density clustering algorithms are used to segment the discrete point cloud data that meet the distance threshold constraint into independent obstacle candidate clusters.

[0015] In the aforementioned obstacle avoidance control method for multiple mobile robots based on radar and visual feature fusion, in step S1, each obstacle candidate cluster is projected onto the two-dimensional reference plane where the mobile robot moves, and a boundary search algorithm is used to extract a set of closed boundary points describing the external contour of the obstacle. This yields the discrete obstacle boundary, where, For the first The center of the j-th sampling point associated with each obstacle , In order to be with the first The total number of closed boundary sampling points associated with each obstacle; the set of closed boundary points As a subsequent step S2, it is constructed -norm smoothing mathematical model's center parameter.

[0016] The aforementioned obstacle avoidance control method for multi-mobile robots based on radar and visual feature fusion includes, in step S2:

[0017] Step S21: In the obstacle smooth distance approximation model, using -norm theory transforms the edge lattice of irregular obstacles into a continuous and differentiable mathematical representation, smoothing the distance indicator function. Defined as:

[0018] ;

[0019] In the formula, Let be the position vector of the i-th mobile robot; For the first The center of the j-th sampling point associated with each obstacle; In order to be with the first The total number of closed boundary sampling points associated with each obstacle. For the safe radius of mobile robots, The radius of the boundary sampling point; A positive integer greater than 1, used to adjust the model's approximation accuracy of obstacle boundaries;

[0020] Step S22: Based on the smooth distance indication function obtained in step S21, utilize the safety barrier function. The safety barrier function provides real-time assessment of the collision risk between the mobile robot and irregular obstacles. Represented as:

[0021] ;

[0022] In the formula, Let be the safety barrier function value of the i-th mobile robot relative to the obstacle; The preset normalized obstacle avoidance warning boundary threshold is used to define the range of the dangerous potential energy field around the obstacle; It is a positive constant used to adjust the potential energy gain intensity of the barrier function; Let i be the order of the dynamics system of the i-th mobile robot;

[0023] Step S23: Utilizing the local observation information of each mobile robot and the state deviation of neighboring mobile robots, construct a command regulator based on a consensus protocol to generate obstacle avoidance guidance reference signals. The local observation information includes the position, velocity, and attitude reference values ​​of the leading mobile robot. The state deviation of neighboring mobile robots is the vector difference between the internal states of the current mobile robot and the command regulators of neighboring mobile robots. The command regulator is represented as follows:

[0024] ;

[0025] In the formula, The time derivative of the state of the distributed instruction regulator; The set of neighbors in the communication topology; The set of neighbors within the perception range; Let be the state vector of the distributed command regulator for the i-th mobile robot; Let S be the state vector of the distributed command regulator of the j-th mobile robot; S and F are the system matrix and output matrix of the reference system, respectively. To ensure consistency in updating the gain of the command regulator; The observation weights for the mobile robot to the ideal reference state; The intermediate ideal target signal includes the formation geometric offset vector; This is a comprehensive barrier coupling term that includes obstacle avoidance constraint information; This is the preset formation geometry offset vector; and These are the obstacle avoidance barrier terms and the obstacle avoidance barrier terms between mobile robots, respectively, after filtering. This is the preset time delay parameter; This is the final generated adaptive collision-free reference instruction signal used to guide execution layer tracing.

[0026] The aforementioned obstacle avoidance control method for multi-mobile robots based on radar and visual feature fusion includes, in step S3:

[0027] During the operation of multiple mobile robots, heterogeneous sensor data is fused in real time to obtain environmental obstacle information. The obstacle information is then substituted into the obstacle smooth distance approximation model to calculate the smooth distance indicator function value. Based on the smooth distance indicator function value, the safety barrier function value under the current state is further calculated.

[0028] The aforementioned obstacle avoidance control method for multi-mobile robots based on radar and visual feature fusion includes, in step S4:

[0029] The safety barrier function value calculated in real time in step S3 is compared with the preset safety boundary threshold. If the safety barrier function value is lower than the threshold, it is determined to be safe and the previous step is returned to continue monitoring. If the safety barrier function value exceeds the threshold, it is determined that there is a potential collision risk.

[0030] The aforementioned obstacle avoidance control method for multiple mobile robots based on radar and visual feature fusion further includes: step S5, constructing a cooperative tracking controller based on adaptive backstepping, used to receive the obstacle avoidance guidance reference signal generated in step S2, and using the obstacle avoidance guidance reference signal as the control target, controlling the real-time compensation system disturbance through backstepping and driving the mobile robot to track.

[0031] The aforementioned obstacle avoidance control method for multi-mobile robots based on radar and visual feature fusion includes, in step S5:

[0032] Step S51: Construct a multi-order tracking error dynamics and coordinate transformation model to receive the obstacle avoidance guidance reference signal generated in step S2. In multi-order tracking error dynamics, position tracking error is defined as follows: In the formula Let be the output state vector of the i-th mobile robot, referring to the actual position coordinate vector of the i-th mobile robot in the workspace; in the inverse step coordinate transformation model, define the virtual error variable of the p-th order subsystem. :

[0033] ;

[0034] In the formula, Let p be the internal state of the p-th order subsystem in the i-th robot. A virtual control law designed for the (p-1)th order subsystem in the i-th robot;

[0035] Step S52: Using the backstepping method, perform recursion and select Lyapunov candidate functions that include parameter estimation errors for the first-order subsystem. :

[0036] ;

[0037] In the formula, T is the transpose of the vector. For the unknown parameter matrix of the nonlinear system; For the unknown parameter matrix Online estimates; For parameter estimation error, The gain matrix is ​​positive definite; for the subsequent p-th order subsystem, Recursively construct composite Lyapunov functions Represented as:

[0038] ;

[0039] In the formula, Let Lyapunov function be the (p-1)th order subsystem of the i-th mobile robot;

[0040] Step S53: Calculate and generate the final actual control input acting on the mobile robot actuator. Total Lyapunov function With the corresponding control input Represented as:

[0041] ;

[0042] ;

[0043] In the formula, For the i-th mobile robot, the n-th... i Lyapunov function of -1 order subsystem and Error variables for each order of subsystems defined for the backstepping method. It is a positive definite control gain constant; This is a nonlinear compensation term that includes information on the derivatives of state evolution and obstacle avoidance constraints; This is a comprehensive term that includes adaptive parameter compensation and recursive error correction.

[0044] A computer system includes a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of the aforementioned obstacle avoidance control method for multi-mobile robots based on radar and visual feature fusion. The beneficial effects of this invention are as follows: The obstacle avoidance control method for multi-mobile robots based on radar and visual feature fusion provided by this invention constructs a multi-mobile robot system model based on distributed hierarchical control under irregular obstacle constraints. It analyzes obstacle boundaries in complex environments, constructs an obstacle smooth distance approximation model, and characterizes environmental obstacle avoidance features in real time. It compares the real-time position of the mobile robot with the preset safety barrier function boundary to determine whether there is a potential collision risk. By combining multimodal perception, distributed command adjustment, and adaptive backstepping control, safe navigation of the multi-mobile robot system in complex environments is achieved. The obstacle smooth distance approximation model significantly reduces the conservatism of traditional geometric models and improves environmental space utilization. The obstacle smooth distance approximation model accurately improves the fitting accuracy of irregular obstacle boundaries. It can eliminate the influence of sensor blind spots and system nonlinear disturbances, enabling not only collision-free movement of the swarm but also accurate maintenance of formation shape.

[0045] Meanwhile, after determining that there is a collision risk, the present invention uses an adaptive reference signal based on the output of a distributed command regulator to adaptively adjust the navigation layer commands; at the same time, it uses a cooperative tracking controller based on an adaptive backstepping method to estimate unknown system parameters and external disturbances in real time and generate control inputs to drive the mobile robot to accurately track the adaptive reference signal, ultimately achieving collision-free motion and accurate maintenance of the predetermined formation shape of the multi-mobile robot system. Attached Figure Description

[0046] Figure 1 This is a flowchart of a multi-mobile robot obstacle avoidance control method based on radar and visual feature fusion provided in Embodiment 1 of the present invention;

[0047] Figure 2 This is a schematic diagram of the simulation results of obstacle avoidance trajectories of multi-mobile robot formations in a complex obstacle environment using a simplified elliptical geometry modeling method based on existing technology.

[0048] Figure 3 This is a schematic diagram of the simulation results of the obstacle avoidance trajectory of a multi-mobile robot formation in a complex L-shaped obstacle environment in Embodiment 1 of the present invention;

[0049] Figure 4 It corresponds to Figure 2 The relative distance change curves between the mobile robots during the simulation process and the schematic diagram of the safe collision threshold;

[0050] Figure 5 This corresponds to Embodiment 1 of this application. Figure 3 The simulation process includes curves showing the relative distance changes between the mobile robots and a schematic diagram of the safe collision threshold. Detailed Implementation

[0051] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this application. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application.

[0053] Example 1

[0054] like Figure 1As shown, this embodiment aims to overcome the problem that simplified geometric models in the prior art cannot meet the requirements of obstacle avoidance accuracy and space utilization in complex environments. It provides a multi-mobile robot obstacle avoidance control method based on radar and visual feature fusion, comprising the following steps:

[0055] Step S1: Collect multimodal sensor point cloud data of the mobile robot, perform spatiotemporal fusion on the multimodal sensor point cloud data, preprocess the spatiotemporally fused global point cloud data, and perform obstacle clustering segmentation on the preprocessed data to obtain obstacle candidate clusters. Then, calculate the discrete obstacle boundaries. This step can eliminate the perception limitations of a single sensor. Through spatiotemporal alignment, clustering segmentation, and contour extraction algorithms, unstructured obstacles in complex environments are transformed into closed boundary point sets with clear geometric features, thus providing a basis for subsequent construction. -The normal smooth mathematical model provides accurate physical constraint parameters, including:

[0056] Step S11: Collect multimodal sensor point cloud data of the mobile robot and perform spatiotemporal fusion of the multimodal sensor point cloud data, including:

[0057] The system uses LiDAR and depth camera to simultaneously acquire the original 3D point cloud data of the mobile robot. It then uses a pre-calibrated external transformation matrix to map the heterogeneous point cloud data to a unified local coordinate system of the mobile robot. Finally, it uses data completion to eliminate the near-range perception blind spots or limited field of view that exist with a single sensor.

[0058] Step S12: Preprocess the spatiotemporally fused global point cloud data, and perform obstacle clustering segmentation on the preprocessed data to obtain obstacle candidate clusters, including:

[0059] Ground filtering and outlier removal are performed on the spatiotemporally fused global point cloud data. A neighborhood search is performed using a spatial indexing algorithm (such as KD-tree). The physical adjacency relationship between the current query point and its neighboring search points is determined based on the geometric connectivity of the point cloud data in the local space. Then, Euclidean clustering or density clustering (DBSCAN) algorithms are used to segment the discrete point cloud data that meet the distance threshold constraint into independent obstacle candidate clusters, thereby realizing the spatial entity recognition of unstructured obstacle targets in complex environments.

[0060] Step S13: Project each obstacle candidate cluster onto the two-dimensional reference plane where the mobile robot moves, and use a boundary search algorithm (Alpha-shape or Hull algorithm) to extract a set of closed boundary points describing the outer contour of the obstacle. This yields the discrete obstacle boundary, where, For the first The center of the j-th sampling point associated with each obstacle , In order to be with the first The total number of closed boundary sampling points associated with each obstacle; the set of closed boundary points As a subsequent step S2, it is constructed -norm smoothing mathematical model's center parameter.

[0061] Step S2: Construct a smoothed obstacle distance approximation model to transform the discrete obstacle boundaries extracted from multimodal point cloud data into a continuously differentiable mathematical representation, obtain a smoothed distance indicator function, and then calculate the safety barrier function value of the mobile robot relative to the obstacle using the smoothed distance indicator function, and generate an obstacle avoidance guidance reference signal, including:

[0062] Step S21: In the obstacle smooth distance approximation model, in order to overcome the non-smoothness of the discrete point set when judging the boundary, the following is used: -norm theory transforms the edge lattice of irregular obstacles into a continuous and differentiable mathematical representation, smoothing the distance indicator function. Defined as:

[0063] ;

[0064] In the formula, Let be the position vector of the i-th mobile robot; For the first The center of the j-th sampling point associated with each obstacle; In order to be with the first The total number of closed boundary sampling points associated with each obstacle. For the safe radius of mobile robots, The radius of the boundary sampling point; It is a positive integer greater than 1, used to adjust the model's approximation accuracy of obstacle boundaries.

[0065] Step S22: Based on the smooth distance indicator function obtained in step S21, utilize the safety barrier function with singular repulsion properties. Real-time assessment of collision risks between mobile robots and irregular obstacles; safety barrier function. Represented as:

[0066] ;

[0067] In the formula, Let be the safety barrier function value of the i-th mobile robot relative to the obstacle; The preset normalized obstacle avoidance warning boundary threshold is used to define the range of the dangerous potential energy field around the obstacle; It is a positive constant used to adjust the potential energy gain intensity of the barrier function; Let be the order of the dynamics system of the i-th mobile robot. When the mobile robot approaches the obstacle boundary, it causes... At this point, the value of the safety barrier function increases dramatically to generate an infinitely large potential energy constraint, and the constrained system enters the safe set.

[0068] Step S23: Utilizing the local observation information of each mobile robot and the state deviation of neighboring mobile robots, construct a command regulator based on a consensus protocol to generate obstacle avoidance guidance reference signals. The local observation information includes the position, velocity, and attitude reference values ​​of the leading mobile robot. The state deviation of neighboring mobile robots is the vector difference between the internal states of the current mobile robot and the command regulators of neighboring mobile robots. The command regulator is represented as follows:

[0069] ;

[0070] In the formula, The time derivative of the state of the distributed instruction regulator; The set of neighbors in the communication topology; The set of neighbors within the perception range; Let be the state vector of the distributed command regulator for the i-th mobile robot; Let S be the state vector of the distributed command regulator of the j-th mobile robot; S and F are the system matrix and output matrix of the reference system, respectively. To ensure consistency in updating the gain of the command regulator; The observation weights for the mobile robot to the ideal reference state; The intermediate ideal target signal includes the formation geometric offset vector; This is a comprehensive barrier coupling term that includes obstacle avoidance constraint information; This is the preset formation geometry offset vector; and These are the obstacle avoidance barrier terms and the obstacle avoidance barrier terms between mobile robots, respectively, after filtering. This is the preset time delay parameter; This is the final generated adaptive collision-free reference instruction signal used to guide execution layer tracing.

[0071] Step S3: Real-time acquisition of multimodal point cloud data of the mobile robot, and calculation of the safety barrier function value using the obstacle smoothing distance approximation model from step S2, including:

[0072] During the operation of multiple mobile robots, heterogeneous sensor data is fused in real time to obtain environmental obstacle information. The obstacle information is then substituted into the obstacle smooth distance approximation model to calculate the smooth distance indicator function value. Based on the smooth distance indicator function value, the safety barrier function value under the current state is further calculated.

[0073] Step S4: Determine whether the obstacle avoidance warning threshold is triggered based on the safety barrier function value, including:

[0074] The safety barrier function value calculated in real time in step S3 is compared with the preset safety boundary threshold. If the safety barrier function value is lower than the threshold, it is determined to be safe and the process returns to the previous step to continue monitoring. If the safety barrier function value exceeds the threshold, it is determined that there is a potential collision risk and the process proceeds to the next step.

[0075] This embodiment uses Lyapunov stability theory and proof by contradiction to prove the reference signal generated by the distributed command regulator. The ability to consistently keep the mobile robot within a safe set is analyzed as follows:

[0076] Construct the obstacle avoidance potential energy function and safety constraints. According to the safety barrier function in step S22, when the mobile robot i approaches the obstacle boundary, i.e. When mobile robot i approaches other mobile robots, i.e. The value of the safety barrier function will tend to infinity, and its mathematical singularity characteristic is expressed as: , .

[0077] Establish the Lyapunov energy function for system consistency error, and define the consistency error vector of the command regulator as follows: Construct Lyapunov candidate functions Taking the derivative of the consistency error system of the command regulator, the energy decay characteristic satisfies:

[0078] ;

[0079] In the formula, For matrix The smallest eigenvalue. Integrating the above equation, we can see that the system energy... At any time Both are bounded, that is The security proof based on proof by contradiction assumes that at a certain moment... The risk of collision exists, meaning there is a mobile robot causing the barrier function value to approach infinity. At this point, the integral value of the Lyapunov function will satisfy the following divergence condition:

[0080] ;

[0081] In the formula, These are positive constants related to system parameters. Therefore, the above results... The proven system energy in the Lyapunov energy function for establishing system consistency error. This is a bounded contradiction. Therefore, the assumption is false, thus proving that at any time... Barrier function Always maintain boundaries, meaning that the mobile robot always maintains a safe distance from obstacles and its neighbors. and This achieves rigorous theoretical obstacle avoidance.

[0082] Step S5: Construct a cooperative tracking controller based on adaptive backstepping, which receives the obstacle avoidance guidance reference signal generated in step S2, and uses the obstacle avoidance guidance reference signal as the control target. It then uses backstepping to control and compensate for system disturbances in real time, driving the mobile robot to track the obstacle. This includes:

[0083] Step S51: Construct a multi-order tracking error dynamics and coordinate transformation model to receive the obstacle avoidance guidance reference signal generated in step S2. In multi-order tracking error dynamics, position tracking error is defined as follows: In the formula Let be the output state vector of the i-th mobile robot, referring to the actual position coordinate vector of the i-th mobile robot in the workspace. To solve the control problem of high-order nonlinear systems, in the inverse step coordinate transformation model, a virtual error variable of the p-th order subsystem is defined. :

[0084] ;

[0085] In the formula, Let p be the internal state of the p-th order subsystem in the i-th robot. The virtual control law is designed for the (p-1)th order subsystem of the i-th robot.

[0086] Step S52: Using the backstepping method, perform recursion and select Lyapunov candidate functions that include parameter estimation errors for the first-order subsystem. :

[0087] ;

[0088] In the formula, T is the transpose of the vector. For the unknown parameter matrix of the nonlinear system; For the unknown parameter matrix Online estimates; For parameter estimation error, The gain matrix is ​​positive definite; for the subsequent p-th order subsystem, Recursively construct composite Lyapunov functions Represented as:

[0089] ;

[0090] In the formula, Let Lyapunov function be the Lyapunov function of the (p-1)th order subsystem of the i-th mobile robot; based on composite Lyapunov functions. To meet stability requirements, virtual control laws are used. and parameter adaptive update law It cancels out nonlinear dynamic terms and ensures the energy decay of the subsystem.

[0091] Step S53: Based on the total Lyapunov function To satisfy the stability requirement of global asymptotic convergence, the actual control inputs that ultimately act on the actuators of the mobile robot are calculated and generated. Total Lyapunov function With the corresponding control input Represented as:

[0092] ;

[0093] ;

[0094] In the formula, For the i-th mobile robot, the n-th... i Lyapunov function of -1 order subsystem and Error variables for each order of subsystems defined for the backstepping method. It is a positive definite control gain constant; This is a nonlinear compensation term that includes information on the derivatives of state evolution and obstacle avoidance constraints; This is a comprehensive term that includes adaptive parameter compensation and recursive error correction. For the overall Lyapunov function... Based on the above actual control inputs With parameter adaptive update law This proves that the closed-loop system satisfies ,in, , The value is a positive constant, thus reducing the tracking error. It can asymptotically converge globally to a small neighborhood of the origin, enabling accurate tracking of the reference trajectory.

[0095] The effectiveness of the method of this invention is further illustrated below through simulation experimental data. This experiment uses the MATLAB simulation platform to build a multi-mobile robot cooperative control simulation model containing three nonholonomic mobile robots. The physical layer includes three mobile robots subject to nonholonomic constraints and a complex simulation environment containing two L-shaped polygonal obstacles, with dimensions of approximately 0.7m × 0.7m and center coordinates at (1.9, 1.1) and (2.3, 1.5), respectively. The control architecture layer includes a perception module based on multimodal point cloud fusion, a navigation module based on a distributed command regulator, and a control module based on an adaptive backstepping method. The main dynamic parameters of the mobile robots are set to their own safe radius. Formation control and environmental parameters are set as the initial formation geometric offset vector. and external time-varying perturbations applied to the robot .

[0096] From simulation results Figure 2 and Figure 3 It can be seen that the existing formation control method based on elliptical approximation modeling and the present invention are based on... The -norm modeling methods were compared in complex environments. In open areas, both methods could drive robot motion. However, with the addition of irregular L-shaped obstacles, the results showed that existing elliptical modeling methods lacked flexibility in non-convex obstacle environments. The models were too conservative and failed to describe local concave features, leading to limitations in robot movement. Unable to effectively avoid the collision, a clear collision and trajectory overlap occurred; for example Figure 3 As shown, under the same working conditions, the method of the present invention can accurately fit the boundary of the obstacle, drive the robot group to smoothly pass through the narrow channel, and realize collision-free formation maneuvering, which shows that the proposed method has the ability to adapt to complex geometric constraints.

[0097] From simulation results Figure 4 and Figure 5 As can be seen, the changes in the relative distance between the robots under the two methods were compared. Ideally, the relative distance... It should always be kept above the safety threshold (dashed line). Add irregular L-shaped obstacles, such as... Figure 4 As shown, the results indicate that existing ellipse modeling methods react sharply when faced with complex obstacle compression, leading to a decrease in robot spacing. If the safety threshold is breached, resulting in one or more collision failures, further improvements are needed; such as... Figure 5 As shown, using the method proposed in this invention, even in extremely limited environmental spaces, the relative distance between the robots is always kept stable above the safety threshold, which meets the obstacle avoidance requirements. This demonstrates that the proposed distributed command regulator has the ability to resist complex environmental interference and strictly ensure system safety.

[0098] Example 2

[0099] A computer system includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the obstacle avoidance control method for multiple mobile robots based on radar and visual feature fusion as described in Embodiment 1.

[0100] In summary, the method proposed in this application can effectively counteract the effects of complex and irregular obstacle environments and nonlinear disturbances in the system. It can not only achieve collision-free safe movement of multi-mobile robot systems, but also accurately maintain the predetermined formation shape and reference trajectory.

[0101] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0102] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A multi-mobile robot obstacle avoidance control method based on radar and visual feature fusion, characterized in that, Includes the following steps: Step S1: Collect multimodal sensor point cloud data of the mobile robot, perform spatiotemporal fusion on the multimodal sensor point cloud data, preprocess the spatiotemporal fusion global point cloud data, and perform obstacle clustering and segmentation on the preprocessed data to obtain obstacle candidate clusters, and then calculate the discrete obstacle boundaries. Step S2: Construct an obstacle smooth distance approximation model to transform the discrete obstacle boundary extracted from multimodal point cloud data into a continuously differentiable mathematical representation, obtain a smooth distance indicator function, calculate the safety barrier function value of the mobile robot relative to the obstacle through the smooth distance indicator function, and generate an obstacle avoidance guidance reference signal. Step S3: Collect multimodal point cloud data of the mobile robot in real time, and calculate the safety barrier function value using the obstacle smoothing distance approximation model in step S2; Step S4: Determine whether the obstacle avoidance warning threshold is triggered based on the safety barrier function value; Step S2 includes: Step S21: In the obstacle smooth distance approximation model, using -norm theory transforms the edge lattice of irregular obstacles into a continuous and differentiable mathematical representation, smoothing the distance indicator function. Defined as: ； In the formula, For the first The position vector of a mobile robot; For the first The first obstacle associated with the The center of each sampling point; In order to be with the first The total number of closed boundary sampling points associated with each obstacle. For the safe radius of mobile robots, The radius of the boundary sampling point; A positive integer greater than 1, used to adjust the model's approximation accuracy of obstacle boundaries; Step S22: Based on the smooth distance indication function obtained in step S21, utilize the safety barrier function. The safety barrier function provides real-time assessment of the collision risk between the mobile robot and irregular obstacles. Represented as: ； In the formula, For the first The safety barrier function value of a mobile robot relative to an obstacle; The preset normalized obstacle avoidance warning boundary threshold is used to define the range of the dangerous potential energy field around the obstacle; It is a positive constant used to adjust the potential energy gain intensity of the barrier function; For the first The order of a mobile robot dynamics system; Step S23: Utilizing the local observation information of each mobile robot and the state deviation of neighboring mobile robots, construct a command regulator based on a consensus protocol to generate obstacle avoidance guidance reference signals. The local observation information includes the position, velocity, and attitude reference values ​​of the leading mobile robot. The state deviation of neighboring mobile robots is the vector difference between the internal states of the current mobile robot and the command regulators of neighboring mobile robots. The command regulator is represented as follows: ； In the formula, The time derivative of the state of the distributed instruction regulator; The set of neighbors in the communication topology; The set of neighbors within the perception range; For the first The state vector of the distributed command regulator of a mobile robot; Let be the state vector of the distributed command regulator for the j-th mobile robot; and These are the system matrix and output matrix of the reference system, respectively; To ensure consistency in updating the gain of the command regulator; The observation weights for the mobile robot to the ideal reference state; The intermediate ideal target signal includes the formation geometric offset vector; This is a comprehensive barrier coupling term that includes obstacle avoidance constraint information; This is the preset formation geometry offset vector; and These are the obstacle avoidance barrier terms and the obstacle avoidance barrier terms between mobile robots, respectively, after filtering. This is the preset time delay parameter; This is the final generated adaptive collision-free reference instruction signal used to guide execution layer tracing.

2. The obstacle avoidance control method for multiple mobile robots based on radar and visual feature fusion according to claim 1, characterized in that, In step S1, multimodal sensor point cloud data of the mobile robot is collected, and the multimodal sensor point cloud data is spatiotemporally fused, including: The system uses LiDAR and depth camera to simultaneously acquire the original 3D point cloud data of the mobile robot. It then uses a pre-calibrated external transformation matrix to map the heterogeneous point cloud data to a unified local coordinate system of the mobile robot. Finally, it uses data completion to eliminate the near-range perception blind spots or limited field of view that exist with a single sensor.

3. The obstacle avoidance control method for multiple mobile robots based on radar and visual feature fusion according to claim 1, characterized in that, In step S1, the spatiotemporally fused global point cloud data is preprocessed, and obstacle clustering is performed on the preprocessed data to obtain obstacle candidate clusters, including: Ground filtering and outlier removal are performed on the spatiotemporally fused global point cloud data. A neighborhood search is performed using a spatial indexing algorithm. The physical adjacency relationship between the current query point and its neighboring search points is determined based on the geometric connectivity of the point cloud data in the local space. Then, Euclidean clustering or density clustering algorithms are used to segment the discrete point cloud data that meet the distance threshold constraint into independent obstacle candidate clusters.

4. The obstacle avoidance control method for multiple mobile robots based on radar and visual feature fusion according to claim 1, characterized in that, In step S1, each obstacle candidate cluster is projected onto the two-dimensional reference plane where the mobile robot moves, and a set of closed boundary points describing the external contour of the obstacle is extracted using a boundary search algorithm. This yields the discrete obstacle boundary, where, For the first The first obstacle associated with the The center of each sampling point , In order to be with the first The total number of closed boundary sampling points associated with each obstacle; the set of closed boundary points As a subsequent step S2, it is constructed -norm smoothing mathematical model's center parameter.

5. The obstacle avoidance control method for multiple mobile robots based on radar and visual feature fusion according to claim 1, characterized in that, Step S3 includes: During the operation of multiple mobile robots, heterogeneous sensor data is fused in real time to obtain environmental obstacle information. The obstacle information is then substituted into the obstacle smooth distance approximation model to calculate the smooth distance indicator function value. Based on the smooth distance indicator function value, the safety barrier function value under the current state is further calculated.

6. The obstacle avoidance control method for multiple mobile robots based on radar and visual feature fusion according to claim 1, characterized in that, Step S4 includes: The safety barrier function value calculated in real time in step S3 is compared with the preset safety boundary threshold. If the safety barrier function value is lower than the threshold, it is determined to be safe and the previous step is returned to continue monitoring. If the safety barrier function value exceeds the threshold, it is determined that there is a potential collision risk.

7. The obstacle avoidance control method for multiple mobile robots based on radar and visual feature fusion according to claim 1, characterized in that, Also includes: Step S5: Construct a cooperative tracking controller based on adaptive backstepping, which receives the obstacle avoidance guidance reference signal generated in step S2, uses the obstacle avoidance guidance reference signal as the control target, and controls the real-time compensation system disturbance through backstepping to drive the mobile robot to track.

8. The obstacle avoidance control method for multiple mobile robots based on radar and visual feature fusion according to claim 7, characterized in that, Step S5 includes: Step S51: Construct a multi-order tracking error dynamics and coordinate transformation model to receive the obstacle avoidance guidance reference signal generated in step S2. In multi-order tracking error dynamics, position tracking error is defined as follows: In the formula For the first The output state vector of the i-th mobile robot refers to the i-th The actual position coordinate vector of a mobile robot in the workspace; in the inverse step coordinate transformation model, the virtual error variable of the p-th order subsystem is defined. : ； In the formula, For the first The first robot The internal state of the subsystem. For the first The first robot Virtual control law for subsystem design; Step S52: Using the backstepping method, perform recursion and select Lyapunov candidate functions that include parameter estimation errors for the first-order subsystem. : ； In the formula, T is the transpose of the vector. For the unknown parameter matrix of the nonlinear system; For the unknown parameter matrix Online estimates; For parameter estimation error, The gain matrix is ​​positive definite; for the subsequent p-th order subsystem, Recursively construct composite Lyapunov functions Represented as: ； In the formula, For the first The first mobile robot Lyapunov functions of the order subsystem; Step S53: Calculate and generate the final actual control input acting on the mobile robot actuator. The total Lyapunov function With the corresponding control input Represented as: ； ； In the formula, For the first The first mobile robot Lyapunov functions of the order subsystem, and Error variables for each order of subsystems defined for the backstepping method. It is a positive definite control gain constant; This is a nonlinear compensation term that includes information on the derivatives of state evolution and obstacle avoidance constraints; This is a comprehensive term that includes adaptive parameter compensation and recursive error correction.

9. A computer system comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the obstacle avoidance control method for multi-mobile robots based on radar and visual feature fusion as described in any one of claims 1-8.

Images