Uncertain nonlinear agent obstacle avoidance control method based on tracking

Abstract

The invention relates to the technical field of intelligent control, in particular to an uncertain nonlinear intelligent agent obstacle avoidance control method based on tracking. According to the method, an intelligent agent is modeled into an uncertain nonlinear model and converted into an uncertain nonlinear differential equation, a dynamic tracking controller is designed based on input state stability, and an obstacle avoidance controller is designed based on feasible set projection. The invention provides an uncertain nonlinear intelligent agent obstacle avoidance control method based on tracking, and solves or at least alleviates the problems that an existing intelligent agent obstacle avoidance control method is adaptive to an uncertain nonlinear dynamic model and a complex environment, is easy to fall into local optimum, is poor in real-time performance and the like, so that the requirement of obstacle avoidance control on a system is reduced, and the system reliability is improved. And moreover, the obstacle avoidance controller is designed based on the first-order integrator, so that the design process of the obstacle avoidance controller is simplified, and a simple low-order system can control a complex high-order intelligent agent.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to a tracking-based method for obstacle avoidance control of uncertain nonlinear intelligent agents. Background Technology

[0002] In many key areas such as industrial production, public services, and national defense, intelligent agents, represented by unmanned vehicles, drones, and unmanned ships, are achieving unprecedented widespread application and rapid popularization. In real-world operational scenarios, ensuring these intelligent agents can achieve safe, reliable, and efficient autonomous movement in environments filled with complex obstacles has become one of their core functions. However, the dynamic equations of real physical intelligent agents are highly nonlinear and contain unavoidable model uncertainties. They are also highly susceptible to parameter noise (such as slight deviations in inertia tensors and damping coefficients), external disturbances (such as instantaneous wind and water flow forces), and unmodeled dynamics (such as vibrations of flexible structures or inherent sensor noise). All of these factors make it difficult for traditional control methods to maintain and guarantee the good autonomy and reliability of these intelligent agents, which involve multidisciplinary constraints encompassing perception, computation, decision-making, control, and execution, in real, dynamic environments.

[0003] Meeting the above requirements is a technical solution currently available in the field. For example, in a nonlinear teleoperation multilateral control method considering formation obstacle avoidance disclosed in Chinese Patent Publication No. CN112859596B, an obstacle avoidance planner that integrates an artificial potential function obstacle avoidance algorithm and a leader switching algorithm is designed during the process of handling obstacle environments. Virtual force feedback is provided at the master end, enabling the operator to understand the movement status of the slave leader in a timely and accurate manner. Furthermore, through the slave formation controller, the slave robot can perform effective obstacle avoidance and formation movement under nonholonomic constraints. In addition, considering the inherent nonlinearity of the system and various uncertainties, a master trajectory planner and a master controller are designed to obtain excellent tracking performance of the master robot. Finally, through the cooperative action of the controllers, the global stability of the multilateral teleoperation system is ensured.

[0004] However, with the gradual maturation of intelligent agent technology, the expansion of its application boundaries, and the increasing complexity of operating environments, the inherent characteristics of the aforementioned technical principles have gradually exposed their inherent defects when facing new demands, leading to new contradictions. For example, obstacle avoidance planners based on artificial potential functions, such as CN112859596B, rely on a virtual potential field to force the agent to avoid obstacles. However, due to the gradient descent characteristics of artificial potential functions, the agent is prone to getting trapped in local optima when facing complex, dense, or concave obstacle environments, making it unable to effectively plan a globally optimal or feasible obstacle avoidance path, or even escape the obstacle altogether. The root cause is the lack of global exploration capabilities, which is prevalent in the real world, such as in urban canyons, narrow valleys, or dynamic obstacle clusters. Furthermore, this method also heavily relies on communication between the master and slave ends, and in teleoperation scenarios, communication delays, bandwidth limitations, and even communication interruptions are common. Once communication quality deteriorates, the system is forced to slow down or adopt a more conservative strategy in order to maintain global stability. This greatly reduces the timeliness and effectiveness of obstacle avoidance, making the autonomy and reliability of the entire system subject to the vulnerability of the external communication link.

[0005] In summary, current agent-based obstacle avoidance control methods face fundamental problems when dealing with uncertain nonlinear models, including insufficient adaptability to uncertain environments, susceptibility to local minima, high requirements for communication or prior information, high computational complexity, and difficulty in implementation. The underlying reason is that these methods require high model accuracy or high-bandwidth communication to obtain certain performance indicators. However, the real physical world inherently contains model uncertainties, external interference, and resource limitations. Current methods cannot comprehensively optimize these fundamental contradictions in complex environments, thus limiting their robustness, autonomy, and ease of implementation.

[0006] Therefore, designing an obstacle avoidance control method that can effectively cope with the uncertain nonlinear dynamics of intelligent agents, reduce dependence on accurate models and communication, and simultaneously ensure real-time performance and theoretical safety has become a key challenge and an urgent technical problem for those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a tracking-based uncertain nonlinear intelligent agent obstacle avoidance control method, which can solve or at least alleviate the problems of existing intelligent agent obstacle avoidance control methods in the face of uncertain nonlinear dynamic models, insufficient adaptability to complex environments, easy getting trapped in local optima, excessive dependence on communication or prior information, high computational complexity, and poor real-time performance.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a tracking-based uncertain nonlinear intelligent agent obstacle avoidance control method, the method comprising:

[0009] Step S1. Establish an uncertain nonlinear kinematic model of the intelligent agent system. This model includes the position vector, the velocity vector in the body coordinate system, the inertia matrix, the uncertain damping matrix, and the control force and torque vectors.

[0010] The planar kinematic model is transformed into a state space and rewritten as a differential equation. This form includes an inertial position state vector, a body velocity state vector, redefined control input and output vectors, and a nonlinear term characterizing the uncertainty of the system. This nonlinear term satisfies the quasi-local Lipschitz property.

[0011] Step S2. Design an observer-based tracking controller to enable the agent's actual position to track a time-varying reference signal. The design includes:

[0012] Design a virtual controller to obtain a first error subsystem, wherein the tracking error is the difference between the actual position of the agent and the reference signal, and the virtual controller contains a non-decreasing, continuously differentiable positive feedback gain function;

[0013] Design a real controller to obtain a second error subsystem and an observation error subsystem, wherein the virtual tracking error is the difference between the actual speed and the desired virtual control input, and the observation error is the estimation deviation of the unmeasurable disturbance term. The real controller includes another non-decreasing continuously differentiable positive feedback gain function and incorporates a state observer to estimate the unmeasurable disturbance term.

[0014] The dynamic tracking controller ensures that the closed-loop system is stable from input to state when the reference signal is used as input and each error is used as the state by satisfying the multi-loop nonlinear small gain condition.

[0015] Step S3. Design an obstacle avoidance controller based on a first-order integrator, and generate a safe, time-varying speed command as the reference signal for the tracking controller;

[0016] Construct a speed feasible set, which defines the range of safe speed commands that the agent can take given the agent's relative position to the obstacle;

[0017] The obstacle avoidance controller determines whether the speed command is within the feasible speed set and projects it in real time to ensure that the agent can actively avoid stationary obstacles while tracking the desired trajectory.

[0018] Step S4. Select appropriate parameters to complete obstacle avoidance control.

[0019] To further realize the present invention, the following technical solutions may be preferred:

[0020] Preferably, in the kinematic model, both the inertia matrix and the damping matrix are diagonal matrices, wherein the diagonal elements of the damping matrix are nonlinear functions dependent on the corresponding velocity components, and their uncertainty arises from the random fluctuations of the environmental medium and factors that make it difficult for the intelligent agent to model accurately.

[0021] Preferably, the redefined control input is the original control force and torque vector multiplied by the inverse of the inertia matrix, and the nonlinear term is determined by the inverse of the inertia matrix, the damping matrix, and the velocity vector; the model establishment introduces... Class and A class function is used to define the class-local Lipschitz property of the nonlinear term.

[0022] Preferably, the virtual controller is designed as a negative feedback mechanism for tracking errors, and the value of its feedback gain function must satisfy a condition consisting of multiple... Specific inequality conditions are constructed using class functions to ensure the stability of the first error subsystem.

[0023] Preferably, the state observer employs a first-order dynamic structure, and its internal state and output are used to generate estimates of unmeasurable disturbance terms; the real controller is a combination of negative feedback of the virtual tracking error and the estimated values, and the value of its feedback gain function must satisfy another condition consisting of multiple... The specific inequality conditions formed by the class functions, together with the conditions of the virtual controller, constitute a multi-loop nonlinear small-gain condition.

[0024] Preferably, the multi-loop nonlinear low-gain condition includes three sets of composite inequalities, which together ensure the input-to-state stability of the closed-loop system.

[0025] Preferably, the dynamic tracking controller is composed of a set of equations of a virtual controller, a real controller, and a state observer, which can ensure that the closed-loop system is stable from input to state with the derivative of the reference signal as input and the tracking error, virtual tracking error, and observation error as state.

[0026] Preferably, the reference signal is generated by a first-order integrator system; the obstacle avoidance controller defines the relative position of the agent and the obstacle and sets a collision avoidance margin not less than the desired safe distance; the feasible velocity set is constructed based on a combination of a strictly decreasing strictly convex potential function and a strictly increasing scaling function to constrain the safe velocity command.

[0027] Preferably, the obstacle avoidance controller adopts a segmented projection mechanism: when the original speed command generated by the high-level task planner is within the feasible speed set, the command is directly used; when the original speed command is outside the feasible set, it is projected onto the boundary of the feasible set, and the projected command is composed of an avoidance component along the obstacle repulsion direction and an original command component perpendicular to that direction.

[0028] Preferably, the implementation process of the method includes: the sensor system acquiring the position information of the intelligent agent and the obstacle in real time; the central control unit calculating the relative position and the obstacle avoidance controller module generating a safe speed command; the command being input as a reference signal to the tracking controller module; the tracking controller module combining the observer state and disturbance estimation to calculate the final control input and sending it to the actuator; wherein, the gain parameters of each controller and the observer gain are preset through offline analysis to meet the small gain stability condition of the system.

[0029] The beneficial effects of this invention are:

[0030] This invention establishes an uncertain nonlinear kinematic model of the intelligent agent, designs an observer-based dynamic tracking controller, and designs an obstacle avoidance controller based on a first-order integrator, thus forming the overall controller of this control method, thereby realizing obstacle avoidance control of the intelligent agent. Specific advantages are as follows:

[0031] (1) This obstacle avoidance control method is based on tracking control, which reduces the requirements of the intelligent agent and expands the scope of application of this method;

[0032] (2) The obstacle avoidance controller in this obstacle avoidance control method is designed based on a first-order integrator. Compared with designing the obstacle avoidance controller directly based on the agent model, it simplifies the design process and realizes the control of a high-order agent through a low-order system. Attached Figure Description

[0033] Figure 1 A flowchart of the control method of the present invention;

[0034] Figure 2 A schematic diagram of the gain directed graph and the small gain condition satisfied by the error correlation system of the present invention;

[0035] Figure 3 A closed-loop system block diagram of the control method of the present invention;

[0036] Figure 4 A schematic diagram of the motion trajectory of the intelligent agent on a plane according to the present invention;

[0037] Figure 5 A schematic diagram of the obstacle avoidance scenario of the intelligent agent according to the present invention. Detailed Implementation

[0038] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Example 1

[0041] This embodiment discloses a tracking-based uncertain nonlinear intelligent agent obstacle avoidance control method, such as... Figure 1 As shown, this paper presents a safe and efficient autonomous motion solution for intelligent agents in complex, dynamic environments with model uncertainties. The method systematically constructs an uncertain nonlinear kinematic model of the agent, designs a dynamic tracking controller based on input-to-state stability, and combines it with a real-time obstacle avoidance controller based on feasible set projection, forming a hierarchical and collaborative control architecture. This ensures that the agent can effectively avoid obstacles while tracking the desired trajectory.

[0042] Before detailing the method proposed in this invention, it is necessary to first model the dynamic characteristics of the intelligent agent to capture its inherent nonlinearity and uncertainty. In a specific embodiment, the intelligent agent is defined as a physical entity performing planar motion in an inertial coordinate system, its position determined by a vector... It means that among them Represents the eastward position component. This represents the northward position component. The agent's velocity in the body coordinate system is represented by a vector. It means that among them Represents the lower edge of the body coordinate system The velocity components of the shaft, Represents the lower edge of the body coordinate system The velocity components of the axis. This representation is widely applicable to various mobile intelligent agents, such as drones, unmanned vehicles, or underwater vehicles, whose motion is typically dominated in the horizontal plane, and whose dynamics in the vertical direction can be decoupled or simplified.

[0043] The kinematic model of the agent on the plane is established as follows:

[0044]

[0045] in, This represents the control force and torque vectors applied to the agent, constituting the agent's control input. This control input is generated by the agent's drive system, and is formed, for example, by combining multiple propeller thrust vectors, wheel torque, or rudder deflection forces. Let be the inertia matrix of the agent, which is a diagonal matrix, where and These represent the equivalent inertial parameters of the agent in the two main directions of motion. These parameters are typically obtained through physical measurements or system identification and may contain slight uncertainties due to load variations or structural deformations. The damping matrix is ​​uncertain; it is a diagonal matrix with diagonal elements... This represents the nonlinear damping coefficient of the agent under the corresponding velocity component. The uncertainty of this damping matrix mainly comes from the random fluctuations of the environmental medium (such as changes in the density, temperature, and flow rate of air and water), as well as factors that are difficult to model accurately, such as the surface roughness, structural deformation, and fouling of the agent itself.

[0046] To facilitate the subsequent systematic design and analysis of the controller, the aforementioned planar kinematic model of the intelligent agent underwent a state-space transformation, rewriting it as a second-order differential equation. During this process, [missing information - likely related to a specific function or process]. Class and The concept of class functions, which are widely used in nonlinear control theory to describe the gain characteristics and stability of a system. For a continuous function... ,if It is strictly increasing and satisfies , then call Class function. If Class function and satisfying Then it is called These functions are class functions. Their properties ensure that the system state can converge or remain bounded under specific conditions, laying the mathematical foundation for subsequent Lyapunov stability analysis and the application of the small gain theorem.

[0047] Furthermore, define This is the agent's output vector, which typically corresponds to the agent's actual position in the inertial coordinate system. Define the state vector. Represents the inertial position of the agent, and This represents the velocity of the agent in the body coordinate system. Define the new control input. Through this state transformation and input redefinition, the above agent planar kinematics model can be rewritten in the following standard second-order differential equation form:

[0048]

[0049] in, This represents the uncertain nonlinear term in the agent's dynamics. This term arises from the damping matrix. This is due to the uncertainty and its nonlinear dependence on velocity. To characterize the local Lipschitz property and growth limit of this nonlinear term, it is assumed that... - Direction (where) express or The direction satisfies the following properties:

[0050]

[0051] in, For any velocity scalar; It is a non-negative continuous function, mainly used to limit the gain of nonlinear terms; for This type of function ensures that the gain of the nonlinear term is strictly increasing and unbounded in the state space. For example, for a typical quadratic drag model, ,at this time Related items, and It can be taken as a linear function. This property provides a theoretical basis for the subsequent controller design based on the small gain theorem, enabling the controller to effectively handle this uncertain nonlinear term, rather than simply ignoring it as a disturbance.

[0052] After establishing the agent's dynamic model, this invention further designs a dynamic tracking controller, the goal of which is to track the agent's actual position. Capable of accurately and robustly tracking a time-varying reference signal This design effectively mitigates the uncertainties of the aforementioned model and external disturbances. The controller design first addresses position tracking errors by using a virtual controller, and then designs a real controller to handle velocity tracking errors and estimate unknown disturbances.

[0053] Considering the time-varying reference signal is This signal, generated by the subsequent obstacle avoidance controller, indicates the trajectory the agent intends to follow. -direction( The tracking error is defined as follows: ,in It is the actual location of the intelligent agent. Quantity, It is a reference signal Components. This invention first considers the speed of the intelligent agent. As a virtual control input, and defining the virtual tracking error as... ,in It is the desired virtual control input, represented by the agent in The expected speed in the direction.

[0054] Tracking error Perform time differentiation and combine with the agent's kinematic model. We can obtain:

[0055]

[0056] In order to - The error subsystem has the desired stability characteristics, and the virtual controller is designed as follows. :

[0057]

[0058] in, This is a non-decreasing, continuously differentiable positive function that acts as the feedback gain in control. This function adaptively adjusts the feedback strength based on the magnitude of the current tracking error to achieve better convergence performance and robustness. For example, Can be selected as In the form of, The gain can be a positive constant, or a piecewise function can be used to provide different gains within different error ranges. To ensure system stability and satisfy the small gain condition, The function must satisfy the following conditions:

[0059]

[0060] in, It is an arbitrarily specified local Lipschitz Class functions. These functions are used to quantify the gain and decay characteristics between the subsystems of a system, ensuring that the overall system remains stable even when they are interconnected. Specifically, Can describe The impact on the stability of the subsystem, and Then describe the velocity of the reference signal. The impact on the subsystem. By carefully selecting these functions, it can be ensured that, under appropriate conditions, The error eventually converges. For example, these... The function can be selected as or In the form of adjusting coefficients To satisfy the gain condition.

[0061] Substitute the designed virtual controller The expression can be obtained as follows: -Error subsystem:

[0062]

[0063] Next, the virtual tracking error... Perform time differentiation and combine it with the agent dynamics model. We can obtain:

[0064]

[0065] Further exploration yields... The expression is:

[0066]

[0067] in, This represents the nonlinear perturbation of an intelligent agent in a static state, which is usually an unmeasurable constant or slowly varying perturbation component, such as static friction or constant ocean current perturbation. This includes the nonlinear terms of the original system. and virtual controller The complex coupling terms generated by the derivative are specifically expressed as follows:

[0068]

[0069] because It is a non-decreasing, continuously differentiable positive function, and according to The Lipshitz property, by choosing an appropriate functional form (such as a polynomial or a sigmoid function), ensures that a function can be found. , making The gain of the term satisfies the following bounds:

[0070]

[0071] This inequality provides the necessary gain limit information for designing robust controllers, ensuring that stability analysis of the system remains feasible even in the presence of complex coupled nonlinear terms.

[0072] In order to estimate the unmeasurable disturbance term This invention designs a state observer. The observer employs a first-order low-pass filter structure, and its state equation and output equation are as follows:

[0073]

[0074] in, For the internal state of the observer, For unmeasurable terms The estimated value, This is a positive constant observer gain, the value of which determines the observer's convergence speed and its ability to suppress noise. This constant... Optimization can be achieved through offline simulation and online debugging to strike a balance between estimation accuracy and noise robustness. Typically, larger... The value can speed up the estimation convergence, but may be more sensitive to measurement noise.

[0075] The observation error is defined as Take its time derivative, and combine it with the observer equation and The expression can be obtained as follows: -Observation error subsystem:

[0076]

[0077] This subsystem clearly demonstrates the dynamic evolution of the observation error, and its convergence is affected by the observer gain. and nonlinear coupling terms The impact.

[0078] In order to - To maintain stability of the error subsystem, the present invention designs the following real controller. :

[0079]

[0080] in, Let be a non-decreasing, continuously differentiable positive function, as - Feedback gain of the error subsystem. This function is used to counteract the effects of nonlinear terms and ensure system error convergence. For example, It can be adopted The parameters are determined by ensuring the overall system has a small gain condition. To satisfy the overall system's small gain condition, The following conditions must be met:

[0081]

[0082] in, It is any specified local Lipschitz that satisfies the small gain condition. Class functions. These small gain conditions guarantee that in multiple interacting subsystems (e.g., Between these two points, the errors will not amplify each other, causing system instability. Specifically, the small gain condition includes:

[0083]

[0084] in, Let represent the identity function. These conditions ensure the stability of the closed-loop system. In practical implementations, these... functions and The functions are typically designed as linear or nonlinear functions with specific gain characteristics, and the above inequalities are satisfied through parameter adjustments. For example, these functions can be set as... In the form of, It is a positive coefficient. By choosing the appropriate To strictly satisfy the inequality.

[0085] Design the real controller Substitution The expression can be obtained as follows: -Error subsystem:

[0086]

[0087] At this point, - Direction, the present invention obtained by -Error Subsystem -Error subsystem and - A set of dynamic equations composed of the observation error subsystem.

[0088] Based on the above design results, - Direction: The dynamic tracking controller designed in this invention consists of the following set of equations, which work together to achieve tracking for the intelligent agent:

[0089]

[0090] Under the action of this dynamic tracking controller, the closed-loop system is transformed into the following error-correlated system:

[0091]

[0092] To analyze the input-to-state stability (ISS) of this error-correlated system, three Lyapunov functions are introduced. For - Error subsystem, the Lyapunov function can be chosen as .when When, its time derivative satisfies:

[0093]

[0094] This inequality shows that, under certain conditions, - The error subsystem is input-to-state stable, and the error is a function of... The defined rate converges.

[0095] for - Error subsystem, the Lyapunov function can be chosen as .when When, its time derivative satisfies:

[0096]

[0097] This inequality shows that, The error subsystem is input-to-state stable when certain conditions are met, and the error is expressed as a function of... The defined rate converges.

[0098] for -Observation error subsystem, the Lyapunov function can be chosen as .when When, its time derivative satisfies:

[0099]

[0100] This inequality shows that, The observation error subsystem is input to a state-stable system, and the error convergence rate is determined by the observer gain. The decision is made. In the Lyapunov analysis above, "ae" means "almost everywhere," which is a common representation when absolute value functions are Lyapunov functions, implying that the derivative satisfies this condition outside of non-differentiable points (such as the origin).

[0101] In summary, under the action of the dynamic tracking controller, the closed-loop system is transformed into an error-correlated system, and its correlation gain satisfies the multi-loop nonlinear small-gain condition, such as... Figure 2 As shown. That is, in -direction( This dynamic tracking controller can ensure that the closed-loop system operates at a constant speed. As input, The state is input to a stable state. This proves that the designed controller can still enable the agent to accurately track the reference trajectory even in the presence of uncertain nonlinear dynamics and external disturbances.

[0102] Based on this, the present invention designs an obstacle avoidance controller, the core function of which is to generate a safe, time-varying speed command. This command serves as a reference signal for the aforementioned tracking controller, thereby ensuring that the agent can actively avoid stationary obstacles while tracking the desired trajectory. This obstacle avoidance controller achieves real-time obstacle avoidance by constructing a feasible velocity set and projecting velocity commands, without relying on an accurate obstacle model or complex online optimization.

[0103] To generate the time-varying reference signal required by the tracking controller, consider this signal. It is generated by the following first-order integrator system:

[0104]

[0105] in, The state of the first-order integrator represents the instantaneous position that the agent expects to reach. The input is inherently bounded, representing the speed command generated by the obstacle avoidance controller and subject to safety constraints.

[0106] Consider a stationary obstacle on the agent's path, whose position is precisely denoted as . In order for the agent to safely avoid the obstacle, the following safety conditions must be met: for a given initial condition ,in The relative distance between the agent and the obstacle is for all All must meet ,in This is the minimum safe distance between the desired agent and the obstacle to avoid collision. The initial distance is typically determined by a combination of factors, including the agent's size, sensor accuracy, the conservatism of the obstacle avoidance strategy, and system response time. It should be greater than or equal to This is to ensure that the device is in a safe state from the beginning and to avoid an initial collision.

[0107] Define the relative position vector between the agent and the obstacle as: Define the relative position vector between the first-order integrator state (i.e., the reference position) and the obstacle as follows: Then we have:

[0108]

[0109] To achieve obstacle avoidance directly at the speed level, this invention first designs a speed... feasible set This feasible set defines the current relative position of the agent to the obstacle. The range of safe speed commands that an intelligent agent can take. Feasible set. Defined as:

[0110]

[0111] in, It is the unit direction vector of the agent relative to the obstacle, pointing in the direction of the line connecting the center of the obstacle to the center of the agent. For a collision-free margin, and it must satisfy... Its value is greater than or equal to the actual safe distance. This is to provide an additional safety buffer for obstacle avoidance, and its specific value can be adjusted according to the agent's movement speed, response time, and sensor detection range.

[0112] function For a strictly decreasing, strictly convex and satisfying and A continuously differentiable function. For example, Can be selected as Functions of the form, where The coefficient is positive. This function is effective when the agent approaches an obstacle (i.e., ...). (If the value decreases), it will increase rapidly, thereby increasing the urgency of obstacle avoidance and forming a mathematical description of "repulsive force".

[0113] function For a strictly increasing condition that satisfies Locally Lipschitz continuous functions. For example, Can be selected as or A function of the form [etc.]. This function will […]. The "obstacle avoidance strength" generated by the function is converted into a velocity component.

[0114] The positive constants chosen in the design define the functions. and The domain and range of the function are defined to ensure its physical meaning and mathematical properties. For example, the domain and range can be set. For a small positive number, ensure exist It is defined when the value is negative but not excessively large.

[0115] Based on the above feasible set The obstacle avoidance controller is designed to project speed commands in real time.

[0116]

[0117] in, In order to ensure A bounded and essentially bounded speed command signal, which is usually generated by a higher-level task planner of the agent, represents the agent's desired direction and speed of movement in the absence of obstacles.

[0118] When the task planner generates the raw speed command Located in feasible set When the obstacle avoidance condition is met (i.e., the obstacle avoidance condition is satisfied), it means the instruction is safe, and the agent directly adopts it. Used as a reference speed.

[0119] when Not located in the feasible set Internal time (i.e., if directly adopted) (potentially causing collision with obstacles), the obstacle avoidance controller will Project to feasible set On the boundary. Projected velocity command. It consists of two parts: one part is the reverse avoidance velocity component along the direction of the agent relative to the obstacle. The magnitude of this component is determined by the distance between the agent and the obstacle; the closer the distance, the greater the avoidance speed. Its physical meaning is the normal "thrust" applied by the agent to avoid the obstacle. The other part is perpendicular to this relative direction, maintaining the original command. Invariant velocity components .in, for The identity matrix. This projection mechanism ensures that the original task instruction is preserved as much as possible while avoiding obstacles, that is, the velocity components parallel to the obstacle avoidance boundary are retained to maintain task continuity. This obstacle avoidance controller is conceptually similar to a dynamic safety barrier, which is activated and corrects the velocity instruction only when the agent approaches the safety boundary. This avoids the problem of traditional artificial potential functions easily getting trapped in local optima, and its computational complexity is much lower than that of nonlinear optimization methods, which can meet the requirements of real-time obstacle avoidance.

[0120] Under the control of the designed tracking controller and obstacle avoidance controller, appropriate parameters are selected based on the motion state of the agent to complete the obstacle avoidance control. The overall block diagram of the closed-loop system is shown below. Figure 3 .

[0121] Example 2

[0122] The overall implementation process of the tracking-based uncertain nonlinear intelligent agent obstacle avoidance control method proposed in this invention is a multi-level, collaborative control system. The core logic of the system is implemented through a central control unit (MCU) or embedded system, which includes a high-speed processor (e.g., a multi-core ARM processor) and a dedicated digital signal processor (DSP) or field-programmable gate array (FPGA).

[0123] First, the sensor system on the intelligent agent (such as lidar, millimeter-wave radar, visual sensor array, GNSS / IMU integrated navigation system, etc.) continuously acquires the intelligent agent's current position in real time. and the location of obstacles These raw data undergo filtering, denoising, and coordinate transformation by a preprocessing unit, such as Kalman filtering or extended Kalman filtering, to obtain high-precision, low-latency agent attitude and position information, as well as accurate position information of obstacles in the agent coordinate system or inertial coordinate system.

[0124] The central control unit uses the location of the intelligent agent obtained by the sensors. and obstacle location It can calculate the relative position of the agent and obstacles in real time. Subsequently, the obstacle avoidance controller module is activated, which, according to... and preset functions To compute the feasible set in real time The mission planning module generates an initial desired speed command based on the current mission objective (e.g., cruise, stationary navigation, path tracking). Obstacle avoidance controller judgment Is it in If it is inside, then output directly. If not included, a corrected safe speed command is calculated according to the projection mechanism described above. , which serves as the reference speed for the tracking controller.

[0125] This safe speed command It is input into the tracking controller as a time-varying reference signal. The source of its generation, namely The tracking controller module then adjusts the speed based on the agent's current velocity. Expected speed and observer state and estimated disturbance The control input of the intelligent agent is calculated in real time according to the dynamic tracking controller equations. Among them, the function and The specific form can be implemented as a piecewise linear function, a sigmoid function, or a polynomial function, with its parameters preset through offline analysis and simulation to meet the system's small gain condition. Observer gain It is also preset as a positive definite constant, typically determined through system identification and robustness analysis, for example, by optimizing it through Monte Carlo simulations at different disturbance intensities and noise levels. These parameters are stored in the non-volatile memory (such as EEPROM or flash memory) of the control unit and loaded at system startup.

[0126] Finally, the calculated control input The actuator drive module is sent to the intelligent agent. For example, in an autonomous vehicle, this drive module will... The torque or steering angle commands are converted into wheel torque commands and sent to the motor driver and steering actuator via CAN bus or Ethernet. For drones, these commands are converted into propeller speed commands or rudder deflection angle commands and sent to the electronic speed controller (ESC) or servo motor via PWM signals or digital communication protocols. These commands drive the intelligent agent to complete the actual movement, achieving accurate tracking of the reference trajectory and effective obstacle avoidance. A schematic diagram of an intelligent agent obstacle avoidance scenario is shown below. Figure 5 As shown.

[0127] The entire control process is executed periodically at a high frequency (e.g., 100Hz to 1000Hz, depending on the agent's dynamic response speed and environmental dynamics) to ensure the system has sufficient real-time response capability to cope with dynamic environmental changes and model uncertainties. For example, for high-speed drones, the control frequency may be as high as 1kHz, while for low-speed unmanned vehicles, 200Hz may be sufficient. Through this hierarchical control architecture, the obstacle avoidance controller focuses on high-level safety decisions, while the tracking controller is responsible for low-level precise motion execution and uncertainty handling, achieving functional decoupling and performance optimization, and greatly enhancing the agent's autonomy and reliability in complex and uncertain environments.

[0128] Example 3

[0129] The following is an example: Consider a differential equation model in -direction( )satisfy

[0130]

[0131] in, Using the control design method of this invention, the corresponding parameters and functions in the tracking controller can be designed as follows: , , ,in , The corresponding parameters and functions in the obstacle avoidance controller can be designed as follows: , , , Consider an unmanned surface vessel with inertial and damping matrices respectively. and This verifies that, through transformation, the water surface vehicle model meets the requirements of a differential equation model. Therefore, by selecting the starting point of the water surface vehicle... and target point The location of the obstacle Expected non-collision distance The trajectory of a surface vehicle on a plane is as follows: Figure 4 As shown in the figure, the motion trajectory demonstrates that the control method of this invention enables an uncertain nonlinear intelligent agent to achieve obstacle avoidance control.

[0132] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A tracking-based uncertain nonlinear agent obstacle avoidance control method, characterized in that, The method comprises: Step S1. Establishing an uncertain nonlinear kinematic model of the agent system, the kinematic model containing an inertial position, a body velocity, an inertial matrix in a diagonal form and a velocity-related uncertain damping matrix, and converting it into a first-order differential equation form containing uncertain nonlinear terms, the nonlinear terms satisfying a local Lipschitz-like property; Step S2. Designing an observer-based tracking controller, so that the closed-loop system is converted into a coupling system containing a tracking error, a virtual tracking error and an observation error, and by setting a feedback gain function satisfying a multi-loop nonlinear small gain condition, the system input-to-state stability is ensured to make the agent track a time-varying reference signal; Step S3. Designing an obstacle avoidance controller based on a first-order integrator, constructing a velocity feasible set based on the relative position of the agent and the static obstacle, and performing real-time projection correction on the velocity command of the high-level planning to generate a safe time-varying reference signal, so as to ensure that the agent actively avoids obstacles in the tracking process; Step S4. Selecting appropriate parameters to complete the obstacle avoidance control.

2. The tracking-based uncertain nonlinear agent obstacle avoidance control method according to claim 1, wherein, In the kinematic model, the inertial matrix and the damping matrix are both diagonal matrices, wherein the diagonal elements of the damping matrix are nonlinear functions dependent on the corresponding velocity components, and the uncertainty thereof is derived from the random fluctuations of the environmental medium and the factors difficult to accurately model by the agent itself.

3. The tracking-based uncertain nonlinear agent obstacle avoidance control method according to claim 1 or 2, characterized in that, The control input of the transformed differential equation is the original control force and torque vector left multiplied by the inverse of the inertia matrix, and the non-linear term is determined by the inverse of the inertia matrix, the damping matrix and the velocity vector. The model establishment introduces class and class function to define the class local Lipschitz property of the non-linear term.

4. The tracking-based uncertain nonlinear agent obstacle avoidance control method according to claim 3, wherein, The virtual controller is designed as a negative feedback form of tracking error, whose feedback gain function must satisfy an inequality condition composed of multiple Class functions to ensure the stability of the first error subsystem.

5. The tracking-based uncertain nonlinear agent obstacle avoidance control method according to claim 4, wherein, The state observer employs a first order dynamic structure whose internal states and outputs are used to generate an estimate of the unmeasurable disturbance term; the real controller is a combination of a negative feedback of a virtual tracking error and the estimate, whose feedback gain function is chosen to satisfy another inequality condition formed by a set of quadratic functions, and this condition together with the condition of the virtual controller forms a multi-loop nonlinear small gain condition.

6. The tracking-based uncertain nonlinear agent obstacle avoidance control method according to claim 5, wherein, The multi-loop small gain condition comprises three sets of function composite inequalities, which together ensure the input-to-state stability of the closed-loop system.

7. The tracking-based uncertain nonlinear agent obstacle avoidance control method according to claim 6, wherein, The dynamic tracking controller is composed of the equations of a virtual controller, a real controller and a state observer, which ensures that the closed-loop system is input-to-state stable with the reference signal derivative as the input and the tracking error, the virtual tracking error and the observation error as the states.

8. The tracking-based uncertain nonlinear agent obstacle avoidance control method according to claim 1, wherein, The reference signal is generated by a first-order integrator system; the obstacle avoidance controller defines the relative position of the agent and the obstacle and sets an obstacle avoidance margin not less than the expected safety distance; and the velocity feasible set is constructed based on the combination of a strictly decreasing strictly convex potential function and a strictly increasing scaling function to constrain the safe velocity command.

9. The tracking-based uncertain nonlinear agent obstacle avoidance control method according to claim 8, wherein, The obstacle avoidance controller adopts a segmented projection mechanism: when the original velocity command generated by the high-level task planner is located within the velocity feasible set, the command is directly adopted; and when the original velocity command is located outside the feasible set, it is projected to the boundary of the feasible set, and the projected command is composed of an avoidance component in the repulsion direction of the obstacle and an original command component perpendicular to the direction.

10. The tracking-based uncertain nonlinear agent obstacle avoidance control method according to claim 1, wherein, The implementation process of the method comprises: the sensor system acquires the position information of the agent and the obstacle in real time; the central control unit calculates the relative position and generates a safe velocity command by the obstacle avoidance controller module; the command is input to the tracking controller module as a reference signal; the tracking controller module combines the observer state and the disturbance estimation to calculate the final control input and sends it to the actuator; wherein the gain parameters of each controller and the observer gain are preset through offline analysis to meet the small gain stability condition of the system.

Images