A multi-agent and obstacle state cooperative observation method based on leader probe information

Abstract

The application discloses a kind of multi-agent and obstacle state cooperative observation method based on leader probe information, simultaneously consider the unmeasurable state contained in multi-agent system, internal and external uncertainty and safety obstacle avoidance problem in cooperative control, utilize as few as possible network transmission information, for follower design a kind of distributed compound state observer, let each follower can simultaneously realize high-precision estimation to leader's all information, unmeasurable state and unknown disturbance item of itself and relative state information between obstacle, provide effective data for subsequent realization of cluster safety cooperative formation and obstacle avoidance control strategy design.The cooperative observation strategy proposed in the application for nonlinear multi-agent system is based on the relative displacement information between the leader and the obstacle, and the cooperative observation of multiple unmeasurable states in the multi-agent system is realized using less interaction information, which can be widely applied to other multi-agent cooperative observation systems.

Description

Technical Field

[0001] This invention belongs to the field of multi-agent safety cooperative control, and specifically relates to a multi-agent and obstacle state cooperative observation method based on leader detection information. Background Technology

[0002] With the development of artificial intelligence, robotics, and automation technologies, the perception, decision-making, and execution capabilities of individual intelligent agents are constantly improving. However, their limitations are gradually being exposed in complex task scenarios: individual intelligent agents have inherent limitations in computing power, endurance, perception range (such as sensor detection distance), and physical operational capabilities (such as load and operational range). For example, a single drone struggles to complete real-time monitoring of a large area, a single industrial robot cannot efficiently handle multi-process collaborative production tasks, and a single space satellite cannot meet global perception and detection needs. Multi-agent systems overcome the limitations of single intelligent agents through "division of labor and cooperation," possessing advantages such as strong robustness, high efficiency, high flexibility, and wide coverage. For example, the efficiency of 10 drones coordinating a search is far higher than that of a single drone operating alone, and the task can continue to be completed through task reassignment when some drones malfunction. This contradiction between "limited capabilities of single intelligent agents" and "increasing demands of complex tasks" has driven research into multi-agent cooperative control technology—achieving overall efficiency of "1+1>2" by coordinating the behavior of multiple intelligent agents.

[0003] However, multi-agent systems face two core challenges in practical applications: the uncertainty of dynamic disturbances and the obstacle avoidance requirements of multi-agent interactions. On the one hand, when agents operate in complex environments, they are inevitably affected by external disturbances (such as changes in wind speed, fluctuations in ground friction, and communication delays) and internal uncertainties (such as model parameter errors and actuator noise). These disturbances can cause agents to deviate from their expected trajectories, reduce the accuracy of collaborative control, and even lead to system instability. On the other hand, during collaborative operations, multi-agent systems need to avoid collisions with obstacles (static or dynamic) and other agents within a limited space, while maintaining the continuity of the overall task. This places stringent requirements on real-time decision-making and path planning.

[0004] Perturbation estimation is crucial for improving a system's robustness against interference. Traditional methods such as extended Kalman filtering and sliding mode observers are widely used for single-agent perturbation estimation. However, in multi-agent systems, perturbations are spatiotemporally correlated (e.g., drones in the same area are affected by the same wind field), making independent estimation by a single agent difficult to utilize global information, leading to insufficient accuracy. In recent years, cooperative perturbation estimation has become a research focus, with researchers designing distributed observers to estimate perturbation information from multiple agents. However, existing cooperative estimation methods largely rely on ideal communication conditions, and their robustness in scenarios with limited communication (e.g., latency, packet loss) or rapidly changing perturbations still needs improvement. Multi-agent obstacle avoidance requires a balance between safety and task efficiency. Existing methods can be divided into centralized and distributed approaches. Centralized methods (such as path optimization based on mixed-integer programming) achieve optimal obstacle avoidance through global planning, but the computational complexity increases exponentially with the number of agents, making it difficult to apply to large-scale systems. Distributed methods (such as behavior-based obstacle avoidance and optimal collision avoidance algorithms) achieve real-time obstacle avoidance through local information, offering high computational efficiency, but local decisions may lead to a decline in global task performance. In recent years, researchers have combined reinforcement learning and deep learning techniques to design end-to-end obstacle avoidance strategies, which have improved adaptability in dynamic environments. However, how to combine obstacle avoidance behavior with collaborative control objectives remains an unsolved problem.

[0005] Therefore, how to effectively estimate the unmeasurable state, internal and external interference, and obstacle state information in a multi-agent system using as little communication information as possible is a prerequisite for safe and cooperative control of multi-agent systems, and it is also the key technical problem that this invention patent focuses on solving and breaking through. Summary of the Invention

[0006] Purpose of the Invention: To overcome the above shortcomings, the purpose of this invention is to provide a multi-agent and obstacle state collaborative observation method based on leader detection information. For leader-follower multi-agent system collaborative or formation tasks involving moving obstacles, this invention aims to design a composite state observer for followers using minimal network transmission information. This allows each follower to simultaneously achieve high-precision estimation of all information about the leader, its own unmeasurable state and unknown disturbances, as well as its relative state with the obstacle. This provides effective information for the subsequent design of safe collaborative formation and obstacle avoidance control strategies to meet engineering application needs.

[0007] Technical Solution: To achieve the above objectives, this invention provides a multi-agent and obstacle state collaborative observation method based on leader detection information, comprising the following steps: Step 1: Establish a multi-agent and obstacle dynamics model; Consider collaborative observation among n+1 agents, including one leader and n followers; establish dynamic models of the leader agent, follower agents, and dynamic obstacles; provide a directed topology for the n+1 agents to represent the communication relationships between them; follower agents adjacent to the leader can obtain the leader's transmission information, but the leader cannot obtain the follower agents' state information; only the leader can detect its own relative displacement to obstacles. Step 2: Establish a tracking error system; Based on the dynamic model established in the first step, a tracking error system between the i-th follower and the leader, and a relative error system between the i-th agent and the obstacle are established. Step 3: Construct an expanded version of the leader system and the error tracking system; Expansion states are introduced for the leader dynamics system, the tracking error system, and the relative error system, and expansion forms of each system are constructed. Step 4: Design a distributed composite state observer; Based on the expanded systems obtained in the third step, a distributed composite state observer is designed for the i-th follower, enabling each follower agent to obtain all state information of the leader, its own speed and unknown interference information, as well as the relative state information with the obstacle.

[0008] Furthermore, throughout the observation process, each agent only needs to exchange displacement-level state information, without needing to transmit velocity, acceleration, or perturbation information. Compared to traditional methods, this reduces the network transmission burden and effectively mitigates network problems such as communication latency, congestion, and packet loss. Simultaneously, through a single observer framework, it achieves collaborative estimation of the leader's entire state (displacement, velocity, control input), its unknown perturbations, and its relative state with obstacles, integrating the task that traditionally required multiple observers to complete into one.

[0009] Furthermore, in the first step, the dynamic models of the leader agent and the i-th follower agent are as follows: In the formula, x 01 and x 02 f(x) represents the displacement and velocity states of the leader agent, respectively. 01 ,x 02 d0) represents the total internal and external disturbances of the leader agent system; u0 is the control input of the leader agent; x i1 and x i2 f(x) represents the displacement and velocity states of the i-th follower agent, respectively; i1 ,xi2 ,d i ) represents the total internal and external disturbances of the i-th follower agent system; u i This is the control input for the i-th follower agent.

[0010] Furthermore, in the first step, the dynamic model of the dynamic obstacle is as follows: In the formula, x d1 and x d2 f(x) represents the displacement and velocity states of the obstacle, respectively; d1 ,x d2 ,d d ) represents the total internal and external disturbances of the obstacle; u d This serves as the control input for obstacles. Obstacles and agents use the same dynamic description framework, simplifying the observer design logic and eliminating the need to design heterogeneous observers for different types of objects.

[0011] Furthermore, in the first step, in the directed topology G=(ν,ϵ,A), ν represents the set of nodes; ϵ ⊆ ν×ν represents the set of edges; the edge (i,j) in the directed topology represents the transmission of information from agent i to agent j; A=[a ij Let be the connection matrix. If (i,j)∈ϵ, then a ij =1, otherwise a ij =0; b = 0 when agent i is adjacent to the leader. i =1, otherwise b i =0. Using (i,j) to represent the transmission of information from agent i to agent j supports directed communication, which is more in line with practical engineering scenarios than undirected topology (which requires bidirectional communication).

[0012] Furthermore, in the second step, let the tracking error variables of the i-th follower agent relative to the leader's displacement and velocity states be: The tracking error system between the i-th follower and the leader is: .

[0013] Furthermore, in the second step, let the relative displacement and velocity state variables between the i-th agent and the obstacle be: The relative error system between the i-th agent and the obstacle is: .

[0014] Furthermore, in the third step, new state variables are set for the leader dynamics system. and The leader system was expanded to: New state variables are set for the tracking error system. and The tracking error system is expanded to: For a relative error system, a new state variable is set. and The relative error system is expanded to: .

[0015] Furthermore, in the fourth step, the distributed composite state observer is: in, , , , , , , , , These are states x 01 x 02 x 03 x i e1 x i e2 x i e3 x i q1 x i q2 x i q13 The estimated value; a i 1. a i 2. a i 3. a i 4. a i 5. a i 6. a i 7. a i 8. a i 9 represents the gain parameter of the observer; where e i For the leader displacement estimation error combination, η i For the tracking error estimation combination, γ iThis is a combination of relative displacement estimation errors. Each follower operates its own observer independently, eliminating the need for a central node to aggregate global information, and the system communication load does not increase dramatically with the number of agents. Simultaneously, nine gain parameters are set, each corresponding to one of the nine state estimation channels, providing multiple degrees of freedom for engineering debugging.

[0016] Furthermore, in the fourth step, Where, x q1 The relative displacement between the leader and obstacles is obtained through detection. The tracking error information and the leader displacement estimation information are transmitted to the driver of the obstacle relative state observer, so that the relative state estimation of the obstacles and the tracking estimation of the leader are completed collaboratively within the same observer framework.

[0017] Furthermore, define the estimation error variable: , , , , , , , , By designing the observer gain parameters, all estimation errors are made to approach zero. Each estimation error variable is defined as a computable quantity (true value minus estimated value), which can be directly calculated and monitored during actual system construction, facilitating engineering verification.

[0018] As can be seen from the above technical solution, the present invention has the following beneficial effects: 1. This invention provides a multi-agent and obstacle state collaborative observation method based on leader detection information. Addressing the inherent broadband constraint problem of communication networks in multi-agent systems, this invention utilizes the directed topology among multiple agents. Adjacent agents only need to transmit limited displacement-level state information through the communication network to achieve real-time estimation of multiple state variables, reducing network transmission burden and effectively mitigating network problems such as communication transmission delay, congestion, and packet loss. 2. This invention provides a multi-agent and obstacle state collaborative observation method based on leader detection information. Addressing the multi-agent safe collaborative control problem, this invention designs a distributed composite state observer. Based on the relative state information between the leader and obstacles obtained through leader detection, and the leader displacement information estimated by neighboring agents, each agent can simultaneously achieve high-precision estimation of all leader states, control inputs, all states and unknown nonlinear terms of the tracking error system between the leader and the leader, as well as the relative state and control input between the leader and obstacles. Attached Figure Description

[0019] Figure 1 This is a flowchart of a multi-agent and obstacle state collaborative observation method based on leader detection information according to the present invention; Figure 2 This is a block diagram of a multi-agent and obstacle state collaborative observation method based on leader detection information according to the present invention. Figure 3 This is a directed topology diagram of inter-agent communication in this invention; Figure 4 The simulation results of this invention show the estimation error curves of the leader's displacement state by the four follower agents. Figure 5 The simulation results of this invention show the estimation error curves of the leader's velocity state by the four follower agents. Figure 6 The simulation results of this invention show the estimation error curves of the four follower agents to the leader's control input. Figure 7 The above are the estimation error curves of the relative displacement state between the four follower agents and the leader in the simulation results of this invention. Figure 8 The estimated error curves of the relative velocity state between the four follower agents and the leader in the simulation results of this invention are shown. Figure 9 The simulation results of this invention show the estimation error curves for the unknown perturbations of the four follower agents. Figure 10 The estimated error curves of the relative displacement state between the four follower agents and the obstacle in the simulation results of this invention are shown. Figure 11 The estimated error curves of the relative velocity state between the four follower agents and the obstacle in the simulation results of this invention are shown. Figure 12 The figures show the estimation error curves of the four follower agents for unknown inputs to obstacles in the simulation results of this invention. Detailed Implementation

[0020] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention. Example

[0021] In this embodiment, as Figure 1 and Figure 2This invention discloses a multi-agent and obstacle state collaborative observation method based on leader detection information, comprising the following steps: Step 1: Establish a multi-agent and obstacle dynamics model In this example, consider a group of 5 agents, including 1 leader and 4 follower agents. Furthermore, consider the presence of a moving obstacle. Let the dynamic models of the leader and the i-th follower agent be as follows: (1) (2) In the formula, x 01 and x 02 These represent the displacement and velocity states of the leader agent, respectively; u0 is the control input of the leader agent; x i1 and x i2 f(x) represents the displacement and velocity states of the i-th follower agent, respectively; i1 , x i2 , d i ) represents the total internal and external disturbances of the i-th follower agent system; u i This is the control input for the i-th follower agent. In this example, the leader's initial system state is set to x. 01 (0)=0、x 02 (0) = 0, and the control input u0 = 1. The total internal and external disturbances of the follower agent system are denoted as f(x). i1 , x i2 , d i )=0.01 x i1 x i2 + d i The external disturbances experienced by the four follower agents are as follows: , , , And the initial states are x i1 (0)=0、x i2 (0)=0.

[0022] Let the dynamic model of the obstacle be: (3) In the formula, x d1 and x d2 These represent the displacement and velocity states of the obstacle, respectively; u d This is the control input for the obstacle. In this example, the initial state of the obstacle is set to x. d1 (0)=0、x d2 (0) = 0, and the control input ud =0.2.

[0023] In this invention, a directed topology G=(ν,ϵ,A) is provided for n+1 multi-agent systems to represent the communication relationships between agents. Here, ν represents the set of nodes; ϵ ⊆ ν×ν represents the set of edges; if two agents are connected by an edge, they are considered adjacent agents, and adjacent agents can communicate with each other; the edge (i,j) in the directed topology represents the transmission of information from agent i to agent j; A=[a ij ] represents the connection matrix, where if (i,j)∈ϵ, then a ij =1, otherwise a ij =0. Follower agents adjacent to the leader can obtain the leader's transmitted information, but the leader cannot obtain the follower agents' state information. When agent i is adjacent to the leader, b i =1, otherwise b i =0. Furthermore, only the leader can detect its relative displacement to obstacles. In this example, the given undirected topology of the five multi-agent teams is as follows: Figure 3 As shown, the connection matrix can be obtained. And b1=1, b2=b3=b4=0.

[0024] Step 2: Establish a tracking error system; Let the tracking error variables of the i-th follower agent relative to the leader's displacement and velocity states be respectively... (4) Let the relative displacement and velocity state variables between the i-th agent and the obstacle be respectively... (5) Combining equations (1) and (2), we can obtain the following tracking error system between the i-th follower and the leader according to equation (4): (6) Similarly, by combining equations (1), (2), and (3), and according to equation (5), the relative error system between the i-th agent and the obstacle can be obtained as follows: (7).

[0025] Step 3: Construct an expanded version of the leader system and the error tracking system; For the leader dynamics system (1), in this invention, a new state variable is set. and Then the original system (1) is expanded to the following form: (8) In response to the tracking error system (6) between followers and leaders, this invention introduces a new state variable. and Then the original tracking error system (6) is expanded to the following form: (9) Similarly, for the relative error system (7) between the agent and the obstacle, in this invention, a new state variable is set. and Then the original relative error system (7) is expanded to the following form: (10).

[0026] Step 4: Design a distributed composite state observer; In this invention, for the extended systems (8), (9), and (10), the i-th follower is designed with a distributed composite state observer of the following form: (11) In the formula: ; ; ; in, , , , , , , , , These are states x 01 x 02 x 03 x i e1 x i e2 x i e3 x i q1 x i q2 x i q13 The estimated value; a i 1. a i 2. a i 3. a i 4. a i 5. a i 6. a i 7. a i 8. a i 9 represents the gain parameter of the observer; x q1The relative displacement between the leader and the obstacle is detected. In this example, the distributed composite state observer (11) estimates the disturbances and states of different subsystems hierarchically through three cascaded extended state observers, using pole configuration as the skeleton, and combined with a 100Hz sampling frequency. The preset bandwidth corresponding to the first-order gain of each observer is set to... , and According to observer parameters , , Base value (i.e., , , The gain parameters of the observers are set to a, and each channel is independently fine-tuned according to its noise sensitivity and immunity requirements. i 1 = -60, a i 2 = -1200, a i 3 = -8000, a i 4=75、a i 5=2000, a i 6=18000, a i 7=50、a i 8=800, a i 9 = 4000.

[0027] In this invention, the following estimation error variable is defined: , , , , , , , , Combining equations (8), (9), (10) and (11), we can obtain the following estimation error system: (12) Under the action of the above observer, each follower agent can obtain all the state information of the leader (including displacement, velocity and control input), its own velocity and unknown interference information, and the relative state information with the obstacle, which ensures that all estimation errors in the estimation error system (12) tend to zero.

[0028] Simulation results are as follows Figures 4 to 12 As shown. Among them, Figures 4 to 6 The figures show the estimation error curves of the four follower agents for the leader's displacement, velocity, and control input. Figures 7 to 9 The figures show the estimation error curves for the relative displacement, relative velocity, and unknown perturbations between the four follower agents and the leader. Figures 10 to 12The figures show the relative displacement, relative velocity, and estimation error curves for unknown inputs from the obstacles between the four follower agents and the obstacles. Figures 4 to 12 As can be seen, all estimation errors converge to zero within a finite time, verifying the effectiveness of the cooperative observation method proposed in this invention. The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A multi-agent and obstacle state collaborative observation method based on leader detection information, characterized in that: Includes the following steps: Step 1: Establish a multi-agent and obstacle dynamics model; Consider collaborative observation among n+1 agents, including one leader and n followers; establish dynamic models of the leader agent, follower agents, and dynamic obstacles; provide a directed topology for the n+1 agents to represent the communication relationships between them; follower agents adjacent to the leader can obtain the leader's transmission information, but the leader cannot obtain the follower agents' state information, and only the leader can detect its own relative displacement with respect to obstacles; Step 2: Establish a tracking error system; Based on the dynamic model established in the first step, a tracking error system between the i-th follower and the leader, and a relative error system between the i-th agent and the obstacle are established. Step 3: Construct an expanded version of the leader system and the error tracking system; Expansion states are introduced for the leader dynamics system, the tracking error system, and the relative error system, and expansion forms of each system are constructed. Step 4: Design a distributed composite state observer; Based on the expanded systems obtained in the third step, a distributed composite state observer is designed for the i-th follower, enabling each follower agent to obtain all state information of the leader, its own speed and unknown interference information, as well as the relative state information with the obstacle.

2. The multi-agent and obstacle state collaborative observation method based on leader detection information according to claim 1, characterized in that: In the first step, the dynamic models of the leader agent and the i-th follower agent are as follows: In the formula, x 01 and x 02 f(x) represents the displacement and velocity states of the leader agent, respectively. 01 ,x 02 (d0) represents the total internal and external disturbances of the leader agent system, where d is the external disturbance experienced by the leader agent; u0 is the control input of the leader agent; x i1 and x i2 f(x) represents the displacement and velocity states of the i-th follower agent, respectively; i1 ,x i2 ,d i ) represents the total internal and external disturbances of the i-th follower agent system, d i The external disturbance experienced by the i-th follower agent; u i This is the control input for the i-th follower agent.

3. The multi-agent and obstacle state collaborative observation method based on leader detection information according to claim 1, characterized in that: In the first step, the dynamic model of the dynamic obstacle is as follows: In the formula, x d1 and x d2 f(x) represents the displacement and velocity states of the obstacle, respectively; d1 ,x d2 ,d d ) represents the total internal and external disturbances of the obstacle, d d External disturbances affecting the obstacle; u d For the control input of obstacles.

4. The multi-agent and obstacle state collaborative observation method based on leader detection information according to claim 1, characterized in that: In the first step, in the directed topology G=(ν,ϵ,A), ν represents the set of nodes; ϵ ⊆ ν×ν represents the set of edges; the edge (i,j) in the directed topology represents the transmission of information from agent i to agent j; A=[a ij Let be the connection matrix. If (i,j)∈ϵ, then a ij =1, otherwise a ij =0; the adjacency matrix of the leader and followers is B=diag{b1,b2,…,b n When the i-th follower is adjacent to the leader, b i =1, when the i-th follower is not adjacent to the leader, b i =0.

5. The multi-agent and obstacle state collaborative observation method based on leader detection information according to claim 2, characterized in that: In the second step, let the tracking error variables of the i-th follower agent relative to the leader's displacement and velocity states be: The tracking error system between the i-th follower and the leader is: 。 6. The multi-agent and obstacle state collaborative observation method based on leader detection information according to claim 2 or 3, characterized in that: In the second step, let the relative displacement and velocity state variables between the i-th agent and the obstacle be: The relative error system between the i-th agent and the obstacle is: 。 7. The multi-agent and obstacle state collaborative observation method based on leader detection information according to claim 1, characterized in that: In the third step, new state variables are set for the leader dynamics system. and The leader system was expanded to: New state variables are set for the tracking error system. and The tracking error system is expanded to: For a relative error system, a new state variable is set. and The relative error system is expanded to: 。 8. The multi-agent and obstacle state collaborative observation method based on leader detection information according to claim 1, characterized in that: In the fourth step, the distributed composite state observer is: in, , , , , , , , , These are states x 01 x 02 x 03 x i e1 x i e2 x i e3 x i q1 x i q2 x i q13 The estimated value; a i 1. a i 2. a i 3. a i 4. a i 5. a i 6. a i 7. a i 8. a i 9 represents the gain parameter of the observer; where e i For the leader displacement estimation error combination, η i For the tracking error estimation combination, γ i This represents the combination of relative displacement estimation errors.

9. The multi-agent and obstacle state collaborative observation method based on leader detection information according to claim 4 or 8, characterized in that: In the fourth step, Where, x q1 The relative displacement between the leader and the obstacle is detected.

10. The multi-agent and obstacle state collaborative observation method based on leader detection information according to claim 8, characterized in that: Define the estimation error variable: , , , , , , , , By designing the observer gain parameters, all estimation errors can be made to approach zero.

Images