A cooperative navigation method for heterogeneous multi-agent systems

Abstract

This invention discloses a heterogeneous multi-agent cooperative navigation method based on the Three-Zero Path Planning (TZPP) paradigm, satisfying the constraints of zero training, zero prior knowledge, and zero simulation. It employs a coordinator-explorer heterogeneous agent architecture, with a humanoid robot as the coordinator and a quadruped robot as the explorer. This enables autonomous exploration and path planning in unknown and complex environments, eliminating the dependence on training, prior maps, and simulation environments found in existing methods. This significantly reduces the deployment and application costs of the robot system and improves its robustness in real dynamic environments. Furthermore, it achieves specialized task division between the humanoid robot and the quadruped robot. The humanoid robot is responsible for high-level task coordination and navigation decisions, while the quadruped robot utilizes its flexible mobility to undertake environmental exploration and feasible path identification. This fully leverages the complementary advantages of different robot forms, solving the problems of chaotic planning and resource waste in homogeneous multi-agent cooperative systems.

Description

Technical Field

[0001] This invention relates to the field of robot cooperative navigation technology, specifically a cooperative navigation method for heterogeneous multi-agent systems. Background Technology

[0002] Multi-agent systems have significant application potential in scenarios such as logistics transportation, disaster response, and autonomous inspection. Among them, heterogeneous multi-agent systems, by integrating the complementary advantages of robots of different forms, can accomplish complex navigation tasks that a single robot cannot achieve. Path planning is the core link in heterogeneous multi-agent collaboration. Existing multi-agent collaborative path planning methods mainly suffer from three limitations: First, pre-training-based methods rely on large-scale training or fine-tuning, resulting in high training and deployment costs, and the scarcity of real-world data limits the model's generalization ability; second, methods relying on prior maps or scene modeling perform well in known environments but struggle to adapt to real-world dynamic unknown scenarios, exhibiting insufficient generalization ability; third, simulation-based methods are highly dependent on the quality of the simulation environment, and if there are differences between the simulation and the real environment, the actual navigation performance will significantly decrease. Meanwhile, traditional homogeneous multi-agent collaborative architectures lack effective task division mechanisms, easily leading to planning chaos and resource waste; while robot systems integrating visual language models improve environmental adaptability, they have not yet achieved the three zero constraints of no training, no prior knowledge, and no simulation, making efficient deployment in complex real-world unmanned environments difficult. Therefore, there is an urgent need for a heterogeneous multi-agent cooperative navigation method that can break free from dependence on training, prior knowledge, and simulation, and possesses efficient task division capabilities to solve the challenges of autonomous exploration and path planning in unknown and complex environments. Thus, to address these issues, we need to design a heterogeneous multi-agent cooperative navigation method. Summary of the Invention

[0003] The purpose of this invention is to provide a heterogeneous multi-agent cooperative navigation method based on the three-zero path planning (TZPP) paradigm, which satisfies the constraints of zero training, zero prior knowledge, and zero simulation. It adopts a coordinator-explorer heterogeneous agent architecture to achieve efficient division of labor and cooperation between humanoid robots and quadruped robots, solve navigation problems in unknown and complex environments, and improve the robustness and environmental adaptability of the system.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a heterogeneous multi-agent cooperative navigation method, based on the three-zero path planning (TZPP) paradigm, satisfying the constraints of zero training, zero prior knowledge, and zero simulation, adopting a coordinator-explorer heterogeneous agent architecture, with a humanoid robot as the coordinator and a quadruped robot as the explorer, to achieve autonomous exploration and path planning in an unknown and complex environment, specifically including the following steps:

[0005] s1. Path evaluation phase: Determine the location of the mission objective. The humanoid robot acts as a coordinator, collecting environmental perception information through its forward-facing vision. For global objectives With its own environmental perception information Conduct a route feasibility analysis to determine whether the route conditions are favorable.

[0006] s2. Task execution or pilot exploration initiation: If the path conditions are favorable, the humanoid robot directly executes the navigation task, moving autonomously based on current environmental perception information, gradually approaching the target location. Until arrival; if path conditions are unfavorable, the humanoid robot initiates a preliminary exploration phase, assigning waypoint locations described in natural language. To quadruped robots, drive quadruped robots to conduct environmental exploration and identify feasible paths;

[0007] s3. Quadruped robot environment exploration: Location of quadruped robot receiving path points Then, it collects environmental perception information through its own forward-looking vision. First, environmental detection is performed to determine whether the environment is fully accessible. If it is not fully accessible, an exploration strategy with few reference points (mode X) or indirect accessibility (mode Y) is matched based on the environmental characteristics, and the corresponding exploration process is executed. Mode X is suitable for open environments with sparse landmarks, while mode Y is suitable for indirect accessibility environments with dense obstacles.

[0008] s4. Exploration Information Feedback and Waypoint Assessment: The quadruped robot will collect environmental perception information during the exploration process. The data is transmitted back to the humanoid robot in real time. The humanoid robot continuously collects feedback information and evaluates the potential navigation benefits of the waypoints that the quadruped robot has reached. Only waypoints with navigation benefits are included in the path reference, while worthless waypoints are excluded, thereby improving the efficiency of path planning.

[0009] s5. Path Update and Final Navigation: The humanoid robot combines the effective exploration information from the quadruped robot to dynamically update the navigation path and enters the task execution phase, moving along the updated path. During the movement, path evaluation is continuously performed. If path conditions change, the preliminary exploration and path update steps are repeated until the humanoid robot successfully reaches the target location. Complete collaborative navigation.

[0010] Preferably, the quadruped robot performs an exploration process in a few-reference-object mode X, specifically including:

[0011] Sx1. The quadruped robot performs a 360° rotation scan to detect the location of its path. Visibility;

[0012] Sx2. If If invisible, immediately terminate the exploration and report the waypoint as unreachable to the humanoid robot; if It can be seen that the quadruped robot moved to and the environmental awareness information of the current location The data was transmitted back to the humanoid robot.

[0013] Sx3. Quadruped robot in Perform another 360° rotation scan to detect the target location. Visibility, and assess from to The feasibility of the path;

[0014] Sx4. If As it is visible and the path is feasible, the quadruped robot sends a successful exploration signal and current environmental information to the humanoid robot; if If the path is not visible or not feasible, terminate the exploration and report the current environmental status to the humanoid robot.

[0015] Preferably, the quadruped robot performs an exploration process for the indirectly reachable mode Y, specifically including:

[0016] Sy1. The quadruped robot performs a 360° rotation scan to detect the location of the points it passes through. Visibility; if Invisible, feedback indicates unreachable information; if As can be seen, moving to And transmit current environmental perception information back ;

[0017] Sy2. Quadruped robots in 360° rotating scan to detect target position Visibility and path feasibility assessment, if If the path is visible and feasible, send a success signal and environmental information.

[0018] Sy3. If If the path is invisible or infeasible, the quadruped robot uses a preset scanning half-angle. For the range, in − To + Rotating scan within an angle to detect potential passageways in the environment;

[0019] Sy4. If a feasible path is detected, the quadruped robot sends a successful exploration signal, along with the path location and environmental information, to the humanoid robot; if no feasible path is detected, the exploration is terminated and the current environmental status is reported.

[0020] Preferably, the humanoid robot's environmental perception information Environmental perception information of quadruped robots All data is collected using a forward-facing vision sensor. During the exploration process, the quadruped robot calculates the distance factor using Euclidean distance. Real-time acquisition of one's own and target's locations The distance provides data support for path assessment.

[0021] Preferably, the entire process of the method of the present invention is executed in a real physical environment. There is no need to perform any model training or fine-tuning on the humanoid robot or quadruped robot in advance, no need to import prior knowledge such as environmental prior maps and scene modeling data, and no need to train, verify or rehearse through a simulation environment. This completely eliminates the dependence on training, prior knowledge and simulation, and significantly reduces deployment costs.

[0022] Preferably, the navigation task execution process of the humanoid robot is an iterative loop, continuously receiving feedback information from the quadruped robot during movement and dynamically updating the path evaluation results; if changes in path conditions are found during movement, such as the appearance of new obstacles or missing landmarks, the pilot exploration, waypoint evaluation and path update steps are immediately repeated to ensure the robustness of navigation.

[0023] Preferably, the method of the present invention is applicable to a variety of complex indoor and outdoor scenarios, including but not limited to open areas without landmarks, L-shaped turning areas, obstacle areas with single / double-sided passage, Z-shaped turning areas, and areas with a combination of ramps and steps. In particular, it supports intelligent agents to complete autonomous navigation in unknown scenarios where the target point cannot be directly observed from the starting position.

[0024] Compared with the prior art, the beneficial effects of the present invention are:

[0025] This invention optimizes an existing heterogeneous multi-agent cooperative navigation method. Based on a three-zero path planning paradigm, it achieves zero-training, zero-prior-knowledge, and zero-simulation heterogeneous multi-agent cooperative navigation, completely eliminating the dependence of existing methods on training, prior maps, and simulation environments. This significantly reduces the deployment and application costs of robot systems and improves the robustness of the system in real dynamic environments. It employs a coordinator-explorer heterogeneous agent architecture to achieve specialized task division between humanoid robots and quadruped robots. The humanoid robot is responsible for high-level task coordination and navigation decisions, while the quadruped robot utilizes its flexible mobility to undertake environmental exploration and feasible path identification. This fully leverages the complementary advantages of different robot forms, solving the problems of chaotic planning and resource waste in homogeneous multi-agent cooperation. Furthermore, it designs a few-reference-object mode X and an indirectly accessible... The proposed method employs a differentiated exploration strategy for mode Y, adapting to both sparse, open environments and obstacle-dense, indirectly accessible environments. This dynamic matching of exploration strategies significantly enhances the system's environmental adaptability in various complex scenarios. The humanoid robot evaluates the navigation benefits of the quadruped robot's waypoints, including only valid waypoints in the path reference. An iterative path planning and update mechanism effectively avoids ineffective exploration and path detours, improving the efficiency of collaborative navigation and enabling the system to achieve near-human-level navigation performance. The method has been validated in real-world physical environments and is applicable to various unknown and complex indoor and outdoor scenarios, including obstacle-dense, sparse, and indirectly accessible environments. It provides a feasible path for the deployment of heterogeneous multi-agent systems in practical scenarios such as logistics, disaster response, and autonomous inspection, demonstrating broad application prospects. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall workflow in the heterogeneous multi-agent cooperative navigation method of the present invention;

[0027] Figure 2 This is a schematic diagram of the exploration process of the few-reference-object mode X in the cooperative navigation method of heterogeneous multi-agents of the present invention;

[0028] Figure 3 This is a schematic diagram of the exploration process of the indirectly reachable mode Y in the cooperative navigation method of heterogeneous multi-agents of the present invention. Detailed Implementation

[0029] 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. This embodiment uses the Unitree G1 humanoid robot and the Unitree Go2 quadruped robot as hardware platforms, and combines a multimodal large language model to build a perception and decision-making system. The effectiveness of the method of the present invention is verified in five typical unknown and complex scenarios. The five scenarios include: locating a sofa after an L-shaped turn, searching behind a narrow column with one-sided passage, finding a bucket behind a column with passage on both sides, locating a fire extinguisher after a Z-shaped turn, and going around the steps behind a load-bearing column by navigating a ramp.

[0030] In this embodiment, the preset parameter is: the maximum number of displacements of the humanoid robot. =2, maximum rotation angle Rmax=π / 2rad, target arrival distance threshold =0.5m, quadruped robot scanning half angle =π / 2rad; Environmental perception information of a quadruped robot Environmental perception information of humanoid robots All data is collected through forward-facing vision sensors, and a multimodal large language model completes the parsing of natural language instructions and semantic reasoning of environmental information.

[0031] Example 1: L-shaped turn positioning sofa

[0032] This scenario is an open area without landmarks. The target sofa is located behind an L-shaped turn. The humanoid robot cannot directly observe the target from its starting position. Matching the few-reference-object mode X, the specific steps of the method of this invention are executed as follows:

[0033] Determine the location of the mission target The humanoid robot collects forward visual information for the sofa behind the L-shaped turn. The path evaluation result indicates unfavorable conditions; therefore, a pilot exploration is initiated, and waypoint locations are assigned. The L-shaped turn leads to the Go2 quadruped robot;

[0034] Go2 quadruped robot receives Then, matching mode X, a 360° rotational scan was performed and detected. As can be seen, moving to the corner and sensing environmental information The data was transmitted back to the G1 humanoid robot.

[0035] Go2 performed another 360° rotation scan at the corner and detected the target sofa. If the path is visible and feasible, send a successful exploration signal and environmental information to G1;

[0036] The G1 humanoid robot assesses that the waypoint at the corner has navigational benefits, includes it in the path reference, updates the navigation path, moves towards the corner, and continues to move along the path provided by Go2 after reaching the corner, successfully arriving at the target sofa and completing the collaborative navigation.

[0037] In this embodiment, the task completion time for collaborative navigation is 64.00s, the humanoid robot travels a distance of 4.00m, the path score is 68.18, the quadruped robot correctly explores 2 key points, and the effective exploration rate is 100%, achieving navigation performance close to that of a human operator.

[0038] Example 2: Circling around the steps behind the load-bearing column via a ramp

[0039] This scenario is a densely obstacle-filled, indirectly accessible area. The target step is located behind a load-bearing column and must be traversed via a ramp. The humanoid robot cannot directly observe the target from its initial position. Therefore, the indirectly accessible mode Y is matched, and the specific steps of the method of this invention are executed:

[0040] Determine the location of the mission target For the steps behind the load-bearing column, the humanoid robot collects forward visual information. The path evaluation result indicates unfavorable conditions; therefore, a pilot exploration is initiated, and waypoint locations are assigned. The ramp entrance next to the load-bearing column leads to the Go2 quadruped robot;

[0041] Go2 quadruped robot receives Subsequently, matching mode Y was detected by 360° rotational scanning. As can be seen, the device moves to the ramp entrance and transmits environmental perception information back. ;

[0042] Go2 performed a 360° rotating scan at the ramp entrance but did not detect the target step. Then with =π / 2rad is the range of rotational scanning, and the slope was detected as a feasible passage;

[0043] Go2 sends a successful exploration signal and ramp passage information to G1. G1 assesses that the ramp entrance waypoint has navigation benefits, updates the navigation path, and moves to the ramp entrance. After circling around the ramp, it successfully reaches the steps behind the load-bearing column, completing the cooperative navigation.

[0044] In this embodiment, the task completion time for collaborative navigation was 154.89s, the humanoid robot traveled a distance of 14.60m, the path score was 88.18, the quadruped robot correctly explored 6 key points, the effective exploration rate was 100%, and the obstacle avoidance coefficient reached the same level as that of a human operator.

[0045] Example 3: Search behind a narrow pillar with one-sided passage

[0046] This scenario is a one-way obstacle area with the target located behind a narrow pillar, allowing only one-way passage. The humanoid robot cannot directly observe the target from its starting position. Matching the indirect reachable mode Y, after executing the method of this invention, the task completion time is 18.22s, the humanoid robot travels a distance of 2.60m, the path score is 98.08, the quadruped robot correctly explores 2 key points, the effective exploration rate is 100%, and the navigation efficiency is basically the same as that of a human operator.

[0047] Although the present invention has been described above with reference to embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of the invention. In particular, the features in the embodiments disclosed herein can be combined with each other in any manner, and the lack of an exhaustive description of these combinations in this specification is merely for the purpose of saving space and resources. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A cooperative navigation method for heterogeneous multi-agent systems, characterized in that, The method is based on the Three-Zero Path Planning (TZPP) paradigm, satisfying the constraints of zero training, zero prior knowledge, and zero simulation. It employs a coordinator-explorer heterogeneous intelligent agent architecture, with a humanoid robot as the coordinator and a quadruped robot as the explorer, to achieve autonomous exploration and path planning in unknown and complex environments. The method includes the following steps: s1. Determine the location of the mission target. Humanoid robots act as coordinators for overall goals With its own environmental perception information Perform a path evaluation to determine if the path conditions are favorable. s2. If the path conditions are good, the humanoid robot directly executes the navigation task, gradually approaching the target location until it arrives; if the path conditions are poor, the humanoid robot initiates a preliminary exploration phase, assigning waypoint locations described in natural language. To the quadruped robot, drive the quadruped robot to explore the environment; s3. Location of the quadruped robot's receiving path point Then, based on its own environmental perception information Perform environmental detection, match the exploration strategy for the mode with few references X or the mode that is not directly accessible Y, and execute the corresponding exploration process; s4. The quadruped robot transmits environmental perception information during the exploration process back to the humanoid robot in real time. The humanoid robot continuously collects feedback and evaluates the navigation benefits of the waypoints that have been reached, and only includes waypoints with navigation benefits in the path reference. s5. The humanoid robot combines the effective exploration information from the quadruped robot, updates its navigation path, and executes tasks until it reaches the target location. Complete collaborative navigation.

2. The heterogeneous multi-agent cooperative navigation method according to claim 1, characterized in that, The environment detection and pattern matching of the quadruped robot in step s3 specifically includes: the quadruped robot first determines whether the environment is fully accessible. If it is not fully accessible, it matches the less reference object mode X or the indirectly accessible mode Y according to the environmental characteristics. Mode X is suitable for open environments with sparse landmarks, and mode Y is suitable for indirectly accessible environments with dense obstacles.

3. The heterogeneous multi-agent cooperative navigation method according to claim 2, characterized in that, The exploration process of quadruped robot execution mode X in step s3 specifically includes: Sx1. The quadruped robot performs a 360° rotation scan to detect the location of its path. Visibility; Sx2. If If invisible, terminate the exploration and report the waypoint as unreachable to the humanoid robot; if As can be seen, moving to It also transmits environmental perception information back; Sx3. Quadruped robot in Perform another 360° rotation scan to detect the target location. Visibility and assess path feasibility; Sx4. If If the path is visible and feasible, send a successful exploration signal and environmental information to the humanoid robot; if If the path is not visible or not feasible, terminate the exploration and report the current environmental status.

4. The heterogeneous multi-agent cooperative navigation method according to claim 2, characterized in that, The exploration process of the quadruped robot execution mode Y in step s3 specifically includes: Sy1. The quadruped robot performs a 360° rotation scan to detect the location of the points it passes through. The visibility of the information is displayed; if it is not visible, an "unreachable" message is displayed; if it is visible, the user is moved to the desired location. It also transmits environmental perception information back; Sy2. Quadruped robots in 360° rotating scan to detect target position Visibility and path feasibility, if If the path is visible and feasible, send a success signal and environmental information. Sy3. If If the path is invisible or infeasible, the quadruped robot scans half-angles. For the range, in − To + Rotating scan within an angle to detect potential passageways; Sy4. If a feasible path is detected, send a successful exploration signal and path information to the humanoid robot; if no feasible path is detected, terminate the exploration and report the current environmental status.

5. The heterogeneous multi-agent cooperative navigation method according to claim 1, characterized in that, The humanoid robot's environmental perception information Environmental perception information of quadruped robots All data is acquired based on forward vision. During the exploration process of the quadruped robot, the distance factor is calculated using Euclidean distance. Real-time acquisition of target location The distance.

6. The heterogeneous multi-agent cooperative navigation method according to claim 1, characterized in that, The entire process from step s1 to step s5 is executed in a real physical environment, without the need for pre-training of the agent model, importing prior environmental maps or scene modeling data, or training and verification through a simulation environment.

7. The heterogeneous multi-agent cooperative navigation method according to claim 1, characterized in that, The humanoid robot's navigation task execution process is iterative. During the movement, it continuously receives feedback information from the quadruped robot and dynamically updates the path evaluation results. If the path conditions change, the pilot exploration and path update steps are repeated.

8. A cooperative navigation method for heterogeneous multi-agent systems according to any one of claims 1-7, characterized in that, The method is applicable to various complex indoor and outdoor scenarios, including open areas without landmarks, L-shaped turning areas, obstacle areas with single / double-sided passage, Z-shaped turning areas, and areas with a combination of ramps and steps. It also supports navigation in scenarios where the agent cannot directly observe the target point from the starting position.

Images