A multi-robot collaborative system and method based on distributed embodied intelligence

Abstract

This invention discloses a multi-robot collaborative system and method based on distributed embodied intelligence, belonging to the field of distributed artificial intelligence technology. The system consists of multiple peer autonomous robot nodes, each node including: a multimodal embodied perception module, an environmental semantic understanding module, a sparse neighborhood communication module, a collaborative semantic alignment module, a distributed task market module, and a behavior generation module. The collaborative semantic alignment module integrates local semantic information through a distributed consensus algorithm to form a global consensus semantic map. The distributed task market module achieves self-organized decomposition and allocation of tasks through a resource-constrained bidirectional auction mechanism. The system has no logical or physical central control node. This invention, employing the above-mentioned multi-robot collaborative system and method based on distributed embodied intelligence, can improve the robustness, adaptive collaborative capability, and overall task execution efficiency of multi-robot systems in dynamic open environments.

Description

Technical Field

[0001] This invention relates to the field of distributed artificial intelligence technology, and in particular to a multi-robot collaborative system and method based on distributed embodied intelligence. Background Technology

[0002] With the rapid development of industrial automation, smart logistics, disaster relief, and space exploration, multi-robot collaborative systems have become a frontier in research and application due to their potential in efficiency, robustness, and task complexity. In existing technologies, such systems mostly employ centralized or hierarchical control architectures. In a centralized architecture, a central server is responsible for collecting perception data from all robots, performing global environment modeling, task planning and decomposition, and distributing specific action instructions to each robot. While this architecture facilitates global optimization, it suffers from a fatal single point of failure risk, and the central node becomes a communication and computing bottleneck, limiting the system's scalability. Hierarchical architectures attempt to introduce local controllers to distribute the load, but their top-level planner remains centralized, failing to eradicate the fundamental problem.

[0003] Furthermore, the aforementioned architecture relies on precise prior modeling and pre-defined processes for the environment and tasks. When faced with dynamic, open, and uncertain real-world environments—such as the sudden appearance of obstacles, temporary changes to task objectives, or unexpected robot malfunctions—the system lacks the ability to reconstruct collaborative strategies online in real time. Robots are typically viewed as passive execution terminals, their rich embodied perception data compressed into simple state information for central decision-making, failing to fully utilize the local cognitive and decision-making potential of each robot as an intelligent agent. This results in insufficient overall system flexibility, rigid collaborative behavior, and difficulty in effectively resolving "ambiguities in environmental understanding" arising from differences in perspective among robots, creating a "semantic gap" in collaborative efficiency. Therefore, a new paradigm is urgently needed to achieve truly distributed, adaptive, and highly robust collaboration, enabling robot swarms to cooperate intelligently like an organic whole without centralized command. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-robot collaborative system and method based on distributed embodied intelligence to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a multi-robot collaborative system based on distributed embodied intelligence, comprising multiple autonomous robot nodes, each robot node including:

[0006] The multimodal embodied perception module is used to acquire local environmental perception data and label the reliability parameters of each data point;

[0007] The environmental semantic understanding module is communicatively coupled with the multimodal embodied perception module and is used to generate a local semantic map containing confidence information based on local environmental perception data and reliability parameters.

[0008] The sparse neighborhood communication module is used to exchange messages with one or more neighboring robot nodes according to dynamic neighborhood filtering rules;

[0009] The collaborative semantic alignment module is coupled with the environmental semantic understanding module and the sparse neighborhood communication module. It is used to integrate the local semantic map and the messages exchanged with neighboring robot nodes through a distributed consensus algorithm, and update the confidence to form a global consensus semantic map.

[0010] The distributed task market module, which is communicatively coupled with the collaborative semantic alignment module and the sparse neighborhood communication module, is used to receive or publish atomic tasks through a resource-constrained bidirectional auction mechanism, and to decompose and allocate tasks.

[0011] The behavior generation module, which is communicatively coupled with the distributed task marketplace module, is used to generate control instructions based on the assigned atomic tasks.

[0012] The system has no logical or physical central control node.

[0013] Preferably, reliability parameters include sensor accuracy parameters. Environmental occlusion parameters and time validity parameters ;

[0014] ;

[0015] in, The time delay between the data acquisition time and the current time. The attenuation coefficient;

[0016] The specific formula for generating confidence scores by the environmental semantic understanding module is as follows:

[0017] ;

[0018] in, , , These are the weighting coefficients, and .

[0019] Preferably, the dynamic neighborhood selection rule of the sparse neighborhood communication module is based on calculating neighborhood relevance. Select the communication target;

[0020] ;

[0021] in, , , These are the weighting coefficients. For normalization function, For robot nodes and robot nodes The distance between, For robot nodes and robot nodes Inter-communication link quality, This is a task-related factor.

[0022] Preferably, the collaborative semantic alignment module employs a weighted consensus algorithm for any semantic node. Update robot nodes The confidence level is calculated using the following formula:

[0023] ;

[0024] in, For the number of iterations, In the first After rounds of iteration, robot nodes For a specific semantic node in the environment The confidence level value, For the first After rounds of iteration, robot nodes For the same semantic node The updated confidence value, For robot nodes The set of neighboring nodes, For nodes The weight of self-confidence. For nodes One of the neighboring nodes, For nodes with neighboring nodes The weight of confidence level;

[0025] The condition for reaching consensus is that the confidence variance of all nodes is less than a set threshold.

[0026] Preferably, the specific process of the two-way auction mechanism with resource constraints in the distributed task market module is as follows:

[0027] a. Any robot node acting as the bidding party will perform the atomic task. and resource demand vector Broadcast to its neighboring nodes;

[0028] b. The neighboring robot node, acting as the bidder, calculates the expected benefits of performing the task. and resource consumption costs Only when its own resource vector Generate bid value And feedback is sent to the bidding party's node;

[0029] c. The bidding node selects the node with the highest bid value to win the bid and broadcasts the allocation result, thus completing the distributed allocation of the atomic task.

[0030] The preferred formula for calculating resource consumption costs is as follows:

[0031] ;

[0032] in, For resource type indexing, This refers to a collection of resources, including but not limited to energy, time, and computing resources. To estimate resource consumption, This represents the current remaining resource amount. This refers to the resource weighting coefficient.

[0033] Formula for calculating bid value: .

[0034] Preferably, the behavior generation module employs a model prediction control algorithm, with the specific formula as follows:

[0035] ;

[0036] in, To predict the length of the time domain, For robots at all times The predicted state, This is the expected state from the atomic task. The weighted quadratic deviation between the predicted state and the expected state. To control the input vector, To control the weighted quadratic cost of the input, and This is the weight matrix. For the terminal cost function, To optimize the system state at the end of the predicted time domain.

[0037] Preferably, when any robot node detects a task execution failure or a significant change in the environment, it publishes a task anomaly event through the sparse neighborhood communication module, triggering the re-running of the collaborative semantic alignment module and the distributed task market module of robot nodes in the relevant local area, so as to achieve dynamic reallocation of tasks and resources.

[0038] A multi-robot collaborative method based on distributed embodied intelligence, executed by any robot node in a multi-robot collaborative system based on distributed embodied intelligence, includes the following steps:

[0039] S1. Acquire local environmental perception data and label reliability parameters based on sensor accuracy, environmental occlusion, and data timeliness;

[0040] S2. Integrate local environmental perception data with reliability parameters to generate a local semantic map with confidence assessment;

[0041] S3. Based on dynamic neighborhood filtering rules, exchange messages with one or more neighboring robot nodes. The messages include local semantic map fragments, task bidding information, and status updates.

[0042] S4. Through a weighted consensus algorithm, the local and received neighbor semantic information are iteratively integrated to make the confidence of each node on key environmental features consistent, thus forming a global consensus semantic map.

[0043] S5. Based on a global consensus semantic map, it participates in the release, bidding and allocation of atomic tasks through a two-way auction mechanism with resource constraints. The bidding decision is based on a comprehensive calculation of the task benefits and its own resource consumption costs.

[0044] S6. Generate and execute model prediction control instructions based on the assigned atomic tasks;

[0045] S7. If a predetermined change in the environment or a task failure is detected during execution, a local renegotiation is triggered, and steps S4 to S6 are re-executed.

[0046] Therefore, the present invention employs the above-mentioned multi-robot collaborative system and method based on distributed embodied intelligence, which has the following beneficial effects:

[0047] (1) The fully distributed architecture significantly reduces the risk of single point of failure. Each robot node has autonomous perception and decision-making capabilities and works collaboratively through local communication. When an individual node fails, its task can be taken over by a neighboring node through a renegotiation mechanism, thereby improving the overall reliability and robustness of the system.

[0048] (2) Through the distributed semantic alignment mechanism based on confidence transmission, the environmental cognition differences caused by the difference in perspective of each robot are effectively reconciled. It can form a globally consistent semantic map without relying on the central node, providing a reliable basis for collaborative decision-making, while reducing the need for redundant original data transmission.

[0049] (3) A two-way auction mechanism with resource constraints is introduced for task allocation, which realizes dynamic matching of tasks with robot resources and capabilities. This mechanism promotes load balancing at the system level while ensuring the sustainability of individual resources through quantitative cost-benefit calculation, thereby improving the overall task execution efficiency and resource utilization.

[0050] (4) By combining task decision-making, model predictive control and online monitoring, a closed loop of perception-decision-execution is formed. The system can respond to dynamic changes and abnormal states in the environment during task execution and trigger local replanning, thereby enhancing the task adaptability and success rate in complex scenarios.

[0051] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0052] Figure 1 This is a diagram illustrating the architecture of a multi-robot collaborative system based on distributed embodied intelligence, as described in this invention.

[0053] Figure 2 This is a flowchart of a multi-robot collaborative method based on distributed embodied intelligence according to the present invention. Detailed Implementation

[0054] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0055] Example

[0056] This embodiment takes an unmanned warehouse collaborative handling scenario as an example.

[0057] like Figure 1 As shown, this invention provides a multi-robot collaborative system based on distributed embodied intelligence, comprising multiple autonomous robot nodes. Each robot node includes a multimodal embodied perception module 1, an environmental semantic understanding module 2, a sparse neighborhood communication module 3, a collaborative semantic alignment module 4, a distributed task marketplace module 5, and a behavior generation module 6.

[0058] The multimodal embodied perception module 1 is used to acquire local environmental perception data and label the reliability parameters of each data. In this embodiment, the module acquires local environmental perception data through the RGB-D camera, 2D LiDAR and inertial measurement unit (IMU) mounted on the robot, and labels the reliability parameters of each data frame in real time.

[0059] Reliability parameters include sensor accuracy parameters Environmental occlusion parameters and time validity parameters ;

[0060] ;

[0061] in, The time delay between the data acquisition time and the current time. This is the attenuation coefficient.

[0062] The environmental semantic understanding module 2 is communicatively coupled with the multimodal embodied perception module 1. It is used to generate a local semantic map containing confidence information based on local environmental perception data and reliability parameters. In this embodiment, the local semantic map output by this module contains object-level information such as "the cargo box (confidence 0.85) is located at coordinates (X,Y)".

[0063] The specific formula for generating confidence scores is as follows:

[0064] ;

[0065] in, , , These are the weighting coefficients, and For example, the weighting coefficients can be calibrated experimentally. , , .

[0066] The sparse neighborhood communication module 3, coupled with the collaborative semantic alignment module 4 and the distributed task market module 5, is used to exchange messages with one or more neighboring robot nodes according to dynamic neighborhood filtering rules; message types include local semantic map fragments, task bidding information and node status updates.

[0067] Dynamic neighborhood selection rules calculate neighborhood relevance. Select the neighboring node to communicate with first;

[0068] ;

[0069] in, , , These are the weighting coefficients. For normalization function, For robot nodes and robot nodes The distance between, For robot nodes and robot nodes Inter-communication link quality, This is a task-related factor.

[0070] The collaborative semantic alignment module 4 is communicatively coupled with the environmental semantic understanding module 2 and the sparse neighborhood communication module 3. It is used to integrate the local semantic map and the messages exchanged with neighboring robot nodes through a distributed consensus algorithm, and update the confidence to form a global consensus semantic map.

[0071] The distributed consensus algorithm uses a weighted consensus algorithm, which is applied to any semantic node. Update robot nodes The confidence level is calculated using the following formula:

[0072] ;

[0073] in, For the number of iterations, In the first After rounds of iteration, robot nodes For a specific semantic node in the environment The confidence level value, For the first After rounds of iteration, robot nodes For the same semantic node The updated confidence value, For robot nodes The set of neighboring nodes, For nodes The weight of self-confidence. For nodes One of the neighboring nodes, For nodes with neighboring nodes The weights of the confidence level, all weight coefficients satisfy the normalization condition and are positively correlated with the original confidence level and communication reliability;

[0074] The condition for reaching a consensus is that the confidence variance of all nodes is less than a set threshold; through multiple rounds of iteration, all robots eventually reach a consensus on key information such as "where is the cargo box" and "whether the passage is clear".

[0075] The distributed task market module 5 is communicatively coupled with the collaborative semantic alignment module 4 and the sparse neighborhood communication module 3. It is used to receive or publish atomic tasks through a bidirectional auction mechanism with resource constraints, and to decompose and allocate tasks.

[0076] The specific process of a two-way auction mechanism with resource constraints is as follows:

[0077] a. Any robot node acting as the bidding party will perform the atomic task. (e.g., moving a cargo box to point B) and resource demand vector (The required energy, time, etc.) are broadcast to its neighboring nodes;

[0078] b. The neighboring robot node, acting as the bidder, calculates the expected benefits of performing the task. and resource consumption costs Only when its own resource vector Generate bid value And feedback is sent to the bidding party's node;

[0079] The formula for calculating resource consumption costs is as follows:

[0080] ;

[0081] in, For resource type indexing, This refers to a collection of resources, including but not limited to energy, time, and computing resources. For robot nodes The estimated resource consumption for executing this task can be estimated based on models such as task path length and computational complexity. This represents the current remaining resource amount. This refers to the resource weighting coefficient.

[0082] Formula for calculating bid value: Expected benefits It can be calculated by taking into account factors such as the suitability of the task to its own position and the overall importance of the task.

[0083] c. The bidding node selects the node with the highest bid value to win the bid and broadcasts the allocation result, thus completing the distributed allocation of the atomic task.

[0084] Behavior generation module 6, which is communicatively coupled to distributed task market module 5, is used to generate control instructions based on the assigned atomic tasks;

[0085] Behavior generation module 6 employs a model predictive control algorithm, the specific formula of which is as follows:

[0086] ;

[0087] in, To predict the length of the time domain, For robots at all times The predicted state, This is the expected state from the atomic task. The weighted quadratic deviation between the predicted state and the expected state. To control the input vector, To control the weighted quadratic cost of the input, and This is a weighting matrix used to balance tracking accuracy and control magnitude. This is the terminal cost function, used to ensure the stability and consistency of the prediction at the end of the time domain. To optimize the system state at the end of the predicted time domain.

[0088] The system has no logical or physical central control node, and all decisions are generated in a distributed manner among peer nodes. During operation, when any robot node detects a task execution failure or a significant change in the environment, it publishes a task anomaly event through the sparse neighborhood communication module 3, triggering the collaborative semantic alignment module 4 and the distributed task market module 5 of the robot nodes in the relevant local area to restart, and quickly and adaptively decompose and redistribute the unfinished or affected tasks to achieve dynamic re-planning.

[0089] like Figure 2 As shown, a multi-robot collaborative method based on distributed embodied intelligence, executed by any robot node in a multi-robot collaborative system based on distributed embodied intelligence, includes the following steps:

[0090] S1. Acquire local environmental perception data and label reliability parameters based on sensor accuracy, environmental occlusion, and data timeliness;

[0091] S2. Integrate local environmental perception data with reliability parameters to generate a local semantic map with confidence assessment;

[0092] S3. Based on dynamic neighborhood filtering rules, exchange messages with one or more neighboring robot nodes. The messages include local semantic map fragments, task bidding information, and status updates.

[0093] S4. Through a weighted consensus algorithm, the local and received neighbor semantic information are iteratively integrated to make the confidence of each node on key environmental features consistent, thus forming a global consensus semantic map.

[0094] S5. Based on a global consensus semantic map, it participates in the release, bidding and allocation of atomic tasks through a two-way auction mechanism with resource constraints. The bidding decision is based on a comprehensive calculation of the task benefits and its own resource consumption costs.

[0095] S6. Generate and execute model prediction control instructions based on the assigned atomic tasks;

[0096] S7. If a predetermined change in the environment or a task failure is detected during execution, a local renegotiation is triggered, and steps S4 to S6 are re-executed.

[0097] Therefore, the present invention adopts the above-mentioned multi-robot collaborative system and method based on distributed embodied intelligence, which solves the problems of single point of failure risk, poor environmental adaptability and rigid collaboration in existing centralized systems, and can improve the robustness, adaptive collaboration capability and overall task execution efficiency of multi-robot systems in dynamic open environments.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A multi-robot collaborative system based on distributed embodied intelligence, characterized in that, It includes multiple autonomous robot nodes, each robot node including: The multimodal embodied perception module is used to acquire local environmental perception data and label the reliability parameters of each data point; The environmental semantic understanding module is communicatively coupled with the multimodal embodied perception module and is used to generate a local semantic map containing confidence information based on local environmental perception data and reliability parameters. The sparse neighborhood communication module is used to exchange messages with one or more neighboring robot nodes according to dynamic neighborhood filtering rules; The collaborative semantic alignment module is coupled with the environmental semantic understanding module and the sparse neighborhood communication module. It is used to integrate the local semantic map and the messages exchanged with neighboring robot nodes through a distributed consensus algorithm, and update the confidence to form a global consensus semantic map. The distributed task market module, which is communicatively coupled with the collaborative semantic alignment module and the sparse neighborhood communication module, is used to receive or publish atomic tasks through a resource-constrained bidirectional auction mechanism, and to decompose and allocate tasks. The behavior generation module, which is communicatively coupled with the distributed task marketplace module, is used to generate control instructions based on the assigned atomic tasks. The system has no logical or physical central control node.

2. The multi-robot collaborative system based on distributed embodied intelligence according to claim 1, characterized in that: Reliability parameters include sensor accuracy parameters Environmental occlusion parameters and time validity parameters ; ； in, The time delay between the data acquisition time and the current time. The attenuation coefficient; The specific formula for generating confidence scores by the environmental semantic understanding module is as follows: ； in, , , These are the weighting coefficients, and .

3. A multi-robot collaborative system based on distributed embodied intelligence according to claim 1, characterized in that: The dynamic neighborhood selection rule of the sparse neighborhood communication module calculates the neighborhood relevance. Select the communication target; ； in, , , These are the weighting coefficients. For normalization function, For robot nodes and robot nodes The distance between, For robot nodes and robot nodes Inter-communication link quality, This is a task-related factor.

4. A multi-robot collaborative system based on distributed embodied intelligence according to claim 1, characterized in that, The collaborative semantic alignment module employs a weighted consensus algorithm for any semantic node. Update robot nodes The confidence level is calculated using the following formula: ； in, For the number of iterations, In the first After rounds of iteration, robot nodes For a specific semantic node in the environment The confidence level value, For the first After rounds of iteration, robot nodes For the same semantic node The updated confidence value, For robot nodes The set of neighboring nodes, For nodes The weight of self-confidence. For nodes One of the neighboring nodes, For nodes with neighboring nodes The weight of confidence level; The condition for reaching consensus is that the confidence variance of all nodes is less than a set threshold.

5. A multi-robot collaborative system based on distributed embodied intelligence according to claim 1, characterized in that, The specific process of the two-way auction mechanism with resource constraints in the distributed task market module is as follows: a. Any robot node acting as the bidding party will perform the atomic task. and resource demand vector Broadcast to its neighboring nodes; b. The neighboring robot node, acting as the bidder, calculates the expected benefits of performing the task. and resource consumption costs Only when its own resource vector Generate bid value And feedback is sent to the bidding party's node; c. The bidding node selects the node with the highest bid value to win the bid and broadcasts the allocation result, thus completing the distributed allocation of the atomic task.

6. A multi-robot collaborative system based on distributed embodied intelligence according to claim 5, characterized in that, The formula for calculating resource consumption costs is as follows: ； in, For resource type indexing, A collection of resources, including energy, time, and computing resources. To estimate resource consumption, This represents the current remaining resource amount. This refers to the resource weighting coefficient. Formula for calculating bid value: .

7. A multi-robot collaborative system based on distributed embodied intelligence according to claim 1, characterized in that, The behavior generation module uses a model prediction control algorithm, the specific formula of which is as follows: ； in, To predict the length of the time domain, For robots at all times The predicted state, This is the expected state from the atomic task. The weighted quadratic deviation between the predicted state and the expected state. To control the input vector, To control the weighted quadratic cost of the input, and This is the weight matrix. For the terminal cost function, To optimize the system state at the end of the predicted time domain.

8. A multi-robot collaborative system based on distributed embodied intelligence according to claim 1, characterized in that: When any robot node detects a task execution failure or a significant change in the environment, it publishes a task anomaly event through the sparse neighborhood communication module, triggering the re-running of the collaborative semantic alignment module and the distributed task market module of robot nodes in the relevant local area, so as to achieve dynamic reallocation of tasks and resources.

9. A multi-robot collaborative method based on distributed embodied intelligence, characterized in that, Applied to a multi-robot collaborative system based on distributed embodied intelligence as described in any one of claims 1-8, and executed by any one of the robot nodes, the system includes the following steps: S1. Acquire local environmental perception data and label reliability parameters based on sensor accuracy, environmental occlusion, and data timeliness; S2. Integrate local environmental perception data with reliability parameters to generate a local semantic map with confidence assessment; S3. Based on dynamic neighborhood filtering rules, exchange messages with one or more neighboring robot nodes. The messages include local semantic map fragments, task bidding information, and status updates. S4. Through a weighted consensus algorithm, the local and received neighbor semantic information are iteratively integrated to make the confidence of each node on key environmental features consistent, thus forming a global consensus semantic map. S5. Based on a global consensus semantic map, it participates in the release, bidding and allocation of atomic tasks through a two-way auction mechanism with resource constraints. The bidding decision is based on a comprehensive calculation of the task benefits and its own resource consumption costs. S6. Generate and execute model prediction control instructions based on the assigned atomic tasks; S7. If a predetermined change in the environment or a task failure is detected during execution, a local renegotiation is triggered, and steps S4 to S6 are re-executed.

Images