Multi-robot Byzantine fault-tolerant consensus method based on semantic verification and dynamic weighting

Abstract

The invention relates to the technical field of collaborative decision and Byzantine fault tolerance of a multi-robot system, discloses a semantic verification and dynamic weighting-based multi-robot Byzantine fault tolerance consensus method, and solves some defects of a traditional PBFT algorithm. The core of the method is to construct a dynamic physical semantic rule base containing physical rules, task scenes and group collaborative constraints, perform double check of rule matching and multi-node cross validation on robot structured messages to eliminate false messages, integrate check results and dynamic weights into a consensus process, and match real-time updating and periodic calibration to optimize performance. The method breaks through the limitation that a traditional consensus only verifies a message form, accurately recognizes illegal false messages, reinforces the effect of high-credibility nodes through dynamic weighting, can still achieve real consensus when a small number of Byzantine nodes exist, adapts to a multi-robot dynamic cooperation scene, does not increase too high resource overhead, improves the system safety and decision reliability, and is suitable for large-scale popularization and application. The method is suitable for key tasks such as regional inspection and fault confirmation.

Description

Technical Field

[0001] This invention belongs to the field of multi-robot system collaborative decision-making and Byzantine fault tolerance technology. Specifically, it relates to a multi-robot Byzantine fault-tolerant consensus mechanism that integrates semantic verification and dynamic credibility weighting, which is applicable to multi-robot collaborative task scenarios such as inspection and fault confirmation. Background Technology

[0002] Multi-robot collaborative systems, through the collaborative perception, decision-making, and execution of distributed nodes, have been widely applied in key areas such as industrial inspection, emergency rescue, and warehousing and logistics. The core requirement of such systems is to ensure that all robot nodes (hereinafter referred to as "nodes") reach a consensus on task status and environmental information in dynamic and complex physical environments. The consensus mechanism, as the core technology to ensure this goal, directly determines the security and reliability of the system.

[0003] Byzantine fault-tolerant consensus mechanisms are key technologies for dealing with interference from malicious nodes (Byzantine nodes). Among them, the Practical Byzantine Fault-Tolerant (PBFT) algorithm, with its strong fault tolerance capability of "tolerating no more than 1 / 3 of malicious nodes" and its high efficiency with polynomial time complexity, has become the mainstream choice for traditional distributed systems. It ensures consensus consistency through a three-stage process of "proposal-pre-voting-final voting," combined with formal methods such as message signature verification and sequence number verification. It has been successfully applied to pure software scenarios such as blockchain and distributed databases. However, it has revealed significant shortcomings in multi-robot collaborative scenarios:

[0004] First, consensus verification lacks physical semantic validity checks. Traditional PBFT only verifies the formal legality of messages (signature, sequence number, identity), without considering the physical semantic validity of the message content. Messages from multiple robots are directly related to the physical world state (position, speed, energy consumption, etc.). Byzantine nodes can forge false messages such as "5% remaining battery, completed 20km inspection" or "moved 1km in 1 second," which are formally valid but violate physical laws. If such messages are included in the consensus, it will lead to decision-making errors, and traditional PBFT cannot identify such problems.

[0005] Secondly, the voting weight design is rigid and fails to consider differences in node credibility. Traditional PBFT uses an equal weighting mechanism, giving all "normal" nodes equal voting influence. However, the reliability of multiple robot nodes varies significantly: some nodes have high sensor accuracy and high task completion rates, exhibiting high credibility; others may exhibit low credibility due to hardware failures or malicious behavior. Equal weighting gives low-credibility nodes the same decision-making power as high-credibility nodes, allowing a few Byzantine nodes to disrupt consensus through spoofing, thus reducing system security.

[0006] Currently, research on improvements to PBFT can be broadly divided into two categories:

[0007] One category is consensus efficiency optimization solutions. For example, the Zyzzyva algorithm shortens consensus latency through "speculation execution," while the HotStuff algorithm simplifies the process using layered voting. Although these solutions improve efficiency, they do not address the core issues of physical semantic verification and dynamic weight allocation, and are still not suitable for multi-robot scenarios.

[0008] Another category is fault tolerance and weight improvement solutions. For example, the RBFT algorithm improves fault tolerance by grouping nodes, but the grouping logic is fixed and cannot adapt to the dynamic movement characteristics of robots. Some solutions allocate weights based on computing power and bandwidth, but these hardware indicators are not directly related to message authenticity and node credibility, and cannot fundamentally suppress Byzantine node interference. In addition, other consensus mechanisms such as PoW and PoS are not suitable for multi-robot scenarios: PoW relies on computing power competition, which contradicts the limited energy and computing power of robots; PoS allocates weights based on stake, without considering physical scenario constraints; DPoS lacks dynamic credibility assessment and still cannot solve the semantic verification problem.

[0009] The unique characteristics of multi-robot collaborative scenarios further amplify the shortcomings of existing technologies: node mobility leads to dynamic changes in communication links and task loads, energy and computing power constraints require lightweight consensus mechanisms, and critical task scenarios place extremely high demands on consensus accuracy and anti-interference capabilities.

[0010] In summary, existing consensus mechanisms generally lack physical semantic verification capabilities and dynamic credibility weighting mechanisms, making them unsuitable for the physical characteristics and security requirements of multi-robot collaboration and hindering their application in critical fields. Therefore, developing an enhanced Byzantine fault-tolerant consensus mechanism that integrates physical semantic rationality verification and dynamic credibility weighting has become a critical issue that urgently needs to be addressed. Summary of the Invention

[0011] To address the aforementioned issues, this invention discloses a multi-robot Byzantine fault-tolerant consensus method based on semantic verification and dynamic weighting.

[0012] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0013] Step 1: Construct a scenario-based physical semantic rule base: Establish a dynamic and scalable rule base that includes physical law constraints, task scenario constraints, and group collaboration constraints, which will serve as the core basis for message semantic verification;

[0014] Step 2: Develop a standardized message format, including key fields such as node ID, timestamp, task type, core data, and supporting data, to support semantic feature extraction;

[0015] Step 3: Through the dual logic of "rule matching verification + multi-node cross-verification", the semantic rationality of the messages uploaded by the robot is determined, and only the messages that pass the verification enter the subsequent consensus process;

[0016] Step 4: Quantify the credibility score based on the node's historical behavior indicators, and dynamically map it to the consensus voting weight to achieve dynamic adjustment of "the better the behavior, the higher the weight";

[0017] Step 5: Deeply integrate the semantic verification results and dynamic voting weights into the PBFT consensus process to optimize the proposal initiation, pre-voting, and final voting stages and achieve reliable consensus;

[0018] Step 6: Optimize rule base parameters and node credibility scores in real time through "event-triggered updates + periodic calibration updates" to adapt to dynamic changes in the system.

[0019] Further improvements include the dynamic credibility score ( The calculation formula is as follows:

[0020] = + + -

[0021] in, , , Weighting coefficients of indicators =1, For task completion rate, To improve message semantic compliance rate, For data consistency rate, This is a penalty for violations (value range: 0-0.5).

[0022] Further improvements include the voting weights. The mapping formula is as follows:

[0023]

[0024] Where N is the total number of nodes participating in the current round of consensus, and the constraint is: =1; if < 0.2, forced setting ( = 0.05 / N), limiting the influence of low-trust nodes.

[0025] Further improvements include the following:

[0026] Rule matching verification: Extract key features of the message and align them one by one with the physical semantic rule base; any violation of a rule is marked as a semantic anomaly.

[0027] Multi-node cross-validation: For fuzzy messages that pass the rule matching, retrieve messages from other nodes in the same task / region during the same period, calculate the data deviation, and determine the semantic abnormality if the deviation exceeds the preset threshold.

[0028] Compared with existing methods, the present invention has the following advantages:

[0029] (1) Breaking through the limitations of traditional consensus which only verifies message form, it accurately identifies false messages that violate physical laws and common sense of scenarios through a physical semantic rule base and a dual verification mechanism, thus blocking the interference path of Byzantine nodes from the source.

[0030] (2) Introduce a dynamic credibility weighting mechanism, quantify the weight based on the actual behavior of nodes, so that high credibility nodes have greater voting influence, automatically suppress interference from low credibility nodes, and improve the accuracy of consensus results.

[0031] (3) The physical semantic rule base supports scenario-based customization and dynamic updates, and the credibility weighting supports event triggering and periodic calibration, adapting to different collaborative tasks and system dynamic changes, and has strong versatility.

[0032] (4) In scenarios with a few Byzantine nodes, consensus that is consistent with the real task state can be reached efficiently, fundamentally improving the inherent security and collaborative decision-making reliability of multi-robot systems. Attached Figure Description

[0033] Figure 1 This is an architecture diagram of a multi-robot Byzantine fault-tolerant consensus method based on semantic verification and dynamic weighting; Detailed Implementation

[0034] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The implementation process of the present invention mainly consists of five core parts: rule base and data preparation, node message processing, physical and semantic dual verification, dynamic credibility weighted calculation, and semantic enhancement consensus execution.

[0035] (1) Rule base and data preparation

[0036] The physical semantic rule base is constructed for multi-robot area inspection tasks, and three core rules are built, supporting dynamic updates on the server side: Physical law constraints: Robot movement speed ≤ 2m / s; Remaining battery power ≥ Energy consumption for executing the current action (power consumption ≥ 10% for 1km inspection); The same node cannot appear in two areas with a distance > 2m / s × time difference at the same time; Task scenario constraints: The normal temperature range of the warehouse inspection area is 5-35℃; Motor current deviation ≥ ±3A when the motor fails; The coverage area of ​​a single inspection task ≤ the node's rated inspection radius; Group collaboration constraints: Data collection deviation of multiple robots in the same area ≤ 5%; Consistency of task progress in adjacent areas ≥ 80%; Fault reporting must be consistent with data from at least one adjacent node.

[0037] Data preparation involves storing historical behavior data (task completion records, message sending records, and raw sensor data) for credibility metric statistics on the robot nodes. The server maintains a public rule base, a node behavior database, and a credibility scoring table to ensure that the data is tamper-proof.

[0038] (2) Node message processing

[0039] Robot nodes generate consensus messages according to a standardized format, which is defined as follows:

[0040] {Node ID: R1, Timestamp: T202406100930, Task Type: Warehouse Area Inspection, Core Data: {Location: Area A-3, Temperature: 26℃, Inspection Progress: 95%}, Supporting Data: {Raw Sensor Data: [25.8℃, 26.2℃], Neighboring Node Communication Records: R2, R3, Remaining Battery: 32%}}. The node uploads the structured message to the server, and the server extracts key features through the semantic parsing module: location, temperature, inspection progress, remaining battery, spatiotemporal information, etc.

[0041] (3) Dual verification of message physical and semantics

[0042] Rule matching verification: Align the extracted message features with the physical semantic rule base one by one. If the message from node R1 displays "temperature 26℃, movement speed 1.2m / s, remaining power 32%, 1km inspection consumed 8% of power", it meets the corresponding constraints and is initially judged to be semantically reasonable; if a message from a node displays "temperature 85℃, remaining power 3%, 1km inspection completed", it violates the constraints of physical laws and task scenario, and is directly marked as semantically abnormal and discarded.

[0043] Multi-node cross-validation: For the message "Area A-3 inspection progress 95%" from node R1, extract the contemporaneous messages from neighboring nodes R2 and R3. If R2 reports "Area A-3 uninspected area ≤ 5%" and R3 reports "temperature data of Area A-3 deviates by 2% from data collected by R1", both satisfy the collaborative constraints, confirming semantic rationality; if R2 reports "Area A-3 still has 30% uninspected", the data deviation exceeds the threshold, and R1 has no valid supporting data, then it is judged as semantically abnormal.

[0044] (4) Dynamic credibility weighted calculation

[0045] Credibility metrics statistics: The server-side provides real-time statistics on three core metrics for each node:

[0046] Task completion rate : = Actual number of tasks completed / Total number of tasks assigned (R1 node had a total of 50 tasks assigned, of which 49 were completed, =0.98));

[0047] Message semantic compliance rate ):( = Number of messages that passed semantic verification / Total number of messages sent (R1 node sent 100 messages, 99 of which passed verification) =0.99);

[0048] Data consistency rate ( )): = Number of messages with data deviation ≤ threshold / Number of messages participating in cross-validation (R1 node participated in 80 cross-validation messages, 78 of which were consistent). =0.975).

[0049] Credibility score calculation: setting indicator weight coefficients =0.3、 =0.4、 =0.3, node R1 has no violations ( ),but: = 0.3×0.98 + 0.4×0.99 + 0.3×0.975 - 0 = 0.9815 If node R4 has 2 semantically abnormal messages ( ), =0.6、 =0.2、 =0.5, then: = 0.3×0.6 + 0.4×0.2 + 0.3×0.5 - 0.2 = 0.15

[0050] Voting weight mapping: There are a total of 4 nodes participating in the current round of consensus (R1 / R2 / R3 / R4). = 0.9815 + 0.952 + 0.931 + 0.15 = 3.0145, then: =0.9815 / 3.0145≈0.3256, =0.952 / 3.0145≈0.3158, =0.931 / 3.0145≈0.3088, =0.05 / 3.0145≈0.0166) (R4 node) <0.2, forced to be processed with the lowest weight), the total weight of high-confidence nodes reaches 95.02%, effectively suppressing the interference of low-confidence nodes.

[0051] Dynamic update execution:

[0052] Real-time updates: The server updates in real time after node R1 completes a task and the message verification passes. =0.9804、 =0.9901, restart calculate And synchronize to all nodes;

[0053] Periodic calibration: After every 10 collaborative tasks are completed, the credibility score is smoothed. = 0.8× +0.2 × {average score over the period}) to avoid drastic weight fluctuations caused by a single action.

[0054] (5) Semantic Enhancement Consensus Implementation

[0055] Proposal initiation: Only allowed A high-trust node (R1 / R2 / R3) with a trust level of ≥0.5 initiates a proposal. The R1 node initiates a proposal that "Inspection of Area A-3 is complete" and broadcasts it to all participating nodes.

[0056] Pre-voting: Each node votes based on the semantic verification results, and the server calculates the total weight of the proposals supported. = + + = 0.3256 + 0.3158 + 0.3088 ≈ 0.9502;

[0057] Final vote: because If the value is ≥66.7 (Byzantine fault tolerance threshold), a final vote is triggered. If the voting result is consistent with the pre-vote, the proposal is passed and a consensus is reached.

[0058] Results Feedback: The consensus result is synchronized to all nodes, and the server updates the node trust score (R1 indicates support for the correct proposal). Fine-tuned to 0.983), and optimized the rule base based on task execution feedback (adding the constraint "inspection energy consumption increases by 20% under high temperature conditions").

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. 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 be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A multi-robot Byzantine fault-tolerant consensus method based on semantic verification and dynamic weighting, characterized in that, Includes the following steps: Step 1: The server builds and initializes a scenario-based physical semantic rule base. The rule base contains three core constraints: physical law constraints, task scenario constraints, and group collaboration constraints. It supports dynamic expansion and updates. Step 2: The robot node generates a consensus message according to a standardized structured format. The message includes node ID, timestamp, task type, core data, and key fields of supporting data. Step 3: The server receives the consensus message uploaded by the node, extracts the key features of the message through the semantic parsing module, and performs a dual physical semantic rationality check of "rule matching verification + multi-node cross-verification"; Step 4: The server collects historical behavior data of each node and calculates the node's credibility score based on three indicators: task completion rate, message semantic compliance rate, and data consistency rate, combined with violation penalty items. Step 5: The server dynamically maps the credibility score to consensus voting weight, completes the weight normalization process, and performs weight suppression on low credibility nodes. Step Six: The server will synchronize the semantically verified messages and dynamic voting weights to all nodes participating in the consensus, initiating the three-stage consensus process of "proposal - pre-voting - final voting"; Step 7: The server tallies the voting results. If the total weight of the proposal is ≥66.7%, a consensus is reached and the results are synchronized to all nodes. If the threshold is not met, return to Step 2 to re-initiate the proposal. Step 8: Based on the consensus results and node behavior feedback, the server updates the physical semantic rule base parameters and node credibility scores in real time to complete dynamic iterative optimization.

2. The multi-robot Byzantine fault-tolerant consensus method based on semantic verification and dynamic weighting as described in claim 1, characterized in that, The credibility score in step four The quantitative calculation formula is as follows: = + + - in, , , Weighting coefficients of indicators =1, which can be dynamically adjusted according to the specific task scenario; The task completion rate of the i-th node ( = Number of tasks actually completed / Total number of tasks assigned (value range: 0-1). The message semantic compliance rate of the i-th node ( =Number of messages that passed semantic verification / Total number of messages sent (value range: 0-1) The data consistency rate of the i-th node ( = Number of messages with data deviation ≤ threshold / Number of messages participating in cross-validation (value range 0-1). For violations, the penalty is 0-0.5 points. Sending one semantically abnormal message will result in a penalty of 0.1 points, falsifying data will result in a penalty of 0.3 points, and failing to complete the task three times in a row will result in a penalty of 0.2 points, with a cumulative penalty not exceeding 0.5 points.

3. The multi-robot Byzantine fault-tolerant consensus method based on semantic verification and dynamic weighting as described in claim 1, characterized in that, The mapping formula for the voting weight (\(w_i\)) in step five is as follows: Where N is the total number of nodes participating in the current round of consensus, and the constraint is: =1; if < 0.2, forced setting ( = 0.05 / N), only basic voting rights are retained.

4. The multi-robot Byzantine fault-tolerant consensus method based on semantic verification and dynamic weighting as described in claim 1, characterized in that, The specific logic of "rule matching verification" in step three is as follows: the extracted key features of the message are aligned with the physical semantic rule base one by one. If any constraint rule is violated, it is directly marked as semantic anomaly and rejected from entering the subsequent consensus process. The specific logic of "multi-node cross-verification" is as follows: for messages that pass the rule matching, the messages from other nodes in the same task / region are retrieved, the core data deviation value is calculated, and if the deviation exceeds the preset threshold and there is no valid supporting data, it is judged as semantic anomaly and discarded.

5. The semantically enhanced multi-robot Byzantine fault-tolerant consensus mechanism as described in claim 1, characterized in that, The "dynamic update mechanism" in step eight includes two parts: real-time triggered updates, where the server updates the corresponding node's credibility score and voting weight in real time when a node completes a task, sends a message, or commits a violation, and synchronizes this information to all participating nodes; and periodic calibration updates, where the credibility scores of all nodes are smoothed after every 10 collaborative tasks are completed. = 0.8× Add 0.2 × {average score over the period} to avoid drastic fluctuations in weight due to a single action.

6. The multi-robot Byzantine fault-tolerant consensus method based on semantic verification and dynamic weighting as described in claim 1, characterized in that, The physical constraints include robot movement speed ≤ maximum design speed, remaining battery power ≥ energy consumption for performing the current action, and the same robot cannot appear in two physical locations simultaneously; the task scenario constraints include environmental parameters of the area to be inspected being within the normal operating range of the equipment, and fault type matching equipment parameters; the group collaboration constraints include data deviation of multiple robots collecting data in the same area ≤ threshold, and task progress being consistent with feedback from adjacent nodes.

Images