Decentralized allocation method and system for autonomous driving model based on smart contract

Abstract

This invention discloses a decentralized allocation method and system for autonomous driving models based on smart contracts, relating to the field of autonomous driving. The method includes: receiving a model acquisition request, wherein the model acquisition request includes a data packet describing the requirements and a budget limit; the model acquisition request is sent through a purchasing node; acquiring autonomous driving model-related information and a sealed bid corresponding to each autonomous driving model based on the model acquisition request, wherein the sealed bid matches the data packet containing the requirements description, and the autonomous driving model-related information and the sealed bid corresponding to each autonomous driving model are sent through a provider node; evaluating the autonomous driving models based on preset multi-dimensional indicators to obtain a model quality score; filtering the autonomous driving models based on the model quality score and the sealed bid corresponding to each autonomous driving model to obtain a target autonomous driving model; performing payment settlement based on the target autonomous driving model and outputting the target autonomous driving model to the purchasing node.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving, specifically to a decentralized allocation method and system for autonomous driving models based on smart contracts. Background Technology

[0002] To enable autonomous vehicles to quickly adapt to different environments, a feasible solution is to enable model switching and reuse in cross-domain scenarios. This means that when faced with a fundamental shift in tasks, vehicles can directly call mature models that have been fully validated in the target domain, instead of starting data collection and model training from scratch. This significantly reduces adaptation costs and time, and maximizes the benefits for model developers.

[0003] However, current model resources are highly fragmented, and due to their strategic value, companies are generally unwilling to share them. Even when they are willing to share, there is a lack of efficient circulation mechanisms. This leads to the widespread existence of resource silos across enterprises and regions, making direct acquisition and integration of mature external models face significant structural barriers. Although platforms such as Hugging Face and Replicate provide infrastructure support for model sharing, they still have significant shortcomings in reliable model quality assessment and contributor incentive mechanisms, making it difficult to achieve a sustainable supply and circulation of high-quality models. Summary of the Invention

[0004] To address the shortcomings mentioned in the background section, the present invention aims to provide a decentralized allocation method and system for autonomous driving models based on smart contracts.

[0005] Firstly, the objective of this invention can be achieved through the following technical solution: a decentralized allocation method for an autonomous driving model based on smart contracts, the method comprising the following steps: Receive a model acquisition request, wherein the model acquisition request includes a data packet containing a requirement description and a budget limit; the model acquisition request is sent through the purchasing party node; Based on the model acquisition request, information related to autonomous driving models and the sealed quotation corresponding to each autonomous driving model are obtained. The sealed quotation is matched with a data packet containing a requirement description. The information related to autonomous driving models and the sealed quotation corresponding to each autonomous driving model are sent based on the provider node. The autonomous driving model is evaluated based on preset multi-dimensional indicators to obtain a model quality score. Based on the model quality score and the sealed bid corresponding to each autonomous driving model, the autonomous driving models are screened to obtain the target autonomous driving model. Payment and settlement are performed based on the target autonomous driving model, and the target autonomous driving model is output to the purchasing node.

[0006] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the autonomous driving model is stored in IPFS after being verified by the contract, and the metadata and hash value of the autonomous driving model are recorded on the blockchain.

[0007] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the preset multi-dimensional indicators include four quality dimensions: safety indicators, efficiency indicators, compliance indicators, and comfort indicators.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the process of evaluating the autonomous driving model based on preset multi-dimensional indicators to obtain a model quality score, as follows: The MEREC (Objective Weighting Method Based on Criterion Removal Effect) method is used to calculate the quantitative weight index of the four quality dimensions. MEREC uses mathematical simulation to quantify the change in the overall performance of the model after removing each dimension in turn. Based on these weights, the quality score of each model is calculated using the MARCOS (Solution Measurement and Ranking Method Based on Compromise Solution) method.

[0009] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the processing flow of the MEREC method is as follows: Using formula (1), a decision matrix X is constructed based on m models and n evaluation indicators, where, This represents the raw score of the i-th model on the j-th metric: (1) MEREC uses specific standardization methods based on the type of indicator: benefit-type indicators are standardized using formula (2), and cost-type indicators are standardized using formula (3); After matrix normalization, the composite score for each indicator is calculated using formula (4). MEREC uses the natural logarithm and exponential function for the calculation: (4) Remove each indicator j in turn, and recalculate the overall score of each model using formula (5). : (5) For each indicator j, the absolute score changes of all models after its removal are summarized, as shown in formula (6): (6) The total removal effect of each indicator is normalized to obtain the final weight, as shown in formula (7): (7).

[0010] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: the processing flow of the MARCOS method is as follows: The MARCOS method is employed to simultaneously measure the relative utility of the model with respect to both the ideal solution AI and the anti-ideal solution AAI. MARCOS incorporates extended decision matrices of both the ideal solution AI and the anti-ideal solution AAI. As shown in formula (8): (8) For the extended initial matrix Standardization is performed. For benefit-based standards, formula (9) is used for standardization; for cost-based standards, formula (10) is used for standardization. Using formula (11), the standardized extended matrix is... Multiplying each element in the matrix by its corresponding weight constructs the weighted normalized matrix V: (11) Calculate the utility of each model, and sum the weighted values ​​of each model across all metrics, denoted as . The utility of the model relative to the ideal solution and the antiideal solution is calculated using formulas (12) and (13) respectively, and the final utility score is determined by formula (14). As a model quality score: (12) (13) (14) The final utility score is standardized to the range of 0 to 100, and the conversion formula is shown in (15): (15) in and These represent the maximum and minimum utility scores of the candidate models, respectively.

[0011] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: a payment mechanism based on the VCG principle for payment settlement performed on the target autonomous driving model. Under the VCG mechanism, the winning bidder w does not need to pay its own bid, but instead pays for the externalities it causes to other participants: that is, the total social welfare loss caused by w's participation in the auction. The amount paid by the winning bidder is... The calculation formula is as follows: , in This represents the maximum social welfare achievable without w. This represents the sum of benefits for all other bidders when w wins the bid.

[0012] Secondly, in order to achieve the above objectives, this invention discloses a decentralized allocation system for autonomous driving models based on smart contracts, comprising: The request receiving module is used to receive model acquisition requests, wherein the model acquisition request includes a data packet containing a requirement description and a budget limit; the model acquisition request is sent through the purchasing party node; The data acquisition module is used to acquire information related to autonomous driving models and the sealed quotation corresponding to each autonomous driving model based on the model acquisition request. The sealed quotation is matched with a data packet containing a requirement description. The information related to autonomous driving models and the sealed quotation corresponding to each autonomous driving model are sent based on the provider node. The screening and processing module is used to evaluate autonomous driving models based on preset multi-dimensional indicators to obtain model quality scores. Based on the model quality scores and the sealed bids corresponding to each autonomous driving model, the autonomous driving models are screened to obtain target autonomous driving models. Payment settlement is performed based on the target autonomous driving models, and the target autonomous driving models are output to the purchasing node.

[0013] In another aspect of the present invention, in order to achieve the above-mentioned objective, a terminal device is disclosed, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The memory stores the computer program capable of running on the processor, and when the processor loads and executes the computer program, it employs the decentralized allocation method of the autonomous driving model based on smart contracts as described above.

[0014] In another aspect of the present invention, in order to achieve the above-mentioned objective, a computer-readable storage medium is disclosed, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is loaded and executed by a processor, it employs the decentralized allocation method of the autonomous driving model based on smart contracts as described above.

[0015] The beneficial effects of this invention are: This invention incentivizes model providers to contribute resources, enabling vehicle manufacturers to acquire suitable models on demand, thereby reducing the cost of repeated model training and improving the adaptability of autonomous driving systems in cross-domain environments. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the architecture of the present invention; Figure 3 This is a schematic diagram of the reverse auction process of the present invention; Figure 4 This is a schematic diagram of the four-dimensional evaluation index for autonomous driving in this invention; Figure 5 This is a schematic diagram of the weight and total score calculation process of the model of this invention; Figure 6 This is a schematic diagram of the cross-domain model procurement process for cross-city autonomous taxis according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the dynamic evaluation and incentive update process of the unmanned truck model for port logistics according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the system structure of the present invention; Figure 9 This is a schematic diagram illustrating the performance of various indicators related to the utility of both buyers and sellers, total social welfare, and the cost-effectiveness of the final model obtained in the embodiments of the present invention. Figure 10 This diagram illustrates the calculation results of the weights of each dimension and the total model score under six different models within an embodiment of the present invention, and compares them with other calculation methods. Detailed Implementation

[0017] 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.

[0018] Example 1: like Figure 1 As shown, the decentralized allocation method for the autonomous driving model based on smart contracts includes the following steps: S101: Receive a model acquisition request, wherein the model acquisition request includes a data packet containing a requirement description and a budget limit; the model acquisition request is sent through the purchasing party node; S102: Obtain autonomous driving model-related information and sealed quotation corresponding to each autonomous driving model based on the model acquisition request, wherein the sealed quotation is matched with a data packet containing a requirement description, and the autonomous driving model-related information and sealed quotation corresponding to each autonomous driving model are sent based on the provider node; The autonomous driving model is stored in IPFS after being verified by the contract, and the metadata and hash value of the autonomous driving model are recorded on the blockchain.

[0019] S103: Evaluate the autonomous driving model based on preset multi-dimensional indicators to obtain the model quality score. Based on the model quality score and the sealed quotation corresponding to each autonomous driving model, filter the autonomous driving models to obtain the target autonomous driving model. Perform payment settlement based on the target autonomous driving model and output the target autonomous driving model to the purchasing node.

[0020] The preset multi-dimensional indicators include four quality dimensions: safety, efficiency, compliance, and comfort.

[0021] The process of evaluating autonomous driving models based on preset multi-dimensional indicators to obtain model quality scores is as follows: To accurately reflect the relative importance of different quality dimensions and enable the framework (AD-DRAC) to select the truly optimal model, AD-DRAC employs a data-driven approach to determine the weights of each evaluation dimension. To this end, AD-DRAC uses the MEREC method to calculate the quantified weights of the four model quality dimensions. Compared to other methods (such as equal weighting or AHP), MEREC quantifies the importance of each dimension by mathematically simulating the changes in overall model performance after sequentially removing each dimension. This method has a key advantage: MEREC endogeneously derives the weights solely from the statistical distribution and internal consistency of the candidate model data.

[0022] To ensure cross-sectional fairness in the evaluation, AD-DRAC calculates a uniform quantitative weight for each auction. This weight is calculated based on a matrix of all candidate models in a given auction and is then applied to each model to ensure that all scores are generated under the same evaluation criteria. The specific process of MEREC is as follows: Using formula (1), a decision matrix X is constructed based on m models and n evaluation indicators. Wherein, This represents the raw score of the i-th model on the j-th metric.

[0023] Since the units and directions of the various indicators differ, the initial decision matrix needs to be standardized into a dimensionless unified matrix, and the higher the value in the matrix, the better the performance. MEREC adopts a specific standardization method according to the type of indicator: benefit-type indicators (the higher the value, the better) are standardized by formula (2), and cost-type indicators (the lower the value, the better) are standardized by formula (3).

[0024] After matrix normalization, the composite score for each indicator needs to be calculated using formula (4). MEREC uses the natural logarithm and exponential function for calculation, making it highly sensitive to changes in indicator values.

[0025] (4) Next, we remove each indicator j in turn and recalculate the overall score of each model using formula (5). : (5) For each indicator j, we summarize the absolute score changes of all models after its removal, as shown in Equation (6): (6) The larger the value, the greater the interference of removing metric j on global performance, that is, the higher the importance of the metric.

[0026] Finally, the total removal effect of each indicator is normalized to obtain its final weight, as shown in formula (7): (7) To achieve fair and robust model selection across multiple often conflicting evaluation criteria such as safety, efficiency, and compliance, AD-DRAC calculates a comprehensive utility score that reflects the overall performance of each candidate model while avoiding over-reliance on any single dimension. To this end, we employ the MARCOS method, which constructs a trade-off ranking by simultaneously measuring the relative utility of the model with respect to the ideal solution (AI, composed of the best values ​​under all criteria) and the anti-ideal solution (AAI, composed of the worst values ​​under all criteria). Unlike simple weighted sum methods (which may mask key weaknesses by averaging good and bad dimensions), MARCOS explicitly considers both the model's proximity to the optimal scenario and its deviation from the worst scenario. This dual-reference mechanism ensures that models with serious deficiencies in key aspects (e.g., frequent collisions) are appropriately penalized, while favoring candidate models that achieve balanced and reliable performance across all key dimensions. An extended decision matrix incorporating the ideal solution (AI) and the anti-ideal solution (AAI) is used. As shown in formula (8): (8) Subsequently, the extended initial matrix was... Standardization is performed. For benefit-based standards, formula (9) is used for standardization; for cost-based standards, formula (10) is used for standardization.

[0027] Using formula (11), the standardized extended matrix is... By multiplying each element in the matrix by its corresponding weight, a weighted normalized matrix V can be constructed.

[0028] (11) Based on the above steps, the utility of each model is calculated. First, the sum of the weighted values ​​of each model across all indicators is calculated, denoted as . Subsequently, the utility of the model relative to the ideal and antiideal solutions is calculated using formulas (12) and (13), respectively. Finally, the final utility score is determined using formula (14). : (12) (13) (14) After calculating the final utility score for each candidate model To improve the interpretability and comparability of the results, we standardized the scores to the range of 0 to 100, and the conversion formula is shown in (15): , (15) in and These represent the maximum and minimum utility scores of the candidate models, respectively. This transformation ensures that the model with the highest score is assigned 100 points, the model with the lowest score is assigned 0 points, and the scores of other models are linearly distributed between these two values. The final scores can be directly used for market demonstration and decision-making.

[0029] The process of selecting the winning model is as follows: In model procurement under strict budget constraints, selecting solely based on the highest quality or lowest price often leads to suboptimal results: either overspending for marginal performance improvements or sacrificing critical performance. To address the multi-objective trade-off between cost and utility without relying on ad hoc weight settings, AD-DRAC employs Pareto optimization. This method maps each candidate model to a two-dimensional "price-total-score" space and identifies the Pareto front. The Pareto front consists of all non-dominated models that cannot optimize one dimension without sacrificing another. By retaining all valid trade-offs, Pareto optimization avoids prematurely setting subjective quality-price weights while eliminating clearly inferior options.

[0030] From the Pareto frontier, the ultimate winner will be selected at the "knee point": the solution where the performance return from additional cost begins to diminish. This selection explicitly prioritizes cost-effectiveness over maximizing scores, avoiding the pitfall of choosing expensive models (whose performance offers only a negligible improvement over significantly cheaper alternatives). As a data-driven inflection point, this ensures that the buyer obtains near-optimal model quality at the most cost-effective price.

[0031] The process of performing payment settlement based on the target autonomous driving model: A well-designed payment rule is crucial for preventing strategic misreporting and ensuring the continued willingness of high-quality model providers to participate. In the AD-DRAC reverse auction, model performance and cost differences are significant, and simple pricing methods (such as payment based on bids or fixed second-lowest price) cannot reflect the true impact of each bidder on the overall framework welfare. Therefore, we adopt a payment mechanism based on the VCG (Vickrey-Clarke-Groves) principle, which guarantees incentive compatibility: each model provider only needs to report its true costs to maximize its own utility, unaffected by the behavior of other participants. Under the VCG mechanism, the winning bidder w does not need to pay its own bid, but rather pays for the externalities it causes to other participants: the loss of total social welfare due to w's participation in the auction. The winning bidder pays the amount specified in the figure. The calculation formula is as follows: (16) in This represents the maximum social welfare achievable without w. This represents the sum of the benefits of all other bidders when w wins the bid. This design ensures that the payment amount reflects only the opportunity cost of the framework, rather than arbitrary markups or price distortions.

[0032] While VCG-based payment methods and Pareto optimal selection ensure accurate pricing and efficient allocation in a single transaction, they cannot directly incentivize unsuccessful participants to continuously improve model quality. To address this deficiency, AD-DRAC introduces a dynamic reward mechanism: the platform allocates funds from the reward pool... A portion of the funds will be extracted to reward the most cost-effective participants in each round of auctions. The top three models are ranked (based on a non-Pareto dominance principle). This design explicitly rewards models with superior performance per unit cost, incentivizing broader participation and long-term quality investment. Reward amounts are allocated according to preset ranking weights and further adjusted based on the model's absolute score, ensuring that both ranking and absolute quality are considered. This mechanism prevents the market from stagnating due to a few dominant providers.

[0033] Specifically, the present invention will be further illustrated below through embodiments: AD-DRAC's core components include the Ethereum blockchain, smart contracts, model providers, model purchasers, and a trading market. The architecture uses the Ethereum blockchain as its underlying infrastructure, with all nodes collectively maintaining a decentralized, immutable, and trusted trading environment, independent of any third-party institutions. Smart contracts perform multiple key functions: preventing duplicate uploads and tampering of models by recording and comparing model hash values; automatically executing the auction process after a purchase request is initiated by the buyer; escrowing the tokens pledged by both buyers and sellers to ensure contract fulfillment; and writing key information from the transaction process into the blockchain for transparent verification and oversight by all participants.

[0034] In the reverse auction process, model providers must deposit a security deposit (to ensure the integrity of their actions), an initial CARLA Leaderboard score, the model's applicability, and security credentials, including scene suitability level, security certification number, and hardware compatibility specifications, into the smart contract when uploading their models. After contract verification, the model is stored in IPFS, and its metadata and hash value are recorded on-chain. Autonomous driving manufacturers with purchasing needs submit transaction requests and budgets to the market, staking tokens equal to the budget. Upon receiving the requests, the market filters out all eligible models. Model providers submit sealed bids to the smart contract based on their actual costs and expected returns. Simultaneously, the market calculates the quality score of each model off-chain using the MEREC and MARCOS methods, considering four dimensions: security, efficiency, compliance, and comfort. After eliminating unqualified models with scores below the threshold and models exceeding the budget, the winning model is selected through Pareto optimization, taking into account the bids. The contract will retrieve the winning model file from IPFS based on the model hash value stored on the blockchain and deliver it to the purchaser. At the same time, the purchaser's deposit will be automatically deducted from the transaction price and paid to the winning model provider, and the balance will be refunded.

[0035] The specific process for cross-regional model procurement for intercity autonomous taxis is as follows: This example applies to an autonomous taxi operating company (the purchaser), whose fleet originally operated in cities with warm climates and simple road rules (such as Shenzhen). They now plan to expand their operations to cities with complex climates and varied road conditions (such as Harbin, involving icy and snowy roads). Because retraining the model is time-consuming and expensive, the buyer decided to purchase a mature autonomous driving model that had already been validated in the target cities through a model trading platform.

[0036] General process: Demand Posting: Buyers post their model procurement needs and budgets for specific scenarios (ice and snow, urban roads) on the model trading market.

[0037] Model bidding: Multiple model vendors (sellers) upload model metadata that matches the tags and submit sealed bids.

[0038] Automated verification and testing: The smart contract verifies the model's security certificate and triggers the off-chain computing environment to calculate the model's score.

[0039] Optimal decision-making: The system is based on the Pareto optimization algorithm to find the "knee" model with the best cost performance between "price" and "performance".

[0040] Transaction and Delivery: The smart contract automatically executes VCG payments, and the purchaser obtains the right to download the model.

[0041] Detailed implementation process of the technical solution: Step S1: The buyer requests initialization. Buyer node. Send a data packet containing a request description to the smart contract and budget cap The requirements description includes tags for the target application scope (such as "Snowy", "Urban", "Traffic Light"). The purchaser also pledges tokens as collateral to the smart contract.

[0042] Step S2: Seller Model Registration and Bidding. Multiple model suppliers (such as...) , , ... The system receives a request. If its model label matches, it participates in the bidding.

[0043] The model files are encrypted and stored on IPFS (InterPlanetary File System), and the smart contract puts the hash value and metadata on the blockchain.

[0044] Suppliers submit an initial benchmark score (CARLA Leaderboard score) and security certificate. The smart contract automatically verifies the certificate's validity.

[0045] Suppliers based on their own costs Submit a sealed quote To ensure honest pricing, the system design satisfies individual rationality constraints. ≥ .

[0046] Step S3: Multidimensional Data Acquisition and Scoring Calculation. The system triggers off-chain computing nodes to collect runtime data of each candidate model in the test environment. Data dimensions include: Safety: Time to Collision (TTC), Collision Rate (CR), Number of times manual intervention is required.

[0047] Efficiency: Average speed, maximum lane center deviation.

[0048] Compliance: Number of traffic violations.

[0049] Comfort: Rate of change of acceleration (Jerk), maximum acceleration. A comprehensive score for each model was calculated using the MEREC-MARCOS algorithm.

[0050] Step S4: Winner selection based on the Pareto front. The system maps all candidate models in a two-dimensional "quote-score" coordinate system.

[0051] Construct the frontier: filter out all uncontrolled solutions (i.e., there is no other model that has a lower price and a higher score).

[0052] Knee point localization: Calculate the slope change between adjacent points on the Pareto front (i.e., marginal score gain rate).

[0053] Final selection: The point with the largest change in slope is chosen as the winner. This point represents the critical point where, within the budget, the performance improvement from adding additional costs begins to show diminishing marginal returns, thus ensuring the buyer obtains the best cost-effectiveness.

[0054] Step S5: Execution based on the VCG payment mechanism. The smart contract calculates the final payment price based on the VCG-based payment mechanism. The payment amount is not equal to the seller's offer price, but rather to the social welfare loss caused to other participants by the seller's participation.

[0055] like Figure 9 As shown, the proposed framework can help autonomous taxis select the most cost-effective model while maximizing the utility of both the model purchaser and the model provider. Specifically, the cost-effectiveness of the winning model selected by this framework is 110.01%, 9.80%, and 82.92% higher than that of the arbitrary selection method, the lowest bid selection method, and the highest score selection method, respectively. The total social welfare generated by the buyer and seller using this framework method is 34.86%, 35.67%, and 33.70% higher than that of other schemes such as MP (winner pays the highest loser's bid), GFP (payment equal to the winner's own bid), and RRAFL (payment based on the unit bid of the next selected bidder), respectively.

[0056] The process of dynamic evaluation and incentive update for unmanned trucks in port logistics is as follows: This embodiment applies to a closed smart port. The port operator (buyer) needs to regularly update the autonomous driving algorithms of the unmanned automated guided vehicles (AGVs) within the port to adapt to the new cargo yard layout. Since the weight of "efficiency" and "safety" in port operations varies with the seasons (e.g., efficiency is emphasized during busy seasons and safety during rainy seasons), a system is needed that can automatically adjust the evaluation weights and incentivize unsuccessful suppliers to continuously optimize their models.

[0057] General process: Data collection: Extract historical operating indicators for candidate models from port operation logs.

[0058] Dynamic Weight Calculation (MEREC): Using the MEREC algorithm, objective weights of each evaluation index are automatically calculated based on the data distribution characteristics of the current candidate model, without the need for human expert scoring.

[0059] Comprehensive Utility Score (MARCOS): Construct ideal and negative ideal solutions, calculate the distance between each model and the best / worst state, and derive the final ranking.

[0060] Incentive allocation: In addition to paying the winner, the system distributes rewards from the reward pool to the top three unselected models with the best cost performance to prevent market monopoly.

[0061] Detailed implementation process of the technical solution: Step S1: Construct a standardized decision matrix. Collect n indicator data from m candidate models to construct the matrix. .

[0062] Standardize the data: For benefit-oriented indicators (such as average speed), use... For cost-related metrics (such as collision rate, Jerk rate), the following approach is adopted: Step S2: Objective weight calculation based on MEREC. This step aims to address the problem of traditional AHP methods relying on subjective expert scoring.

[0063] Calculate the overall performance of all models Removal effect simulation: Assume the j-th indicator is removed sequentially, and recalculate the model performance. .

[0064] Deviation Calculation: Calculate the sum of the absolute values ​​of the deviations of all model performances from the original performance after removing this indicator. .

[0065] Weight determination: The larger the value, the more important the indicator is in distinguishing the quality of the models. Normalization Obtain the final weight Effect: If all models perform similarly on the "safety" metric, the weight of this metric will automatically decrease; if there are significant differences in "efficiency," the weight of this metric will automatically increase.

[0066] Step S3: Utility classification based on MARCOS.

[0067] Define reference solution: construct ideal solution (AI, where all indicators are optimal) and negative ideal solution (AAI, where all indicators are worst).

[0068] Weighted normalization: The weights obtained in step S2 are... Applied to standardized matrices.

[0069] Utility calculation: Calculate the utility of each model relative to the ideal solution. and the utility relative to the negative ideal solution .

[0070] Final score function: via formula Integrate the two utilities and linearly map them to a 0-100 scale.

[0071] Step S4: Incentive pool allocation based on cost-effectiveness ranking. To maintain ecosystem activity and prevent a winner-takes-all scenario that leads to other developers leaving, a reward mechanism is introduced.

[0072] Cost-effectiveness ranking: The system does not rely on Pareto results, but rather on the "cost-effectiveness ratio" (Score / Bid) to rank all candidate models.

[0073] Reward distribution: A certain percentage will be extracted from the platform's reserved reward pool and distributed to the top three models in terms of cost-effectiveness (even if they did not win this auction).

[0074] Weighted Allocation: Reward amounts are allocated based on a weighted average of ranking weights and model absolute scores. This incentivizes "potential" models that, while not currently top-tier in performance, possess significant cost advantages to remain on the platform and continue optimization.

[0075] like Figure 10 As shown, the model score calculation scheme proposed in this framework takes into account the distribution of scores in different dimensions of each candidate model, which can help unmanned trucks to quantitatively calculate the quality scores of candidate models and effectively distinguish the merits of models.

[0076] Example 2: To achieve the above objective, such as Figure 8 As shown, based on Embodiment 1, this invention discloses a decentralized allocation system for autonomous driving models based on smart contracts, comprising: The request receiving module 11 is used to receive a model acquisition request, wherein the model acquisition request includes a data packet containing a requirement description and a budget limit; the model acquisition request is sent through the purchasing party node; Data acquisition module 12 is used to acquire autonomous driving model-related information and sealed quotation corresponding to each autonomous driving model based on the model acquisition request, wherein the sealed quotation is matched with a data packet containing a requirement description, and the autonomous driving model-related information and sealed quotation corresponding to each autonomous driving model are sent based on the provider node; The screening and processing module 13 is used to evaluate the autonomous driving model based on preset multi-dimensional indicators to obtain the model quality score, screen the autonomous driving model based on the model quality score and the sealed quotation corresponding to each autonomous driving model to obtain the target autonomous driving model, perform payment settlement based on the target autonomous driving model, and output the target autonomous driving model to the purchasing node.

[0077] Based on the same inventive concept, this invention also provides a computer device, comprising: one or more processors, and a memory for storing one or more computer programs; the programs include program instructions, and the processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, used to implement one or more instructions, specifically for loading and executing one or more instructions stored in a computer storage medium to implement the above-described method.

[0078] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, performs the above-described method. This storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0079] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0080] The foregoing has shown and described the basic principles, main features, and advantages of this disclosure. Those skilled in the art should understand that this disclosure is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this disclosure. Various changes and modifications can be made to this disclosure without departing from its spirit and scope, and all such changes and modifications fall within the scope of this disclosure as claimed.

Claims

1. A decentralized allocation method for autonomous driving models based on smart contracts, characterized in that, The method includes the following steps: Receive a model acquisition request, wherein the model acquisition request includes a data packet containing a requirement description and a budget limit; the model acquisition request is sent through the purchasing party node; Based on the model acquisition request, information related to autonomous driving models and the sealed quotation corresponding to each autonomous driving model are obtained. The sealed quotation is matched with a data packet containing a requirement description. The information related to autonomous driving models and the sealed quotation corresponding to each autonomous driving model are sent based on the provider node. The autonomous driving model is evaluated based on preset multi-dimensional indicators to obtain a model quality score. Based on the model quality score and the sealed bid corresponding to each autonomous driving model, the autonomous driving models are screened to obtain the target autonomous driving model. Payment and settlement are performed based on the target autonomous driving model, and the target autonomous driving model is output to the purchasing node.

2. The decentralized allocation method for autonomous driving models based on smart contracts according to claim 1, characterized in that, The autonomous driving model is stored in IPFS (InterPlanetary File System) after being verified by the contract, and the metadata and hash value of the autonomous driving model are recorded on the blockchain.

3. The decentralized allocation method for autonomous driving models based on smart contracts according to claim 1, characterized in that, The preset multi-dimensional indicators include four quality dimensions: safety indicators, efficiency indicators, compliance indicators, and comfort indicators.

4. The decentralized allocation method for autonomous driving models based on smart contracts according to claim 1, characterized in that, The process of evaluating the autonomous driving model based on preset multi-dimensional indicators to obtain a model quality score is as follows: The MEREC (Objective Weighting Method Based on Criterion Removal Effect) method is used to calculate the quantitative weight index of the four quality dimensions. MEREC uses mathematical simulation to quantify the change in the overall performance of the model after removing each dimension in turn. Based on these weights, the quality score of each model is calculated using the MARCOS (Solution Measurement and Ranking Method Based on Compromise Solution) method.

5. The decentralized allocation method for autonomous driving models based on smart contracts according to claim 4, characterized in that, The processing flow of the MEREC method is as follows: Using formula (1), a decision matrix X is constructed based on m models and n evaluation indicators, where, This represents the raw score of the i-th model on the j-th metric: (1) MEREC uses specific standardization methods based on the type of indicator: benefit-type indicators are standardized using formula (2), and cost-type indicators are standardized using formula (3); After matrix normalization, the composite score for each indicator is calculated using formula (4). MEREC uses the natural logarithm and exponential function for the calculation: (4) Remove each indicator j in turn, and recalculate the overall score of each model using formula (5). : (5) For each indicator j, the absolute score changes of all models after its removal are summarized, as shown in formula (6): (6) The total removal effect of each indicator is normalized to obtain the final weight, as shown in formula (7): (7)。 6. The decentralized allocation method for autonomous driving models based on smart contracts according to claim 5, characterized in that, The processing flow of the MARCOS method is as follows: The MARCOS method is employed to simultaneously measure the relative utility of the model with respect to both the ideal solution AI and the anti-ideal solution AAI. MARCOS incorporates extended decision matrices of both the ideal solution AI and the anti-ideal solution AAI. As shown in formula (8): (8) For the extended initial matrix Standardization is performed. For benefit-based standards, formula (9) is used for standardization; for cost-based standards, formula (10) is used for standardization. Using formula (11), the standardized extended matrix is... Multiplying each element in the matrix by its corresponding weight constructs the weighted normalized matrix V: (11) Calculate the utility of each model, and sum the weighted values ​​of each model across all metrics, denoted as . The utility of the model relative to the ideal solution and the antiideal solution is calculated using formulas (12) and (13) respectively, and the final utility score is determined by formula (14). As a model quality score: (12) (13) (14) The final utility score is standardized to the range of 0 to 100, and the conversion formula is shown in (15): (15) in and These represent the maximum and minimum utility scores of the candidate models, respectively.

7. The decentralized allocation method for autonomous driving models based on smart contracts according to claim 1, characterized in that, The payment settlement mechanism based on the target autonomous driving model adopts the VCG principle. Under the VCG mechanism, the winning bidder w does not need to pay its own bid, but instead pays for the externalities it causes to other participants: that is, the total social welfare loss caused by w's participation in the auction. The amount paid by the winning bidder is... The calculation formula is as follows: （16） in This represents the maximum social welfare achievable without w. This represents the sum of benefits for all other bidders when w wins the bid.

8. A decentralized allocation system for autonomous driving models based on smart contracts, employing the decentralized allocation method for autonomous driving models based on smart contracts as described in any one of claims 1 to 7, characterized in that, include: A request receiving module is used to receive a model acquisition request, wherein the model acquisition request includes a data packet containing a requirement description and a budget limit; the model acquisition request is sent through the purchasing party node; The data acquisition module is used to acquire information related to autonomous driving models and the sealed quotation corresponding to each autonomous driving model based on the model acquisition request. The sealed quotation is matched with a data packet containing a requirement description. The information related to autonomous driving models and the sealed quotation corresponding to each autonomous driving model are sent based on the provider node. The screening and processing module is used to evaluate autonomous driving models based on preset multi-dimensional indicators to obtain model quality scores. Based on the model quality scores and the sealed bids corresponding to each autonomous driving model, the autonomous driving models are screened to obtain target autonomous driving models. Payment settlement is performed based on the target autonomous driving models, and the target autonomous driving models are output to the purchasing node.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, The memory stores a computer program that can run on the processor. When the processor loads and executes the computer program, it employs the decentralized allocation method for the autonomous driving model based on smart contracts, as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is loaded and executed by the processor, it employs the decentralized allocation method for the autonomous driving model based on smart contracts, as described in any one of claims 1 to 7.

Images