A self-driving car model sharing method using blockchain

Abstract

The invention discloses a self-driving car model sharing method using blockchain, aiming at improving the accuracy and efficiency of decision-making of the self-driving car. The implementation steps include: 1) building a mobile edge computing network; 2) generating a key pair for each node in the mobile edge computing network; 3) building a local model set of mobile node sets in the mobile edge computing network; 4) each mobile node Communicate with the nearest mobile edge computing node; 5) The mobile edge computing node set obtains the super node sequence; 6) Build a blockchain based on the super node sequence; 7) Update the local model set. Compared with the existing automatic driving vehicle model training method, the present invention effectively improves the decision-making accuracy and efficiency when the model is used for decision-making during the driving process of the automatic driving vehicle.

Description

technical field [0001] The invention belongs to the technical field of automatic driving, and relates to a method for sharing an automatic driving vehicle model using a block chain, which can be used to realize safer and more reliable automatic driving. Background technique [0002] When a self-driving car encounters an obstacle while driving, it needs to make decisions about deceleration, acceleration, and turning. The learning-based method is a typical decision-making method for a self-driving car. The model needs to be trained during the decision-making process, and the accuracy of the model and The efficiency of the training model is a key factor in determining the accuracy and efficiency of decision-making. [0003] At present, the methods for training models are divided into bicycle-based model training methods and cloud-based model training methods. Among them, the bicycle-based model training method is that the self-driving car independently completes data collection...

AI Technical Summary

Problems solved by technology

[0003] At present, the methods for training models are divided into bicycle-based model training methods and cloud-based model training methods. Among them, the bicycle-based model training method is that the self-driving car independently completes data collection and model training. The number of car sensors and driving scenarios are limited. When the obtained model is used for automatic driving decision-making, the accuracy of decision-making is difficult to guarantee, and due to the limited computing power of a single car, the efficiency of decision-making is low.

[0004] The cloud-based model training method is that the self-driving car uploads the data collected by the sensor to the cloud center, trains the model on the cloud center, and then the self-driving car downloads the model from the cloud. This method solves the problem of the number of sensors of a single self-driving vehicle, The problem of limited driving scenarios and computing power is currently a widely used method for training models of autonomous vehicles. For example, the application publication number is CN110196593A, and the patent application titled "A System and Method for Automatic Driving Multi-Scenario Environment Detection and Decision-making", A cloud-based model training method is disclosed, which compresses and stores the data collected by the self-driving car through the on-board core sensor and uploads it to the cloud center regularly, and trains the model through machine learning on the cloud center. The training method obtained When the model is applied to the decision-making process of autonomous vehicles, it has high decision-making accuracy and efficiency, but its shortcomings are: 1. The model can only be trained in the cloud center. Once the cloud center server fails, the model training will If it cannot be completed, the self-driving car will not be able to download the model from the cloud center to make decisions; 2. Training the model on the cloud cannot guarantee that the data will not be tampered with by malicious nodes during the model training process. Once the data is tampered with, the model trained by the cloud center will be inaccurate and automatically Driving a car and then downloading an inaccurate model from the cloud center for decision-making, the accuracy of the decision-making will be low; 3. When the self-driving car uploads data to the cloud, the cloud does not verify the identity of the self-driving car. Once a malicious node uploads Wrong data, the cloud uses wrong data to train the model, and then the self-driving car downloads the wrong model from the cloud for decision-making, and the accuracy of decision-making is low; Causes a lot of burden, the data upload speed will be slow, resulting in low efficiency of model training, so the efficiency of automatic driving decision-making is low

Images