Use Case 3:
Intelligent connected vehicle, mission-critical services, OTA updates, FL/ML and vehicle to infrastructure communication
This use case will provide mission-critical services in the context of secure Vehicle to Infrastructure (V2I) communications, and OTA (over-the-air) vehicle system updates, with distributed learning.
The use case will be simulated using autoPI Raspberry Pi car modules deployed in real vehicles, with SIM 4G/5G interface. autoPI will be used for collection of the data from a number of vehicles to perform AI/ML based on the data shared among the vehicles and the roadside infrastructures. In this case, distributed
Learning and data sharing will provide great benefits to enhance road safety and driving experiences by supporting driving decision making.
The particular goal is given to the evaluation of the effectiveness of the proposed technology for security in future connected car scenarios. Further to this, OEMs and car manufacturers can provide the ability to train their own sensing or driving data by the on-board computing devices, and obtain corresponding parameters (e.𝑔. models/policies) as refined knowledge.
Various levels of knowledge (such as traffic statistics, traffic control, driving rules, sensing, and driving models) would be possible to share among the vehicles and the roadside infrastructures.
Data on connected vehicles will be used to emulate FL and ML problems related to network connectivity 5G/6G.
This use case will leverage cryptographic schemes developed in WP2. It will further apply findings and advancements of WP4 in the domain of DLT, decentralized identifiers and federated AI/ML.
The key validation: (i) OTA updates to help manufacturers to securely transfer software updates as part of “Connected Car” systems; (ii) adaptive asynchronous distributed learning (ADL)-based scheme will be used for model uploading and downloading car data; (iii) use blockchain to guarantee the security and tamper-proof of the shared ADL models that are important for safety and driving applications; (iv) blockchain to protect sensitive data from manipulation and to guarantee immutability of SW updates (the hash is on the ledger and the car can verify that it is getting the right one; (v) Vehicle digital identity allows to verify which car needs updates and those already updated (i.e., they have a credential certifying the performed update); (vi) evaluate secured and privacy-preserving communication and aggregation of the models in the FL server, and how to protect these from FL attacks such as membership inference, attribute inference, data reconstruction attacks, or data poisoning attacks, and potentially side-channel attacks.
KPIs: (a) Secure OTA updates to patch a vulnerability thanks to the secured update mechanisms supported by the blockchain; (b) A twofold increase in the trustworthiness of the vehicle’s digital identity with the usage of hardware security modules to embed on-chip identities. (c) 20% increase in the update cadence of the ML models, thanks to the novel distributed train and deploy frameworks. (d) FL modeling performance with serverless computing functions, using secure/privacy-preserving containers. (e) Efficiency and scalability of privacy-preserving communication and aggregation of the FL model.