Federated learning (FL) has emerged as a robust privacy-aware decentralized computing approach for personalized data in a network of nodes. It has the potential to be the boon for future autonomous vehicles. FL can significantly improve performance while ensuring privacy preservation for all vehicles. It can exploit distributed set of training data while maintaining robust local learning within each vehicle. However, standard FL incurs long communication delays and risk of failures due to a single global server used for aggregating local learning. Any unreliable model can be uploaded by an adversary vehicle (if any), leads to deception in the learning of the whole FL system. Some fraud vehicles may often perform unreliable updates intentionally, e.g., the data poisoning attack, or unintentionally, e.g., low-quality data caused by energy constraints or high-speed 5G mobility. Therefore, investigation of efficient and trustworthy FL for autonomous vehicles becomes critical and is a challenging task. Therefore, we design a lightweight multilevel Blockchain framework for improving the end-to-end trustworthiness of the FL system of Internet of Vehicles (IoV). At the heart of the solution is the integration of trustworthiness and reputation modules, which not only learns and jointly evaluates the trustworthiness of vehicle‘s observations during data collection but also adapts prompt block verification at the Blockchain level.

联合学习（FL）已成为一种强大的隐私感知分散式计算方法，用于节点网络中的个性化数据。它有可能成为未来自动驾驶汽车的福音。FL可以显着提高性能，同时确保所有车辆的隐私保护。它可以利用分布式的训练数据集，同时在每辆车内保持强大的本地学习能力。但是，由于用于集成本地学习的单个全局服务器，标准FL会导致较长的通信延迟和失败的风险。敌方车辆（如果有）可能会上载任何不可靠的模型，从而导致整个FL系统学习的欺骗。一些欺诈工具可能经常有意执行不可靠的更新（例如，数据中毒攻击），或无意执行（例如，由能源限制或高速5G移动性引起的低质量数据。因此，研究自动驾驶车辆的高效和值得信赖的FL变得至关重要，这是一项艰巨的任务。因此，我们设计了一个轻量级的多层区块链框架，用于改善车联网（IoV）FL系统的端到端可信度。解决方案的核心是可信度和信誉模块的集成，它不仅可以学习并共同评估数据收集过程中车辆观测值的可信度，而且还可以在区块链级别进行及时的区块验证。