This chapter provided a comparative analysis of academic breakthroughs and industrial practices in Federated Learning (FL) for smart mobility via key case studies. Academic research (Sect. 4.1) has focused on solving specific hard problems: FADNet was designed to overcome Non-IID data challenges in vehicular perception; the FedGRU scheme was proposed to handle spatio-temporal heterogeneity in traffic flow prediction (TFP); and Byzantine-robust aggregators were developed to secure open V2X networks. In contrast, industrial practice (Sect. 4.2) is driven by real-world “pain points”: Tesla’s “Shadow Mode” represents a pragmatic data loop balancing cost and efficiency; Baidu’s Apollo platform employs FL as a governance tool to break “data silos” between competing OEMs and government; and Huawei’s V2X solution highlights robustness as a fundamental prerequisite for deploying distributed security (IDS). The comparative analysis (Sect. 4.3) revealed a key gap between academia (pursuing “model optimality”) and industry (pursuing “system viability”) in performance, privacy, and deployment strategies. A core trend emerged: FL is currently used primarily as an “offline optimization” tool rather than for “real-time online learning,” and it relies heavily on complex hybrid architectures like Vehicle-Edge-Cloud (V-E-C). The real-world bottlenecks in heterogeneity, scalability, and governance revealed in this chapter establish the foundation for the discussion of future challenges in Chap. 5 .