Autonomous vehicles (AVs) increasingly rely on computation-intensive and delay-sensitive applications that must process large volumes of sensor data in real time. However, limited onboard computational resources and constrained wireless bandwidth make it challenging to meet stringent latency requirements, especially under high mobility and dense traffic. This thesis investigates the use of mobile edge computing (MEC) to enhance task offloading in vehicular networks, with a focus on reducing latency. Three complementary studies are conducted. First, an assistant vehicle offloading scheme is proposed, where idle vehicles share their unused computational resources via vehicle-to-vehicle (V2V) communication. By exploiting nearby idle vehicles without adding infrastructure or modifying the traditional roadside unit (RSU) based MEC architecture, this approach alleviates both computation and transmission bottlenecks and improves task latency. Second, building on assistant vehicle offloading, a sub-area division strategy is introduced to exploit the short-range nature of V2V links for frequency reuse. A rectangular sub-area division scheme is introduced based on vehicle positions to managed interference. The impact of task feedback and vehicle mobility across sub-areas is analyzed, and an optimal sub-area division strategy is derived to further reduce latency. Third, an MEC-assisted adaptive hybrid fusion system is proposed for cooperative perception. This system extends hybrid fusion by combining early and late fusion with an MEC-based early fusion mode at the RSU. It dynamically allocates perception tasks across these fusion schemes according to accuracy requirements, MEC computational capacity, and available bandwidth, thereby minimizing latency while maintaining target perception performance. Simulation results of all three studies are presented to verify the improvement of the latency performance of the proposed methods.
Yilun Zhang (Sun,) studied this question.