Why Oranits' 12.5% Boost in Autonomous Vehicle Task Offloading Is a Game-Changer for Edge AI

A team of researchers from multiple institutions has published Oranits, a novel system model that addresses mission assignment and task offloading in Open Radio Access Network (Open RAN)-based intelligent transportation systems (ITS). The work, now available on arXiv and accepted by IEEE Transactions on Intelligent Transportation Systems, targets the growing need for efficient edge computing as autonomous vehicles generate massive amounts of data requiring real-time processing. Existing approaches often ignore the interdependencies between missions and the costs of offloading tasks to edge servers, leading to suboptimal performance. Oranits explicitly models these dependencies and offloading costs while leveraging vehicle cooperation to optimize overall system benefit.

The proposed solution employs a twofold optimization approach. First, the Chaotic Gaussian-based Global ARO (CGG-ARO) metaheuristic algorithm serves as a baseline for single-slot optimization, improving mission completion rates by 7.1% and overall benefit by 7.7%. Second, the Multi-agent Double Deep Q-Network (MA-DDQN) framework integrates both multi-agent coordination and multi-action selection mechanisms. This deep reinforcement learning approach significantly reduces mission assignment time and improves adaptability, achieving even greater gains: 11.0% improvement in completed missions and 12.5% improvement in overall benefit. These results demonstrate Oranits's ability to enable faster, more adaptive, and more efficient task processing in dynamic ITS environments, particularly as Open RAN architectures become more prevalent in 5G and beyond networks.