Mobility-aware predictive task offloading using deep reinforcement learning for future 6G-inspired vehicular networks

Engineering notes

In comparison to current reactive and reinforcement learning-based baselines, the proposed PODRL framework achieves up to 40% lower end-to-end latency ( p < 0.001) and up to 45% lower vehicular energy consumption ( p < 0.001) while maintaining high deadline satisfaction rates, according to extensive simulation-based evaluations conducted using realistic vehicular mobility traces and forward-looking 6G-inspired channel models.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Reliable communication and ultra-low latency data processing are essential requirements for emerging Internet of Vehicles (IoV) applications such as cooperative autonomous driving, augmented reality navigation, and real-time safety monitoring. Although mobile edge computing has been widely adopted to reduce computational delays by offloading tasks from vehicles to nearby roadside infrastructure, the highly dynamic mobility of vehicles and rapidly changing wireless channels pose significant challenges for reliable task transmission and execution. Traditional offloading strategies typically rely on instantaneous network conditions and reactive decision mechanisms, which often lead to unstable communication links, frequent handovers, and increased energy consumption. This paper proposes a mobility-aware predictive task offloading framework designed for reliable edge communications in future 6G-inspired vehicular network environments. The proposed framework integrates trajectory prediction, wireless channel forecasting, and Deep Reinforcement Learning to proactively determine optimal task transmission and execution locations. Specifically, a Predictive Offloading Deep Reinforcement Learning (PODRL) architecture is developed, where predicted vehicle mobility and channel conditions are incorporated into an augmented Markov Decision Process (MDP) for intelligent decision making. In addition, predictive transmission path selection and proactive task migration mechanisms are introduced to ensure stable communication during high-speed vehicular mobility. In comparison to current reactive and reinforcement learning-based baselines, the proposed PODRL framework achieves up to 40% lower end-to-end latency ( p < 0.001) and up to 45% lower vehicular energy consumption ( p < 0.001) while maintaining high deadline satisfaction rates, according to extensive simulation-based evaluations conducted using realistic vehicular mobility traces and forward-looking 6G-inspired channel models. All evaluations are based on simulation-based 6G-inspired communication models and planned IMT-2030 capabilities, rather than on deployed commercial 6G infrastructure.

Links and sources

• PDF from original source

Comments