Vehicular Computing Power Networks for IoT-Driven Edge Intelligence: MA-DDPG-Based Robust Task Offloading and Resource Allocation

The deep integration of Internet of Things (IoT) and vehicular networks demands ultrareliable, low-latency computing paradigms to support emerging applications like autonomous driving and smart traffic management. Existing mobile-edge computing (MEC) frameworks, however, struggle with dynamic resource heterogeneity, intermittent connectivity, and inefficient coordination among distributed nodes. To address these challenges, this article proposes vehicular computing power networks (VCPNs), an IoT-driven edge intelligence framework that orchestrates computational resources from mobile user equipments (MUEs), connected vehicles, and edge servers. We formulate a joint optimization problem to minimize end-to-end task latency by finding optimal task offloading decisions and resource allocation (e.g., CPU and bandwidth) policies under time-varying IoT channel conditions and node mobility. To enable decentralized coordination in IoT environment, we model the problem as a multiagent Markov decision process (MDP) and propose a multiagent deep deterministic policy gradient (MA-DDPG) algorithm in which agents (MUEs, vehicles, servers) collaboratively learn policies to optimize task scheduling and resource sharing. Furthermore, we design a robust MA-DDPG variant with error-resilient experience replay and channel-adaptive reward mechanisms to ensure reliable training under packet loss and unstable connectivity. Numerical results demonstrate that VCPN reduces average task latency and improves energy efficiency compared to federated MEC baselines. The proposed MA-DDPG algorithm achieves convergence stability in high-mobility scenarios, outperforming conventional deep reinforcement learning methods.