Vehicle task offloading strategy based on DRL in communication and sensing scenarios

Abstract

Integrated sensing and communications (ISAC) is one of the key technologies of 6th generation mobile communication (6G). In vehicle edge computing (VEC) networks, ISAC can assist vehicles in real-time sensing of the surrounding environment and establish a robust shared sensing network. However, the existing VEC offloading method cannot handle the large amount of computational data generated by joint sensing computation while meeting the task delay and energy consumption requirements. Therefore, in order to solve the above problems, we propose a Vehicle-Assisted Fusion Perception Offloading in vehicular edge computing network (VAFPO) scheme, which utilizes the perception characteristics of ISAC, and the task vehicle adaptively uploads data or instructions to the auxiliary node. Adaptive uploading of data or instructions to the auxiliary node, the auxiliary node uses its own perceived data to calculate when transmitting instructions, at the same time, the scheme constructs a delay and energy consumption priority factor according to the task characteristics to get the delay and energy consumption priority, in order to minimize the total system overhead. Finally, we propose a State Normalized DDPG for Adaptive Offloading (SNDAO) algorithm, which increases the convergence speed of the algorithm to obtain the optimal offloading decision by normalizing the states obtained from the agent's interaction with the environment, and at the same time adopting an actor-critic network with an experience playback mechanism. Simulation results demonstrate the effectiveness of the SNDAO algorithm and verify that the VAFPO scheme is significantly better than the current benchmark scheme.