A collaborative computation and dependency-aware task offloading method for vehicular edge computing: a reinforcement learning approach

Liu, Guozhi; Dai, Fei; Huang, Bi; Qiang, Zhenping; Wang, Shuai; Li, Lecheng

doi:10.1186/s13677-022-00340-3

Journal of Cloud Computing

Advances, Systems and Applications

Table 1 A side-by-side comparison of offloading strategies

From: A collaborative computation and dependency-aware task offloading method for vehicular edge computing: a reinforcement learning approach

Reference	Strategy	Applied metrics	Collaborative way	Advantages	Weakness
[17]	deep deterministic policy gradient (TD3)	Delay, Energy	Vehicle-Edge	Appropriate computation model	Not considering the resources of cloud services
[21]	Machine Learning	Delay, Cost	Vehicle-Edge	Considering the mobility of the vehicles	Not considering the resources of cloud services
[22]	Deep Deterministic Policy Gradient (DDPG)	Delay	Vehicle-Edge	Considering V2V and V2I	High complexity
[23]	A policy gradient deep reinforcement learning	Delay, Energy	Vehicle-Edge	Acceptable system model	Not considering the resources of cloud services
[14]	Takagi-Sugeno fuzzy neural network and game theory	Delay, Energy	Vehicle-Edge-Cloud	Traffic flow prediction combined with task offloading	High complexity
[7]	Deep Q-Network (DQN)	Delay	Vehicle-Edge-Cloud	Acceptable complexity	Not considering task dependencies
[24]	Machine Learning	Profit	Vehicle-Edge-Cloud	Edge and cloud share resources in the form of wholesale and buyback	High dimensional state space
[25]	A Greedy Based	Delay	Vehicle-Edge-Cloud	Considering task dependencies	Single offloading
[26]	Heuristic Algorithm	Energy, Cost	Vehicle-Edge-Cloud	Considering task dependencies	High complexity
[27]	Q-learning	Delay	Vehicle-Edge-Cloud	Acceptable system model	High dimensional state space
[28]	Actor-Critic Mechanism	Delay, Energy	Vehicle-Edge-Cloud	Considering task dependencies	Not considering continuous action space
[29]	Deep Meta Reinforcement Learning-based Offloading (DMRO)	Cost	Vehicle-Edge-Cloud	Fine-grained offloading	Not considering continuous action space
[30]	Deep Q-Network (DQN)	QoE	Vehicle-Edge	Solving for high dimensional state space	Model limitations
[31]	Deep Q-Network (DQN)	Cost	Vehicle-Edge	Appropriate mathematical proof	Not considering continuous action space

Back to article page