Table 1.

Comparative analysis of DRL-Based task scheduling approaches in edge/fog Computing.

Source	Formalisation of the service process	Parametric delay control	Behavioural adaptability	Queue as a controllable object	Depth of DRL integration	QoS adaptivity	Partially observable environment	Analytical tractability
22	MDP without queuing model	Delay treated as output, not controlled	Adapts to workload variation	Ignored	Superficial task assignment	Reward-based delay awareness	No	Empirical optimisation
23	Heuristic, lacks internal structure	Fixed reactive behaviour	Load balancing without phase control	Not modelled	Task offloading focus	Execution time minimisation	No	Simulation-based learning
24	External task distribution, no queuing	No control over delay parameters	Energy-focused, not behavioural	Implicit queue presence	High-level scheduling	Energy- and load-aware	No	Scenario-dependent suitability
25	General optimisation without structure	Delay unparameterised	Responds to state without internal dynamics	Not represented	No direct control mechanisms	Aggregate delay targeted	Yes	Approximate algorithmic control
26	Partial queue modelling	Jitter acknowledged, no direct control	Queue-aware dynamics	Limited queue control	DRL based on queue states	Precise QoS control	No	Partial analytical feasibility
Our approach	Fully formalised G/G/1 model	Delay shift as controllable variable	Adaptive to service state and load	Mathematically integrated queue	DRL fused with service mechanics	Delay and variation jointly optimised	Yes (queue and service phase)	Fully analytically tractable