STEPS: Semantic-Contract-Guided Scheduling for LLM-Assisted Natural-Language-Driven Edge AI Services
2026-06-08 • Networking and Internet Architecture
Networking and Internet Architecture
AI summaryⓘ
The authors created a new system called STEPS to help schedule AI services on edge devices based on natural language user requests rather than strict numbers like latency or cost. STEPS uses large language models to understand user intent and turn it into clear contracts that guide how resources are allocated and tasks are assigned. Their approach models the problem as a special game that finds stable, efficient ways to run services while adapting to changes and uncertainty. Tests showed STEPS better meets user expectations and adapts well even when requests are unclear or conditions change.
edge computingAI service schedulinglarge language modelssemantic parsingpotential gameNash equilibriumresource allocationnatural language processingsemantic contractsadaptive systems
Authors
Houyi Qi, Minghui Liwang, Xianbin Wang, Seyyedali Hosseinalipour
Abstract
Networked AI services are increasingly delivered through edge infrastructures to support latency-sensitive applications. Edge scheduling is critical for deciding where and how AI services are executed under limited communication and computing resources. Existing frameworks usually assume that requirements are given as numerical constraints, such as latency bounds, energy budgets, or cost limits. In practice, users often express expectations through ambiguous natural language, creating a gap between user intent and resource constrained scheduling. To bridge this gap, we propose semantic-contract-guided edge potential scheduling (STEPS), a natural language driven scheduling framework for LLM assisted edge AI services. STEPS introduces semantic contracts as executable interfaces between user-side semantics and edge-side decision making. An LLM assisted semantic parser extracts service levels and confidence scores, which are converted into service preferences, fulfillment bounds, and semantic uncertainty. Based on these contracts, STEPS formulates edge scheduling as a contract-guided potential game that jointly determines execution-node selection, computing-resource provisioning, and bandwidth allocation. It also builds feedback signals from semantic request drift, fulfillment drift, fulfillment pressure, and admission pressure to adjust semantic admission, contract conservativeness, and edge coordination. We characterize the exact potential game structure, establish pure strategy Nash equilibrium existence, and prove convergence and stability properties. Experiments show that STEPS improves semantic contract fulfillment, reduces contract guided service loss, and maintains robust adaptation under ambiguous requests and non-stationary edge environments.