Multi-Agent Reinforcement Learning for Dynamic TSN Scheduling in MEC
A recent paper on arXiv (2608.05346) introduces a multi-agent reinforcement learning (MARL) framework designed for scheduling in time-sensitive networking (TSN) within mobile edge computing (MEC) settings, with a focus on extended reality (XR) applications. This cross-type submission tackles the shortcomings of traditional TSN scheduling methods that depend on static optimization or centralized learning models tied to fixed traffic patterns. In real-world MEC scenarios, multiple XR traffic flows coexist with changing characteristics, leading to intricate inter-queue dependencies that existing schedulers overlook. By treating each TSN queue as an independent agent, the proposed framework facilitates adaptive and decentralized scheduling under fluctuating traffic conditions, aiming to enhance performance for latency-sensitive applications in dynamic MEC environments. The paper can be accessed via the provided arXiv link.
Key facts
- Paper arXiv:2608.05346 proposes MARL for TSN scheduling in MEC.
- Targets XR applications with stringent latency requirements.
- Existing solutions use static optimization or centralized learning.
- MEC environments have multiple co-located XR traffic flows.
- Traffic characteristics evolve over time, creating inter-queue dependencies.
- Framework models each TSN queue as an autonomous agent.
- Enables adaptive, decentralized scheduling coordination.
- Paper announced as cross-type on arXiv.
Entities
Institutions
- arXiv