Lightweight explainable energy-aware deep learning-based task scheduling for microcontroller embedded systems

Engineering notes

Key topics: autonomous driving, deployment, control. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

The development of applications for the Internet of Things (IoT) and edge computing has driven the demand for efficient task scheduling mechanisms that enable devices to operate under strict energy, memory, and computational constraints in embedded systems, which are typically running on microcontrollers. Conventional, predictable scheduling strategies like Earliest Deadline First (EDF) and Rate Monotonic (RM) are nonadaptive to changing loads and do not yield good performance under dynamic loads; numerous artificial intelligence-based schedulers are noninterleigible and incur significant computational overhead. To address these problems, this work proposes an energy-aware task scheduling algorithm with an explainable deep learning model, MicroEdgeXAI, for resource-constrained embedded systems. The framework combines analytical energy modelling with a lightweight GRU-based temporal scheduler that learns task execution patterns, system states, and energy characteristics. It promotes the dynamic selection of tasks with the lowest energy, latency, and deadline impacts while servicing tasks with high energy costs. SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations) are incorporated to explain scheduling decisions, enhancing transparency. Additionally, the proposed MicroSchedNet scheduler adopts an efficient architecture with around 12.4 K trainable parameters, which requires only 49.6 KB of Flash memory, 11.8 KB of SRAM, and has a potential inference latency of 0.8–1.4ms on a representative ARM Cortex-M4-based platform, making it suitable for practical deployment in embedded systems. Results of experimental testing on Embench, BEEBS, and TACLeBench workloads confirm that MicroEdgeXAI can reduce energy consumption by up to 30% and deadline misses by 50–65% compared with conventional scheduling methods, while improving latency and CPU utilisation. The findings show the potential of MicroEdgeXAI as an effective, adaptive, and trustworthy approach to intelligent task scheduling for future IoT/edge computing infrastructures.

Links and sources

• PDF from original source

Comments