Optimization of RTOS Scheduling for Latency-Sensitive Sensor Fusion Systems

Abstract

Efficient sensor fusion is essential when it comes to perception and actuation within context of latency-sensitive applications in the real-time embedded systems, specifically in autonomous vehicles, unmanned aerial vehicles (UAV) and industrial robotics. These systems demand very strong timing assurances and deterministic task chronology so that the data collected by multiple sensors, frequently out of synch, can be latency-minted and integrated. The problem with conventional fixed-priority preemptive scheduling algorithms, such as Fixed-Priority Preemptive Scheduling (FPPS) and Earliest Deadline First (EDF) is that they do not readjust dynamically to the changing workloads caused by real-world sensor variability, and instead cause undesirable jitter, deadline violations of tasks, and degraded system reliability. This paper puts forth a new hybrid scheduling optimization scheme to Real-Time Operating System (RTOS), that is candidate-specific to optimize the temporal aspect of sensor fusion pipelines. This would combine a static Deadline-Monotonic ( DM ) base scheduler with dynamic Slack-Aware Priority Adjuster (SAPA ) in which the priorities of tasks are re-allocated at run time depending on real-time feedback as supplied by the execution profiler, sensor burst detection and slack time calculation. Implementing and testing the suggested strategy using FreeRTOS on an ARM cortex M4-based embedded device may test it under a variety of sensor loads. In experimentation we obtain a 42 percent reduction in average fusion latency, 30 percent better jitter consistency, and a higher utility in CPU utilization than baseline EDF and FPPS schedulers. Moreover, the system would be robust in face of bursty sensor events without compromising quality fusion results and starvation of background processes. Besides offering a scalable and very lightweight extension of basic RTOS kernels, the work also establishes the basis of intelligent scheduling techniques that will become a key in future safety-critical and mission-critical designs and systems where such a real-time sensor fusion is essential. The framework is transportable, simple to merge with current FreeRTOS projects and is flexible to heterogeneous tasking compositions, which renders the framework rightful to embedded AI tasks, robotics, and smart edge operations that work under hard real-time requirements.