Taming Edge Computing for Hard Real-Time Advanced Control of Mechatronic Systems

In novel mechatronics enabled by smart structures and materials, servomechanisms are becoming increasingly complex, requiring computationally intensive advanced control algorithms and diagnostic tools to fully exploit their potential. This calls for a significant increase in computational power while guaranteeing hard real-time features. In this work, we propose to address such an issue by adopting recently-emerged edge-computing solutions exploiting low-cost multicore that combine microcontrollers and microprocessors to boost the computational capability. However, such platforms are usually endowed with nonreal-time software infrastructure, assigning a dominant role to microprocessors and leading to large overheads and unpredictability. Therefore, to tame them for hard real time, we first lighten the infrastructure to enable one or more microprocessors to handle computations with minimal overhead and jitter. Then, we designate a microcontroller as the platform master of time and tasks, off-loading the heavy computations to the "relieved" microprocessor cores, acting now as computational slaves. We assess the potentials of this approach with a basic test using a demanding control algorithm as a benchmark, choosing the STM32MP157 as the reference platform and using the Jailhouse hypervisor to adapt one of its microprocessor cores for hard real-time tasks.