ABSTRACT
This work presents a parallel real-time computing system for controlling and managing robots’ sensing, processing and actuating input/output functionalities. The proposed software architecture approach is purposed to be used in robotic platforms deploying computers that are administrated by the Linux operating system. The three main robotic primitives: sense, plan and act are processed and scheduled as multiple threads, and dynamically allocated in multi-core. The real-time architecture is systematically modeled disclosing its formulation scheme considering: dynamic threads priority, runtime threading workloads estimation, and degree of threads complexity with respect to the computing burden delays. The approach is validated through experimental multi-core computer execution. Thus, as a case of exemplification, a robotic arm model with multiple sensors, actuators and an intensive recursive model-based kinematic controller are used. The approach shows a balanced administration of computational resources use and the convenience of real-time parallelism deployed in cyber-physical robotic tasks.
