Worst-Case Execution Time Analysis of Real-Time Robotic Algorithms Using Reinforcement Learning

This study investigates the application of Reinforce-ment Learning (RL) for estimating the Worst-Case Execution Time (WCET) of real-time robotic algorithms, crucial for ensuring reliable robotic navigation and interaction. The research eval-uates these computational techniques through two industrial case studies: (1) collision detection among two mobile cylinders and a stationary box; and (2) a six-degree-of-freedom (6-DOF) robotic model executing complex tasks. By leveraging RL, the study explores extensive input spaces to identify scenarios approaching the WCET within a controlled simulation environment. Through an experimental evaluation, we compare RL against Genetic Algorithms (GA) and random search approaches. The results demonstrate that RL outperforms both GA and random search. This research highlights the potential of RL to provide more accurate and reliable W CET estimations, thereby enhancing the safety and efficiency of real-time robotic systems.