In this paper, we are interested in the design of an autonomous and modular self-reconfigurable robot having self-assembly capability and deterministic behavior. The ability of a modular robot to meet its mission strongly depends on the artificial intelligence software and on the underlying hardware and software architecture. The artificial intelligence software of a robot is mapped into several elementary tasks with different real-time constraints. We propose in this paper a real-time analysis taking into account kernel overheads for the validation of the real-time behavior of an artificial intelligence software. We study the OSEK operating system that requires few hardware resources and is cost effective. The overheads are due to the context switching mechanism which activates, terminates, and reschedules tasks, and to the periodic timer used to create the time base which is necessary for the periodic tasks model. We show how to take into account those overheads in the feasibility conditions. We compare the theoretical worst case response time obtained with kernel overheads to the response time obtained on a a task set, on a real robot, based on the event driven OSEK implementation. We show that the kernel overheads cannot be neglected and that the theoretical results are valid and can be used to ensure a deterministic behavior of the robot.