Despite the significant improvements achieved until now in controlling humanoid robots, they are still a long way from reliably exhibiting human-like behaviors. Various components that formulate humanoids contribute to improving their behaviors. While increasing the physical limitation contributes to dynamic motions, the control architecture is the key to managing those motions and determining robot capabilities to improve its behaviors. This work aims to develop a control architecture emulating the functionality of the human nervous system. Classical architectures dealt with sensorimotor cycles but without intelligence distribution. Whether centralized, where all components are connected to a central unit, or decentralized, where the distributed units are used as an interface between the I/Os and the master controller with no ability to make a decision. The proposed solution is a distributed real-time control architecture with ROS. The joint controllers have the intelligence to make decisions, dominate their actuators, and publish their state. The real- time capabilities are ensured in the master controller by using a Preempt-RT kernel beside OROCOS middleware to operate the real-time tasks. And in the customized joint controllers by FreeRTOS firmware. The experimental validations were performed on our electro-hydraulic humanoid robot (HYDROïD). The results demonstrate 50% advancements in the update rate compared to other humanoids and 30% in the latency of the master processor and the control tasks. The proposed architecture gives the possibility to create and test the systems with distributed artificial intelligence.