摘要：In most advanced real-time control applications such as service robots, the tasks have different criticality, flexible timing constraints and variable execution time. For instance, autonomous service robots perform their activities in dynamic environments, share resources and have to cooperate to fulfill their objectives. These changing environmental conditions produce a variability of the system load. Firstly, the computational requirements of recognition tasks are variable and dependent on the number of objects perceived in scenes. Secondly, the application processes are executed at different frequencies with varying periods and deadlines that are dependent on robot speeds. To tackle these aspects a flexible real-time architecture is implemented in rt-linux in this paper. Moreover, the architecture permits to extract the slack time, enabled in the system due to load variability, and to invest it in improving the communication performances of the robots. Experimental evaluations of the architecture have been carried out with real autonomous robots.
