A Real-Time Embedded Kernel for Nonvisual Robotic Sensors

We describe a novel and flexible real-time kernel, called Yartek, with low overhead and low footprint suitable for embedded systems. The motivation of this development was due to the difficulty to find a free and stable real-time kernel suitable for our necessities. Yartek has been developed on a Coldfire microcontroller. The real-time periodic tasks are scheduled using nonpreemptive EDF, while the non-real-time tasks are scheduled in background. It uses a deferred interrupt mechanism, and memory is managed using contiguous allocation. Also, a design methodology was devised for the nonpreemptive EDF scheduling, based on the computation of bounds on the periodic task durations. Finally, we describe a case study, namely, an embedded system developed with Yartek for the implementation of nonvisual perception for mobile robots. This application has been designed using the proposed design methodology.