Implementation of an embedded system on a TS7800 board for robot control

Growing Functional Modules (GFM) learning based controllers need to be experimented on real robots. In 2009, looking to develop a flexible and generic embedded interface for such robots, we decided to use a TS-7800 single board computer (SBC) with a Debian Linux operating system. Despite the many advantages of this board, implementing the embedded system has been a complex task. This paper describes the implementation of protocols through the TS-7800 different ports (RS232, TCP/IP, USB, analog and digital pins) as well as the connection of external boards (TS-ADC24, TS-DIO64, SSC-32 and LCD display). This implementation was required to connect a large range of actuators, sensors and other peripherals. Furthermore, the architecture of the embedded system is exposed in detail, including topics such as the XML configuration file that specifies the peripherals connected to the SBC, the concept of virtual sensors, the implementation of parallelism and the embedded system interface launcher. Technical aspects such as the optimization of video capture and processing are detailed because their execution required specific compilers versions, EABI emulation and extra libraries (openCV libjpg and libpngand libv4l). The final embedded system was implemented in a humanoid robot and connected to the GFM controller in charge of developing its equilibrium subsystem.