Computing technologies: platforms, processors, and controllers Chapter | 3 39ECUs running on a vehicle to a complete ecosystem comprising all car modules
and ECUs or even a fleet of cars that navigate in groups.
3.3.5 Vehicle software requirements
Inside future AD platforms, both AUTOSAR platforms work closely together
taking care for basic software services and communication on eHCP (Fig. 3.1).
Several prototyping environments have been developed for supporting the
implementation and testing of automated driving scenarios, such as the ROS 2.0
(Robot Operating System) from Open Source Robotics Foundation, or the EB
robinos, a software development framework for AD that employs open inter-
faces and follows an open specification). Such environments, take the existing
automotive eHPC environment for software development to another level and
define the framework for future environments for the design, implementation,
and testing of automated driving solutions.
Several applications that relate to data fusion, computer vision, or percep-
tion tasks, in general, must take advantage of various libraries that have been
developed so far and the future software development must contribute to this
direction. The optimized application libraries can be used for sensor data fusion,
vehicle perception, and analysis of what-if scenarios. The processing of LIDAR
sensor input requires data representation, scene segmentation, and object iden-
tification, which have to use implementations of the Point Cloud Library and
Fast Library for Approximate Nearest Neighbors libraries. Similarly, high-
performance computer vision can be easily achieved by processing the camera
input using the functions that are already available in the OpenCV library. The
same libraries can be found in the OpenVX IDE, which can be used for devel-
oping similar solutions. Sensor fusion is supported by the high-performance
functions of OpenCL, or by the linear algebra libraries BLAS/BLIS and Eigen.
All the above are slowly been integrated as libraries to popular deep learning
FIGURE 3.1 AD architecture with AUTOSAR platform.