OEM & Lieferant Ausgabe 2/2019 - OEM & Supplier 2/2019 by VEK Publishing

134 Engineering Partner Integration of ROS on a PX2 platform for autonomous driving By Markus Haab, Account Manager E/E System, Software Development & Test, ServiceXpert GmbH The demands placed on driver assistance systems for vehicles have grown considerably and continue to rise, especially as a result of autonomous driving functions. This is also increasing the requirements for powerful hardware and software. As the capabilities of autonomous systems are constantly being enhanced and im- proved, the need for expertise in order to efficiently use the powerful hardware and software is growing to the same extent. ServiceXpert Gesellschaft für Service-Infor- mationssysteme mbH functions as an engi- neering partner to large commercial vehicle OEMs on current topics and is involved in developing innovative solutions with state- of-the-art technologies. In an internal R&D project, ServiceXpert engineers investigated the possibilities of using software frameworks such as Robot Operating System (ROS) with the NVIDIA Drive PX2 platform. The focus of such R&D projects is always on gaining expe- rience in handling components and research- ing the options with regard to developing GPU-enabled solutions, hardware integration with real-time capabilities and ADAS expan- sion in general. The research project began with an examina- tion of the key software frameworks currently being used to develop driver assistance sys- tems (ADAS). The engineers then extrapolated how these software frameworks could meet the intensive computational requirements of algorithms based on artificial intelligence us- ing an embedded high-end GPU. The proper- ties that matter in this context – performance, data storage, licensing, etc. – were assessed to enable good use of the findings within cus- tomer projects. In order to review the applicability of ROS in ServiceXpert projects, the framework was compared with the Automotive Data and Time-Triggered Framework (ADTF) and a sample program was developed to show the development steps in ROS with regard to inte- grating a Basler camera. Finally, a USB camera was connected to demonstrate the switch to another camera. The multi-camera interface proved itself easy to handle due to the pres- ence of a previously developed ROS package interface of sensor drivers for the ROS mes- sage type. Finally, ROS was integrated together with key computer-vision-based libraries (OpenCV, CUDA) on the PX2 platform and there was post-processing of the image frames based on edge detection. For the ServiceXpert engineers, ROS, as a non-commercial product, is characterised by enormous flexibility for the provision and integration of ADAS-based improvements, especially through the inherent open source licensing. The next version, ROS2, is expected to fix the current bugs, starting with the pos- sibility of direct DDS support, which is not cur- rently available even in the ADTF. The ServiceXpert team already implemented some of the future improvements that can be made to the existing ROS code skeleton (C++) in the research project. The ROS package integrated into the project used open source dependencies (original BSD licence) so that further functions can be easily built on in the future. In addition, convolution forms the basis of most deep learning algorithms and there- fore enables autonomous driving in the first place. Thus, AI-based solutions can be eas- ily integrated using the graphics processor (GPU). In the current arrangement, the So- bel-filter-based convolutions use the CPU in the background. The camera is an essential component of au- tonomous driving functions. Thus, Service- Xpert has been able to improve route plan- ning and understanding of captured scenes by merging the camera’s sensors with other sensors, such as LIDAR and RADAR. Images: © ServiceXpert GmbH