A German research network with reputable research institutions, universities, IT companies and also companies from the automotive industry has developed a near-series fusion platform with open interfaces (Open Fusion Platform, abbreviated to OFP). It enables automobile manufacturers and their suppliers to cost-effectively integrate highly and fully automated driver assistance functions for the purposes of automated driving.
The OFP was developed by the network coordinator Hella together with the German Aerospace Centre, Elektrobit Automotive, Infineon Technologies AG, InnoSent, Hella Aglaia Mobile Vision, Reutlingen University, the RWTH Aachen Electromobility Centre, Streetscooter Research and TWT GmbH Science and Innovation. In addition, Continental and Nvidia supported the project in the role of associated partners.
The project, funded by the Federal Ministry of Education and Research with €4.4 million ($4.97 million), was successfully completed after just over three years with the integration of the first fully autonomous driving functions in three demonstration vehicles. In the scenario implemented, an electric vehicle drives fully automatically to a free charging station on a parking lot and positions itself over the charging plate. After the charging process has been completed, it automatically searches for a free parking space without a charging plate.
“For such highly or fully automated scenarios, so far only prototypes have existed that were not yet close to series production,” explains Dr. Michael Schilling, Project Manager for Pre-Development of Automated Driving at Hella and Network Coordinator for the OFP project. “Today there already are driver assistance systems in series production, such as traffic jam assistants that merge data from two sensors. But for fully automated driving, it is necessary for the vehicles to be able to perceive their entire surroundings. To achieve this, data from a multitude of sensors and cameras has to be merged in order to create a full model of the surroundings. Such an environmental model then shows the driving environment with the required amount of accuracy, allowing to implement a reliable driving function.”
One challenge here revolves around the interfaces between the individual sensors and the central electronic control unit, which have not been standardised yet. The interfaces of actual driver assistance systems are very function-specific and depend on the individual supplier or automobile manufacturer. This is exactly where the research project started its work.
Four cameras and eight 77 GHz radar sensors, covering 360° around the vehicle, served as input for the OFP. An additional Vehicle to X-Communication module (V2X module) also enables communication between vehicles and external infrastructure, such as a charging plate. The network partners have disclosed the interface descriptions of the individual components in a freely available “Interface Specification”. During the project, an ISO working group for the standardization of the sensor data interfaces was launched by the research group together with other leading automobile manufacturers and suppliers.
Upon completion of the project, an updated interface description is published, which also flows into the ongoing ISO process. This means that for the first time ever all automobile manufacturers and suppliers will have the opportunity of integrating their products quickly and easily into the fusion platform. With the complex environmental model, Hella Aglaia Mobile Vision has developed the central component of the OFP.
Using the OFP visualisation, developers can see how the vehicle perceives the entire environment and then on this basis can decide how the sensor data is to be merged. Whether it is a complex driver assistance function or fully automated driving function – all functions can be programmed in this way.
“This accelerates development work, which ultimately leads to a growing number of functions and thus to automated driving functions moving earlier into the mid-range vehicle segment,” says Dr. Schilling.
Work on the OFP will continue after the project. Crucial questions will be as to how the sensor data can be processed with machine learning in order to improve functions and further accelerate the development work. The parking lot scenario is also to be expanded to include urban driving situations and general driving at over 20 km/h. These scenarios require interaction with other sensors, e.g. LiDAR sensors. It is precisely in this area of multisensory data fusion that OFP will be able to exploit its full potential. Moreover, functional safety will play a major role in further development to ensure that all developed functionalities are and remain fail-safe.