With its open collaboration model, Visteon’s new autonomous driving platform, DriveCore, is intended to speed up the development of autonomous driving technology.
“To meet the computing demands of Level 3-plus solutions, autonomous systems will need highly scalable levels of processing power and the ability to perform sensor fusion across multiple radar, camera and lidar sensors,” said Sachin Lawande, president and CEO of Visteon.
“DriveCore is the first open platform in the industry that offers highly scalable computing power and software to perform late sensor fusion to enable rapid development of these autonomous systems.”
The DriveCore platform includes hardware, in-vehicle middleware and PC-based software, so developers can create machine-learning algorithms for autonomous driving at Level 3 and above. Visteon says the solution offers a failsafe domain controller, with high scalability of computing power, so data from multiple sensors can be integrated.
The modular and scalable computing hardware Compute can deliver everything from 500 gigaflops to 20 teraflops of processing power (with existing Systems on Chip), independent of CPU type. It is designed to support processors from Nvidia, Freescale and Qualcomm, as well as future processor types.
The in-vehicle middleware, Runtime, provides the secure framework needed for applications and algorithms to communicate in real time, and enables sensor-independent sensor fusion, allowing for future sensor developments to be incorporated.
Finally, the PC-based development software Studio provides an open framework for sensor-based AI algorithm development. Third-party algorithms and real-life sensor data can be integrated easily. The solution enables the simulation, validation and benchmarking of algorithms for everything from object to lane detection.
At the introduction of DriveCore, Visteon announced four technology partners. DeepScale utilises deep learning to create an integrated environment model in real time, using any sensor combination, and then deep neural networks to add the perception required for automated driving.
Other partners include Steer, focused on an automated parking solution; StradVision, which uses machine learning algorithms to develop object detection and recognition software; and Automotive AI, which offers a graphical simulation environment with intelligent traffic and traffic scenarios, running with DriveCore Studio.