KI Radar

High-resolution radar system with integrated AI-supported data processing for cooperative autonomous driving

Autonomous driving needs to be safe and reliable. To achieve this, we need trustworthy new electronic components that are not only highly energy-efficient and high-performing, but still guarantee basic functionality even in the event of a fault. In conjunction with artificial intelligence (AI) methods, they build the technological basis that allows the vehicle to reacht appropriately to all driving situations.

The increased resolution of radar sensors through AI-based evaluation and coupling of the sensors and the new antenna structure are significant advances for increasing the effectiveness and safety of autonomous vehicles. The project thus made an essential contribution to the further development of autonomous driving.

Within the subprojetct "Hardware and Software Platform for Edge Computing", Creonic participated in the following work packages:

Development and validation of the AI algorithms
Design of the interface to the AI and integration on the FPGA
Development of the Cognitive Edge
Digital implementation of the transmission system

Project Partners

KSG GmbH, Gornsdorf | Creonic GmbH, Kaiserslautern | Universität Bielefeld | Fraunhofer-Institut für Zuverlässigkeit und Mikro-integration IZM, Berlin

08/2019 - 05/2022

Updates

Implementation of wireless communication on digital level completed

Analog front end
Use of an adaptive system, which can be dynamically adjusted between data rate and reliability
The design of the digital communication system allows to achieve a data rate of 1Gbit/s and still be functional even with poor link quality of -2dB SNR
System supports different code rates and modulations (ModCods), achieving high spectral efficiency under different conditions (200 different gradations possible)

Implementation of wired communication completed

Implementing the Tensorflow-Lite framework on the Xilinx ZCU106 to achieve seamless and accelerated integration and validation,
Using a 10G SFP+ interface as it is flexible and readily available.
Lab setup for demonstration purposes: connecting two FPGA boards as edge nodes, transmitting the images of a camera as secondary sensor and the FFT of the radar sensors as primary data.
Ensuring scaling: connection of the boards was tested via a switch.

Seamless and accelerated validation

Implementation of the Tensorflow-Lite framework on the Xilinx ZCU106.
A software simulation can be performed on the target hardware, as well as a direct hardware implementation.

Model format selection

ONNX Format
Tensorflow

Operating system selection

Linux distribution Ubuntu
Integration of the browser-based development environment Jupyter, to enable easy programmability
Linking to the hardware, so that software programming can be done largely without hardware knowledge

Problem Definition

In the concept of edge computing, a reliable and fast connection is a critical component here: multiple participants (edge nodes) communicate with each other, performing different tasks, but at the same time complementing and replacing each other in case of failure. Within a vehicle, wired communication offers the highest reliability. Due to the potentially large amounts of data generated by AI, a sufficiently fast solution must be considered. At the same time, for safety reasons, the results must be available in a short time so that another system or the driver can react. In addition to a high bandwidth for the data, latency should therefore be low.

To increase the field of view of a vehicle, information and data from other vehicles are used. Only wireless communication and a system that can reliably transmit high volumes of data over a short distance can be considered here.