Mentor/Siemens aims for Level 5 with autonomous driving platform

Mentor/Siemens aims for Level 5 with autonomous driving platform

Technology News |
Mentor, now part of Siemens, has consolidated all of its activities that relate to the automotive sector in one division; and that division has, in a new departure for the company, announced a complete platform of compute hardware and function-related software, for autonomous driving systems, and for ADAS functions on the way towards full autonomy.
By Graham Prophet


A key premise of the system’s philosophy is that all sensor data carries useful information and therefore should not be reduced or pre-processed, or be subject to delay; the platform is a centralised resource that will accept ‘raw’ data from all available sources, and carry out the necessary sensor fusion at that level, so that system decision-making has the maximum possible information.


That fusion function will be executed on hardware based on a Xilinx all-programmable FPGA with four ARM Cortex-A53 cores. The other major block in the system is a full-custom SoC, in which higher-level routines such as artificial intelligence, machine learning and neural net computing will reside. Software, quality-assured to established automotive standards, will be available for key functions (image processing/feature recognition, for example). Hence, Mentor’s summary of, “Centralized raw data fusion and direct real-time sensing”.

The hardware par of the equation, Mentor emphasises, is not only a development vehicle but is capable of being taken through to full production; one of the constraints is that it operates within a 100W power budget. It is presented as having the capability to be developed to full Level-5 autonomy, but being applicable to Level 3 or 4; or to Level 2, in a short timescale.


DRS360, Mentor says, is a comprehensive automated driving solution … that captures, fuses and utilizes raw data in real time from a wide range of sensing modalities, including radar, LIDAR, vision and other sensors. Data from other cars, or from the infrastructure – V2V or V2X – is treated as an additional sensor data flow, as is data from mapping. Using FPGA to pull in raw sensor data results in latency reduction, sensing accuracy and overall system efficiency, that will be required for SAE Level 5 autonomous vehicles.


The platform employs what Mentor terms “raw data sensors”, which are not restricted by the power, cost and size penalties of microcontrollers and related processing in the sensor nodes, [developed] in partnership with leading sensor suppliers, including Sony for image sensing. Eliminating pre-processing microcontrollers from all system sensor nodes enables a broad array of advantages, including real-time performance, significant reductions in system cost and complexity, and access to all captured sensor data for the highest resolution model of the vehicle’s environment and driving conditions. Mentor indicates that a 1:1 correspondence of hardware to data stream and to function is maintained – the scheme does not virtualise processing on its compute resources.


The platform’s data transport architecture lowers system latency by minimizing physical bus structures, hardware interfaces and complex, time-triggered Ethernet backbones, enabling “situation-adaptive redundancy and dynamic resolution”, by using centralized, unfiltered sensor data to ensure enhanced accuracy and reliability. The solution’s optimized signal processing software, advanced algorithms, and compute-optimized neural networks for machine learning run on a seamlessly integrated, automotive-grade platform.


The DRS360 platform, Mentor reiterates, is engineered for production to meet the safety, cost, power, thermal and emissions requirements for deployment in ISO 26262 ASIL D-compliant systems. DRS360 exploits the flexibility and signal processing efficiency of FPGAs, in the shape of a Xilinx Zynq UltraScale+ MPSoC device in the first generation, while accommodating SoCs and safety controllers based on either X86- or ARM-based architectures.






Linked Articles
eeNews Embedded