Uncrewed Systems Technology 048 | Kodiak Driver | 5G focus | Tiburon USV | Skypersonic Skycopter and Skyrover | CES 2023 | Limbach L 2400 DX and L 550 EFG | NXInnovation NX 100 Enviro | Solar power focus | Protegimus Protection

Rather than using general-purpose high-performance vision processors for autonomous vehicles, Recogni and Renesas have developed an embedded platform, which can deliver up to 2000 TOPS (writes Nick Flaherty). The Phoenix ECU system combines Recogni’s Scorpio AI inference processor and Renesas Electronics’ R-Car V4H ADAS/AD SoC. Both chips are built on a 7 nm CMOS process technology. The Scorpio chip’s architecture provides a latency of less than 10 ms from the last pixel out to perception results. That provides ample reaction time for the car to navigate safely, along with an object detection accuracy range of up to 300 m in real time under various road and environmental conditions. “Vision is fundamental to accurate perception processing, and essential to autonomous driving platforms,” said RK Anand, founder and chief product officer at Recogni. “From the beginning, we took an approach of processing high-resolution images at the edge to achieve near- perfect object detection and classification, and enable autonomous driving stacks to make for better driving decisions. “The Scorpio can process multiple 8 MP streams at 30 frames per second in less than 10 ms using only 25 W. That’s a performance order of magnitude greater than anything else on the market and will, we believe, help to accelerate autonomous driving become a reality.” The chip can process 1000 TOPS in the 25Wpower envelope, and two chips can be used in the platformwith a V4Hcontroller for L3 and L4 autonomous applications. The V4H is also used for machine learning inference with dedicated deep learning and computer vision processing blocks with an overall performance of 34 TOPS. The blocks work with four ARM Cortex-A76 general-purpose processor cores running at 1.8 GHz. Three ARM Cortex-R52 cores run in lockstep at 1.4 GHz for a total of 9 kDMIPS to support ASIL D real-time operation and eliminate the need for external microcontrollers. The lockstep allows the processing functions to be checked by each processor. Renesas also provides a dedicated power solution for V4H based around the RAA271041 pre-regulator and the RAA271005 PMIC. This enables a highly reliable power supply for the R-Car V4H and peripheral memories from the vehicle battery’s 12 V supply. These features enable low-power operation while targeting ASIL D compliance for systematic and random hardware faults. Recogni has an open software platform for the Scorpio with an L2+ capable AI vision perception stack and a variety of object classification and detection features. This includes vehicles, traffic signs and traffic light detection and lane detection, but developers can run their own perception and driving functions software on the chip. Driverless vehicles Fast, low-power vision 16 February/March 2023 | Uncrewed Systems Technology The Phoenix system can process multiple 8 MP streams at 30 fps in less than 10 ms We process images at the edge to achieve near- perfect object detection and classification, to make for better driving decisions