Ambarella demonstrates its fully autonomous vehicle on Silicon Valley roads
Ambarella, Inc., a leading developer of low-power, HD and Ultra HD video processing semiconductors, announced it is demonstrating its fully autonomous EVA (Embedded Vehicle Autonomy) vehicle on Silicon Valley roads to industry analysts and customers. EVA has been trained to deal with the various traffic scenarios presented by Silicon Valley’s challenging urban environment. The fully autonomous car combines software and algorithms developed over 20 years of autonomous vehicle research with Ambarella’s low-power CV1 embedded computer vision processors based on its CVflow architecture. EVA’s high-resolution stereovision cameras deliver the 360-degree short- and long-distance viewing capability required for advanced perception and precise self-location. EVA includes sensor fusion of the vision information with radar and map data to provide the information necessary for path planning and merging maneuvers without the need for additional LiDAR systems.
“High-resolution 8-megapixel stereovision combined with superior perception in challenging lighting conditions allows EVA to “see” its surroundings with much higher reliability than was previously possible,” said Professor Alberto Broggi, general manager of Ambarella Italy. “Moving to an implementation based on dedicated Ambarella CVflow processors brings us much closer to making self-driving cars a practical reality.”
EVA’s CV1-based stereovision cameras provide a perception range of over 150 m for stereo obstacle detection and over 180 m for monocular classification. Stereovision processing enables detection of generic obstacles without training, allowing more robust decisions to be made. EVA also uses stereovision to recognize visual landmarks and uses HD map information for high precision localization, even when the GPS signal is weak or not available, in dense urban locations, for example. EVA features include automatic calibration, stereo generic obstacle detection, terrain modeling, traffic light detection, 3D free space detection, lane detection, curb and barrier detection, and CNN classification for vehicle, pedestrian, and bicycle/motorcycle.
Ambarella also announced its next-generation CV2 computer vision processor, which will provide up to 20 times the computer vision performance of CV1 in a fully integrated SoC, delivering higher perception accuracy and further reducing the total number of chips required for a fully autonomous vehicle.
The CV2 4K Computer Vision SoC with CVflow™ architecture and stereovision targets advanced automotive, IP security, drone, and robotic applications. In automotive applications such as advanced driver assistance systems (ADAS) and self-driving systems, its ability to run multiple algorithms simultaneously delivers higher perception accuracy and reduces the total number of chips required. Fabricated in advanced 10-nm process technology, CV2 offers extremely low power consumption.
“With CV2 we have dramatically increased our computer vision performance and combined it with full SoC functionality,” said Fermi Wang, CEO of Ambarella. “As the highest performance member of our new CVflow family, CV2 delivers both the deep neural network and stereovision processing required for the most advanced automotive and security cameras.”
The CV2’s CVflow architecture provides computer vision processing with up to 4K or 8-MP resolution, to enable object recognition and perception over long distances and with high accuracy. Its stereovision processing provides the ability to detect generic objects without training in ADAS and autonomous vehicle applications. Advanced image processing with HDR (high dynamic range) processing delivers outstanding imaging even in low light and from high-contrast scenes. Its efficient 4Kp60 AVC and HEVC video encoding supports the addition of video recording to automotive ADAS and self-driving systems and enables the design of both multi-stream and multi-imager IP security cameras. CV2 includes a suite of advanced security features to prevent hacking, including secure boot, TrustZone™, and I/O virtualization.
A complete set of tools is provided to help customers easily port their own neural networks onto the CV2 SoC. This includes compiler, debugger, and support for industry standard training tools including Caffe™ and TensorFlow™, with extensive guidelines for CNN (convolutional neural network) performance optimizations.
For more information about Ambarella, visit www.ambarella.com.