Arriver presents the capabilities of its next generation software stack for Advanced Driver Assistance and (ADAS) and semi-autonomous driving. The presentation by Arriver’s Head of Technology, Salah Hadi at Qualcomm’s annual Automotive Summit outlines how Arriver will provide a leading software stack enabling Level 1 ADAS to Level 3 semi-autonomous driving in 2024.
Arriver is developing this next generation software stack to be an open, flexible and scalable solution which will allow automotive manufacturers and automotive suppliers to, based on a robust software platform, tailor and customize solutions to their specific needs based on their individual product development strategies.
The software stack consists of a fifth-generation perception stack which is built on almost two decades of development of vision-based perception. The fifth generation will provide high resolution, using 8-megapixel cameras, allowing for higher framerates and provide a wider field of view of 120 degrees, compared to today’s solutions.
The Arriver stack will provide full NCAP ratings and include features like Automatic Cruise Control, Pilot Assist, Glare Free High Beam, Lane Keep Assist, and Traffic Jam Assist supported by state of the art deep learning based detection functions like full 3D vehicle detection, pedestrians, bicycles, motor cycles as well as traffic signs, traffic lights. On top of that it will enable innovative solutions for determining drive free space and challenging police restricted and road work scenarios.
The development uses the next generation of Arriver‘s Convolutional Neural Network (CNN), which is up to 100 time more powerful than the current generation and will allow for the higher accuracy necessary for Level 3 applications and beyond, when the car takes over responsibility for the driver.
The Arriver software will include up-to-date fusion capabilities, catering for simultaneous input from a number of different sensor types beyond vision, including radar, lidar, driver monitoring, thermal sensing, and ultrasonic sensor for short range (parking) application.
Arriver’s second generation drive policy; which will drive the vehicles decision making and actuation of the cars movement; will be developed and offered as part of the full software solution.
The drive policy stack will further allow for up to 12 cameras, creating a high resolution, real time image of the environment around the vehicle. This capability will pave the way for fully autonomous driving in most scenarios and be Level 3 and higher capable from its introduction in production vehicles by 2024.
Using the latest artificial intelligence and deep learning techniques both the perception and drive policy software will be able to support active learning, allowing for improved functionality over time with limited need for human interaction. In addition, Arriver is using advanced annotation concepts like non-causal stereo vision for automatic ground truth generation, leveraging the strong stereo vision knowledge and extensive data base. Non-causal stereo is moving detailed stereo information from future frames to the current frame, generating accurate ground truth for training and validation.
As its main System on a Chip (SoC) hardware platform Arriver will use the Qualcomm Snapdragon Ride platform expanded to support all tiers from L1 to L3 and above systems, which will bring industry leading performance and power management and a range of options allowing for a scalable system suitable for mass market cars as well as pushing the envelope for what is possible in the high-end premium vehicle market.
In summary, Arriver software will be scalable and flexible, delivering industry leading performance in a new way not available in the market today. The software is going to be available to all auto manufacturers and Tier1 suppliers, and this will allow car manufacturers to create fantastic driving experience tailored to their own specific needs.