Processing efficiency is particularly important for extending battery driving range in the growing electric vehicle (EV) market. With external and internal camera-based perception capabilities essential, a full stack of software and AI modules that reduce energy consumption for AI processing and increase performance is targeting the autonomous driving market.

Ambarella, an edge AI semiconductor manufacturing company, announced on the 12th local time an autonomous driving (AD) software stack based on deep learning AI processing.

The newly launched solution includes modular components such as △environmental recognition △sensor fusion △vehicle path planning.

The latest autonomous driving stack has been designed together with Ambarella’s CV3-AD AI domain controller SoC family, and the stack runs on the SoC’s CVflow AI engine. The chip, which runs on Arm cores, is expected to minimize power consumption and processing load.

Likewise, these SoCs provide the processing required for vision, radar detection and fusion, based on their knowledge and experience with sensing modalities. Ambarella’s stack also provides flexibility in its implementation model, allowing automotive OEMs to leverage part or all of the module in combination with their own software IP.

“Our AD stack builds on more than 25 years of experience and builds on the legacy of our VisLab automotive R&D team, which has consistently evolved our AD stack to leverage the latest AI SoCs,” said Fermi Wang, CEO of Ambarella. “Today, we are delivering on the vision we achieved with our 2015 acquisition of VisLab, which combined with our algorithm-first philosophy for purpose-built SoC design, has resulted in a platform that enables automakers to deliver on the promise of autonomous driving.”

One of the features of Ambarella’s AI-based software stack is its deep learning-based planner. This dynamic vehicle path planning module enables a natural driving experience. This allows the ‘deep planner’ module to be trained with data from a variety of real-world scenarios.

The latest configuration of the R&D vehicle includes a detection suite with mono and stereo cameras and the Ambarella Oculii 4D imaging radar. All processing is performed by the CV3-AD. While this optimal configuration is possible thanks to many years of AD and sensor processing experience, the company emphasized that the solution is flexible enough to support OEM-specific detection suites, including the option to add LiDAR.

A key feature of the software stack is that it only requires standard definition (SD) maps. There is no need for pre-generated high-definition (HD) maps; instead, the stack running on CV3-AD generates HD maps in real time using real-time environmental data from the vehicle detection suite.
In contrast, pre-generated HD maps used in other AD systems are fragile and expensive to maintain, requiring centimeter-level positioning capabilities. They also become unreliable in dynamic situations such as road construction or accidents.

Ambarella’s real-time HD map generation can be ideal for handling challenging AD scenarios, including: urban areas of large cities with roundabouts; narrow roads with parked cars; high traffic volumes; construction sites; high pedestrian density; cyclists; and other vulnerable roadways.

The modular AD software and hardware solution provides OEMs with a scalable platform to select members of the CV3-AD family that provide the right mix of performance and capabilities for each model of vehicle, enabling the expansion of autonomous capabilities in mass-market vehicles.

This includes a wide range of autonomous and semi-autonomous vehicle implementations, from fully featured L2+ systems to higher levels of autonomy, Ambarella emphasized.

Ambarella emphasizes that the CV3-AD SoC runs the entire AD stack with low power consumption, delivering superior AI performance per watt and simplifying thermal management. Processing efficiency is particularly important for increasing battery driving range in the growing electric vehicle (EV) market.

Additionally, the software and hardware are designed to work together optimally, as the stack operates using only a small portion of the CV3-AD’s Arm processing power. This reportedly leaves automakers with ample headroom to integrate additional software to create differentiated features, including functional safety.

In addition to optimized AD software and hardware solutions, Ambarella offers automotive OEMs a supply chain and ecosystem of both in-house and ecosystem partners. This includes custom tools for data acquisition, simulation and annotation, including the CV3-AD’s on-chip encoder. Accelerated software-in-the-loop (SIL) simulations using the CV3-AD PCIe server card simplify development, and Ambarella also provides an automated 3D annotation pipeline to support AI training.

The new Ambarella AD software stack will be demonstrated in fully autonomous test drives at an invitation-only exhibit during CES in Las Vegas.