Ambarella has revealed its centralized 4D imaging radar architecture, which allows for both the central processing of raw radar data, and deep low-level fusion with other sensor inputs – including cameras, lidar and ultrasonics. The architecture promises greater environmental perception and safer path planning for L2+ to L5 automated driving systems and AI-based ADAS.
Some of its key capabilities stem from its integration of Ambarella’s Oculii radar technology, such as its AI software algorithms (which dynamically adapt radar waveforms to the surrounding environment). These allow for a high angular resolution of 0.5 degrees, a dense point cloud scaling up to thousands of points per frame, and a detection range exceeding 500 meters. These functions are mutually achieved with an order of magnitude fewer antenna MIMO channels, reducing the data bandwidth and power consumption over similar 4D imaging radars. The radar’s support for Oculii is targeted at system integrators looking to future-proof new radar designs with a flexible, high-performance perception architecture.
In developing the new architecture, Ambarella optimized Oculii’s algorithms for its CV3 AI domain controller SoC family – adding specific radar signal processing acceleration. The CV3’s AI performance per watt offers the compute and memory capacity needed to achieve high radar density, range, and sensitivity. The company has said that a single CV3 alone can efficiently provide high-performance, real-time processing for perception and low-level sensor fusion, as well as path planning within autonomous vehicles and robots centrally and simultaneously. The new architecture works in a similar way to maximize efficiency – helping reduce operational costs by consuming less power at the maximum duty cycle. Here, it reduces the bandwidth for data transport by 6x, while eliminating the need for pre-filtered, edge processing and its resulting loss in sensor information.
By applying AI software to dynamically adapt the radar waveforms generated with existing monolithic microwave integrated circuit (MMIC) devices, and using AI sparsification to create virtual antennas, Oculii reduces the antenna array for each processor-less MMIC radar head to 6 transmit x 8 receive. Overall, the number of MMICs is drastically reduced, while achieving 0.5 degrees of joint azimuth and elevation angular resolution.
This cost-effective, software-defined centralized architecture also supports the dynamic allocation of the CV3’s processing resources (based on real-time weather conditions between sensor types and among sensors of the same type). In extreme rainy conditions that diminish long-range camera data, for example, the CV3 can shift some of its resources to enhance radar inputs. Likewise, if it is raining while driving on a highway, the CV3 can focus on data coming from the vehicle’s front-facing radar sensors to further its detection range while providing faster reaction times.
Within CV3 is Ambarella’s new, next-generation, CVflow architecture. Provided with a neural vector processor and a general vector processor designed to include radar-specific signal processing enhancements. These processors work together to run Oculii’s advanced radar perception software while enhancing its performance – with Ambarella claiming it can achieve speeds up to 100x faster than traditional edge radar processors.
Additional benefits of the new centralized architecture include easier OTA software updates, for continuous improvement and future proofing. A single OTA update can be pushed out for the CV3 SoC and aggregated across the system’s full suite of radar heads. These radar heads eliminate the need for a processor, reducing the upfront cost and any subsequent maintenance costs in the event the system is damaged in a crash.
Target applications for the new centralized radar architecture include ADAS and L2+ to L5 autonomous vehicles, as well as autonomous mobile robots and automated guided vehicle robots. These designs are streamlined by Ambarella’s unified and flexible software development environment, which provides automotive and robotics designers with a software-upgradable platform for scaling performance from ADAS and L2+ to L5. Ambarella confirmed that its new centralized architecture will be demonstrated at its invite-only event during CES 2023, held at the Las Vegas Convention Center from January 5 to January 8.