What does it look like if a drone no longer merely functions as a “flying camera” and instead begins to think for itself? At the event in Santa Clara this week, Ambarella gave quite an interesting presentation, speaking about how its CV5 AI system-on-chip solution is revolutionizing air intelligence and how the newly introduced Antigravity’s A1 drone, which embodies this capability, will bring this on-orbit intelligence capability.
The Ambarella CV5 is based on its CVflow® AI engine, intended to deliver robust computer vision and deep learning inferencing to power- and thermally-constrained edge devices. This is much bigger than simply enabling faster image processing. This is about being able to run AI processing right on the drone and enabling use cases like intelligent subject tracking, obstacles, and scene awareness, all without needing to have constant connectivity to the cloud. As Ambarella President and CEO Fermi Wang described their new offering: “Ambarella’s heritage in high-quality imaging, combined with our CVflow AI roadmap, enables drone makers to push more autonomy and more insight onto the drone itself, where every millisecond and every milliwatt matters.”
For UAV engineers, the implications of edge AI are a reduction in latency as well as increases in the reliability of operational capabilities. The ability to perform operations locally reduces the required bandwidth in the backhaul connection. This is a highly desirable characteristic in areas of application such as infrastructure scanning, agricultural surveying, or public security. These trends are reflected in the larger industry landscape as described in the context of AI-enhanced eco-efficient UAV design.
The Antigravity A1 is the first UAV that incorporates an “all-in-one 8K 360-degree camera,” which provides immersive and high-resolution video capture with no need for the use of accessories. But the important part is the ability of this camera to work with the CV5’s onboard AI processor to enable inferencing from the panoramic image in real time. What this means is that the UAV has the capability to record high-resolution video and analyze an environment simultaneously. The message from Michael Shabun, the Antigravity UAVs CEO, is that they aimed to be efficiency and performance-driven, and they can do so “By leveraging Ambarella’s CV5 SoC for high-quality imaging with its integrated CVflow architecture for on-device AI acceleration and inferencing, we can deliver intelligent capabilities at the edge while staying focused on efficiency and performance.”
From the hardware side, the low-power system of the CV5 promotes extended flight time and lowers thermal management complexity, making it easier to integrate the system into compact drones. This becomes even more important as the number of sensor payloads in drones is set to increase, as forecasted by IDTechEx, to 10-15 sensors per platform by 2036, which include LiDAR, radar, and multi-IMU systems in industrial drones as well as in BVLOS drones. The challenge of integrating compute intensity in connection with SWaP (size, weight, and power) requirements also applies to other AI systems similar to the CV5, including nano-drone navigation applications in which systems operate under resource-constrained environments.
The Antigravity A1, for example, relies on a CV5 with an imaging pipeline that is capable of handling tough functions such as HDR imaging and noise reduction during dim lighting. Such integration of cinematographic imaging and AI perception capabilities is set to revolutionize self-guiding mapping activities, immersive media capture, and safety-critical inspection missions.
Another thing that Ambarella plans with its roadmap is the development of scalable AI SoCs with the potential of managing degrees of autonomy from L1 to L4-level autonomy in aerial robots. For drone developers, this means they would have drones capable of executing complex tasks like multi-object classification and swarm intelligence without having to compromise on flight duration or carrying capacity. The CV5 series is optimized with architecture capable of seamless integration with sophisticated sensor bundles. With the global drone industry racing towards an estimated 147.8 billion dollars by 2036, the likes of the Antigravity A1 demonstrate how edge AI and next generations of imaging can be the differentiators that distinguish products within a crowded sector. With the coming together of the successful imaging lineage of Ambarella and the real-time AI inferencing capabilities, the symbiosis indicates a pivot within the UAV sector from the capture of the world to the understanding of the world—and this all happens immediately.