NVIDIA Launches Jetson Xavier NX Module, Auterion Builds PX4 Drone AI Unit Around It

NVIDIA this week launched the Jetson Xavier NX module, a credit-card-sized board (70 by 45 millimeters) priced at $399 that delivers up to 21 TOPS while consuming just 15 watts. Auterion, the commercial company behind the open-source PX4 flight-control software, has already built a product around the module called AI Node, designed to run artificial intelligence directly on drones and ground robots.

A supercomputer that fits in a drone's cargo bay

NVIDIA is marketing the Xavier NX as its new calling card for robotics and edge computing. It's an extremely compact physical module, but one built on a compute architecture that squeezes processing power once reserved for large servers into a package no bigger than a playing card.

• Price: $399 per unit
• Dimensions: 70mm by 45mm
• AI performance: up to 21 TOPS (trillion operations per second)
• Power consumption: 15 watts
• Processor: six-core NVIDIA Carmel CPU
• GPU: Volta architecture with 384 cores, 48 Tensor Cores and two deep-learning accelerators
• Memory: 8GB RAM

The ability to run multiple neural networks in parallel while simultaneously processing data from several high-resolution sensors marks the real gap versus earlier solutions in this category. In the drone world, where every gram and every watt counts, this is exactly the kind of trade-off payload designers look for: high compute power without significantly cutting into flight time or the aircraft's overall payload capacity.

Why is Auterion betting on local processing instead of raw data transmission to the ground?

Auterion, the company that commercializes and maintains the open-source PX4 flight-control project, didn't wait around. It built a dedicated product called AI Node that uses NVIDIA's module as the processing core for drones and ground robots running PX4. The core idea: give a drone the ability to detect objects in real time and make operational decisions without streaming all the raw sensor data down to a ground station for analysis.

This approach solves a familiar operational problem in the industry. Real-time high-resolution video transmission demands high bandwidth, suffers from latency, and depends on a stable link between the drone and the ground, a limitation that's especially acute in rural or industrial areas with weak cellular coverage. Processing data locally on board the aircraft itself reduces this dependency, effectively letting the drone act as an independent smart sensor rather than a pipe for raw data transfer.

What exactly does this mean for developers of commercial and industrial drones?

For companies developing drones for infrastructure inspection, precision agriculture or security, combining relatively cheap hardware with mature open-source software like PX4 could significantly lower the barrier to entry for building advanced AI applications. Instead of building proprietary compute infrastructure from scratch, developers can now assemble a solution around an off-the-shelf module and open software.

Open questions remain. A 15-watt TDP module is not negligible for a small drone, and the real-world impact on flight time depends heavily on each aircraft's specific configuration. A $399 price tag for the module alone, before factoring in integration, cooling and accompanying software, is also a parameter developers will need to weigh against their overall project budget.

Why is NVIDIA expanding cloud-based developer support across the entire Jetson family?

At the GTC 2020 developer conference NVIDIA is holding this week, the company also announced cloud-native support for its development kits, an expansion that applies to the entire Jetson edge-computing product line and not just the new Xavier NX. The move signals a clear direction: NVIDIA wants robotics and drone developers to be able to develop, test and deploy AI models remotely, without depending on physical access to each individual hardware unit.

It's hard not to be impressed by the pace at which one chip company manages to calibrate an infrastructure product so precisely that within weeks it becomes a key component in another independent software company's solution. This is exactly the kind of ecosystem that could determine who wins the race to set the next development standard in the world of autonomous drones.