NVIDIA Aerial and Sionna pass 1M downloads for 6G

What shipped?

According to an NVIDIA blog post, a year ago the company open sourced its Aerial wireless software. It joins Sionna in a broader open toolkit for next-generation 6G development, linking simulation and AI training with real over-the-air experiments.

Over the past year, global developer adoption of AI Aerial and Sionna grew 10x. NVIDIA's open source 5G/6G tools surpassed 1 million downloads. More than 8,000 research and academic institutions use the platforms.

That fits the same open NVIDIA line as the Cosmos 3 world sim models and Nemotron 3.5 Transcribe ASR: the company is opening research stack pieces, not only language models.

Where are the experiments running?

At King's College London, researchers connected a camera-equipped robot arm to an NVIDIA Jetson AGX Orin running Sionna. The goal is goal-oriented semantic communications: instead of full images, the robot sends compact descriptions of objects, positions and relationships. In fruit-packing and cube-stacking tests with deliberately introduced faults, task completion time fell by up to 52.6% versus periodic image sending, and task success improved.

SoftBank integrated a neural receiver into NVIDIA Aerial CUDA-Accelerated RAN (ACAR) and tested it with a commercial 5G smartphone. In controlled tests with interference, the neural receiver beat a conventional receiver on uplink throughput at all 15 settings, with a maximum gain of roughly 60%.

Using the Aerial Testbed, ETH Zurich researchers tuned neural receivers with local measurements. Block error rates fell by more than one-third versus selected pretrained versions.

At the University of Leeds, a system predicts network demand 10 minutes ahead and uses those forecasts to guide base station sleep decisions. Before powering down, it checks that neighboring active stations can still serve affected devices. The team validated the approach with a portable NVIDIA DGX Spark-powered 5G network-in-a-box and a digital twin. Result: a 45.6% cut in RAN power use versus a full-power baseline, while keeping devices connected.

CTTC is wiring GPU-accelerated radio channel models running on ACAR into its open 5G-LENA network simulator. In an early evaluation, runtime fell from about three hours to 10-20 minutes, while preserving expected statistical behavior.

Why it matters

6G talk often centers on hardware and spectrum. Here the open software stack (simulation + RAN + live experiment) already produces measurable results on robotics, receivers and energy.

Practical takeaway: if you experiment with 5G/6G or live robot links, look at the Aerial and Sionna packages on GitHub. The labs above share that common base, so results are easier to compare than closed one-off tests.