Secret FAA Program, Humanoids Cross a Threshold

Date (UTC): 2026-04-21

Hi its Sam Chandra - CEO/Founder of Deepsky and the author of this newsletter. Here's my take on todays news - which represents the last 5 days of AI in Aviation happenings around the world:

Two stories this week are genuinely historic, and I want to flag both up front before anything else. Physical Intelligence released Pi 0.7, which is the starting gun for physical AI in robotics. And the FAA is quietly building SMART, an AI system to de-conflict the entire US national airspace. Either one of these alone would be the lead story of the week. Both of them landing inside 5 days is a moment.

First, Pi 0.7. Physical Intelligence has built an AI model that goes inside robots. You can coach it mid-task - tell it how to do something - and it learns as you go, and then it remembers. Quick history lesson: OpenAI released GPT-1 in 2018. GPT-2 followed in 2019. GPT-3 dropped in 2020, and that was the foundation ChatGPT was built on when it launched in late 2022. Four years from "barely coherent" to "changed the world." Pi 0.7 is that same GPT-1 moment for robotics, and it is hard to overstate how powerful this technology is going to be. Every ground handler, MRO arm, and apron vehicle is a customer a few years out.

Second, FAA SMART. The administrator is personally overseeing this in relative silence, which tells you exactly how seriously they are taking it. This is the application of AI in aviation that I have been talking about for a while, and it is hugely consequential for our air traffic system - and by extension for the whole industry. They are aiming to de-conflict traffic now, stretching the prediction window from 15 minutes out to hours, with the obvious efficiency gains once that is in place. Three vendors are competing in the bake-off: Palantir has the deepest AI chops of the three by a mile, and a huge federal delivery track record. Thales knows aeronautics and air traffic management better than anyone else in the running. Air Space Intelligence is the youngest - they move faster and iterate with more agility than either of the big primes.

Enjoy.

FAA Quietly Building "SMART" — Palantir, Thales, and Air Space Intelligence Compete

FAA Administrator Bryan Bedford is personally leading development of Strategic Management of Airspace Routing Trajectories (SMART), a 4D AI model of the National Airspace System that extends conflict-detection from roughly 15 minutes out to multiple hours. Palantir, Thales, and Air Space Intelligence are competing in a bake-off for the program, which sits inside the FAA $32.5B modernization push and could be operational in some form before the end of 2026.

USAF Experimental Ops Unit Flies Anduril YFQ-44A Fury CCA from a Laptop

The Air Force Experimental Operations Unit executed weapons-load, autonomous taxi, takeoff, and in-flight retasking of Anduril's semi-autonomous YFQ-44A Fury Collaborative Combat Aircraft using only a ruggedized laptop and a small ground crew. A CCA production decision is expected in summer 2026, with the Fury competing head-to-head against Northrop's YFQ-48A Talon Blue.

U.S. Navy Selects Shield AI for Up to $800M in V-BAT Hivemind ISR Services

The U.S. Navy selected Shield AI alongside AeroVironment, Insitu, and Textron to compete task orders under an $800M ceiling contractor-operated ISR vehicle using V-BAT and the Hivemind autonomy stack. It is a structural shift in Navy ISR procurement, framing autonomy as a managed service rather than a traditional platform purchase.

Airbus and Virgin Atlantic Launch "Smart Catering" at Aircraft Interiors Expo Hamburg

Airbus demonstrated "Smart Catering" live on Virgin Atlantic A330 and A350 flights at AIX Hamburg (Apr 14-16), using a crew-tablet vision model to auto-detect unused meals and sync route-level catering orders. The system targets double-digit reductions in cabin food waste and has moved from internal pilot to full commercial pitch.

Anthropic Releases Claude Opus 4.7 — Retakes the Lead on Long-Horizon Agents

Anthropic released Claude Opus 4.7 on April 16, retaking the lead across most major benchmarks and specifically tuned for reliability on long-running agentic tasks. That capability is exactly what aviation products in pilot debriefs, aircraft lease transitions, and MRO parts sourcing need to move from proof-of-concept into production deployment.

Physical Intelligence Pi 0.7 — Coach a Robot Mid-Task and It Remembers

Physical Intelligence's Pi 0.7 is a single vision-language-action model that executes novel robot tasks - cooking a sweet potato in an air fryer, folding laundry on a bimanual UR5e - with zero or near-zero training data, using mid-task verbal coaching that it then internalizes. On the air-fryer task, success climbed from 5% to 95% after 30 minutes of coaching. This is the GPT-1 moment for robotics.

Sam here again - hope you have enjoyed reading this edition. The SMART story and the speed at which AI is landing in safety-critical aviation both point to the same question: can your exposition keep pace with the technology? That is exactly what we do at Deepsky - consulting on AI adoption, aeronautical information management systems, and automating audit and compliance checks across your exposition. If any of that sounds useful to your operation, contact me at admin@deepskyai.com or visit deepskyai.com.