[Jodavaho.io] — Josh Vander Hook

Big news! As of last week, we’re sending fully autonomous flight optimizations directly to aircraft for our partner airlines. Huge personal milestone for me, as I worked exclusively on the optimization and estimation pipelines that do this!

Why am I sending messages to cockpit screens?

Flight Science tracks flights for partner airlines, and helps them improve operations. We plug into various and sundry commercial and customer data streams; collate, estimate, and optimize things in real time; generate alerts and recommendations custom built for their objectives; and send them to the cockpit and/or dispatchers to help improve the experience, safety, and efficiency of the flight.

My background here…

I joined in 2024, and when it was just me and Andrew, we banged out the initial planning system over a couple of months. It’s iterated a dozen times at least, getting faster, more precise, and more reliable over time. I’m fond of saying that 8 my 10 favorite systems I’ve worked on were here at Flight Science.

The system has generated and sent messages for almost 1 year, but it involved me sitting at my machine and approving or denying each message. No longer! It’s stable and the recommendations are good enough that we don’t need to do that anymore.

What do we send?

Notable examples are alerts: “Turbulence ahead”, Replan: “Do this to avoid turbulence”, Efficiency: “FL320 would save xxx kg of fuel from improved following wind”, and Logistical: “Flight YYYY is due to land in 13 minutes”

How does it work?

We have precise models for key parts, optimization engines for estimating some things, and excellent heuristics for others. You can imagine we have to track the weight of all passengers, luggage, fuel, etc, the temperature of every cubic meter of air they fly through, the fractional kilogram of fuel they use for every stage of flight, airline/pilots preferred flight profile, like climb rates and speeds, and so on… And, we have to track in real time any changes to their plans - so we can look ahead for them.

Then all this information has to be piped into a real-time planner and quickly provide a batch of alternative actions, which we can score against their preferences and decide to send an alert or plan or whatever. And for each alternative we have to re-calculate projected values for all of that stuff above.

Then all this goes to the cockpit through a global satellite/vhf/hf network, and appears on one of those awesome looking little green screens, which they can ignore if they like.

Like this screen (not our message):

It’s all a suggestion for situational awareness. We do measure acceptance rate by tracking the activity of the airplane to see if they adopt the plan, it’s much higher than I expected, but more on that later.

I’m super excited. None of this would be possible if our team wasn’t already super plugged into the airline industry. We have excellent early stage partners to develop and test all this.

Amazing time.