World Models
Updated: Sep 24
Written by Jason Rock
We’re excited to see the aviation industry continuing to embrace modern AI approaches. We have long believed in the potential of modern AI to provide advanced automation and safety capabilities for pilots. In the mid to late 2010s, members of our leadership team were at the forefront of demonstrating the value of convolutional neural networks for aerospace at companies like Boeing and Airbus. That experience, combined with the proven impact of modern AI in other industries, helped inspire the founding of VTI Aerospace.
We also think it’s important to recognize the hard lessons that we’ve learned in our earlier roles. While computer vision has shown immense promise for aviation in the lab for years, making the leap to a deployed airborne system is much harder.
The reality is that the hard problem for Aerospace isn’t just about getting a good number on a benchmark, it’s about fitting AI into a system that works day after day, in an uncompromising operating environment and on constrained hardware. The AI also has to fit into a product that pilots actually want to use. Pilots we talk to tell us that what they want is a system that reduces their workload and improves safety in a way that actually makes sense to them.
Just like we were drawn to work on computer vision in the 2010s, we are now extremely excited by the advances demonstrated by multimodal world-models. We see enormous value in AI grounded and informed by the flight deck and broader airspace context, combining what the aircraft sees with communications, aircraft state, maps, and operational situational awareness.
We are continuing to pioneer how this next generation of AI fits into the cockpit!
