New AI Speeds Robot Vision by Skipping Static Pixels
Penn's F³ system runs two to five times faster in testing, a research-stage result that could reach autonomous drones and future event-camera hardware.
Updated on Nov. 20, 2026 in Robotics

University of Pennsylvania researchers have developed Fast Feature Field (F³), an AI system that processes signals from event cameras, which detect changes in brightness at individual pixels. The work is research-stage: the researchers report it was tested on flying drones, four-legged robots and vehicles, and it has not been described as a commercial product.
Why it matters
Event cameras can generate tens of millions of signals every second, and conventional software has struggled to keep up with that volume on robots that must react quickly. F³ targets that bottleneck by focusing computation only where the scene is changing, a design choice that matters for machines operating in fast-moving or poorly lit conditions.
The reported figures are 120 updates per second at high definition and 440 at VGA resolution, with algorithms using F³ running roughly two to five times faster than earlier methods. The source does not identify the baseline systems or the size of the test set behind the speed comparison.
The players
University of Pennsylvania
A research university whose engineering and robotics groups developed F³ for processing event-camera data in autonomous machines.
The details
Event cameras report only the pixels whose brightness changes, rather than capturing full frames at fixed intervals. F³ uses a neural network trained to anticipate upcoming changes in the scene, and it skips calculations at locations where no change is occurring. Because most pixels are static at any given moment, this selective approach lets the system spend its processing on the small share of the scene that is actually moving.
Timeline
The associated article about F³ was published on November 20, 2026.
The Tech Race
Event cameras are an emerging sensing approach that competes with frame-based vision for fast robotic perception. F³ represents one research route to making that hardware practical, and its reported speedups are the kind of result that other labs pursuing event-based vision will need to match.
The work is at the research stage, so F³ is not yet available to users or to commercial drone makers. The researchers say they are using it to build autonomous drones for crowded and low-light settings and expect to integrate it into future event-camera hardware, but no release dates were given.
The takeaway
Watch for the researchers' next demonstrations of F³ on autonomous drones operating in crowded or low-light environments. Whether the two-to-five-times speedup holds on independent benchmarks will indicate how far the approach can carry event-camera perception.
Further reading
For more on the machines this research is aimed at, explore Robotics coverage from the Philadelphia region.
Source note: This article includes information reported by IHLS.




