> Until very recently, autonomous drones took twice as long as those piloted by humans to fly through a racetrack, unless they relied on an external position-tracking system to precisely control their trajectories. Swift, however, reacts in real time to the data collected by an onboard camera, like the one used by human racers. Its integrated inertial measurement unit measures acceleration and speed while an artificial neural network uses data from the camera to localize the drone in space and detect the gates along the racetrack. This information is fed to a control unit, also based on a deep neural network that chooses the best action to finish the circuit as fast as possible.
Well, they've muddied the waters well enough. The b-roll in the last segment of the video you linked shows motion tracked drones and the 2022 event that used motion tracking (the balls used for motion tracking are visible on the drone arms). The voiceover tells us the system won against human racers. So, even if they have a vision only run made on the same track that beat human times, that run wasn't shown (but you couldn't tell that if you just watched that video).
And from BMSThomas' video, it's obvious their vision only system wasn't ready for prime time in 2022, but they were still working on it.
This paper is based on the performance at the 2022 event, but it took this long to write it up and get it published.
The tracking system was used during all of the races for data collection of both the autonomous and human-piloted drones, which is why the reflective markers are visible.
They did do some demonstrations of the drones controlled with the tracking cameras, and they were significantly faster, but the vision-based drones were definitely able to fly faster than the human pilots in some races.
My read is they were only using the motion tracked system to train the residual network, but the actual competitive flights were done without input from the external tracking but they still captured the data, so they could feed it to their training for future runs.
> Until very recently, autonomous drones took twice as long as those piloted by humans to fly through a racetrack, unless they relied on an external position-tracking system to precisely control their trajectories. Swift, however, reacts in real time to the data collected by an onboard camera, like the one used by human racers. Its integrated inertial measurement unit measures acceleration and speed while an artificial neural network uses data from the camera to localize the drone in space and detect the gates along the racetrack. This information is fed to a control unit, also based on a deep neural network that chooses the best action to finish the circuit as fast as possible.
And here's a video that seems to have been uploaded today - https://www.youtube.com/watch?v=fBiataDpGIo