Researchers led by Xiaofeng Jin, with Matteo Matteucci among the authors, have built AGE-Capture, which lets a drone pick its own viewpoints for detailed 3D scene reconstruction and judge when the job is done, according to a paper on arXiv. It works with different range sensors and camera models.
Across ten indoor and outdoor simulated scenes, its reconstructions scored 3.30 to 3.53 dB higher in peak signal-to-noise ratio than online baselines. It used 58.2% of the shared mission-time budget on average and 91.0% to 93.3% less peak GPU memory. Flights in nine real-world scenes showed it running on board the drone.



