We are excited to share our #CoRL2026 paper, "Learning Visuomotor Quadrotor Flight Through Dynamic Narrow Gaps via Differentiable Simulation", where we demonstrate autonomous quadrotor flight through narrow moving gaps using exclusively onboard vision, with policies trained from scratch in just minutes via #DifferentiableSimulation! Code released!
Paper: https://t.co/pOonTWiZEy
Video: https://t.co/PKPsXugoao
Code: https://t.co/hONCfpvweG
Flying through a narrow, moving gap using only egocentric vision is extremely challenging, as it requires centimeter-level precision, split-second timing, and perception-aware control. We use differentiable simulation to train vision-based control policies directly from scratch, without expert demonstrations, planners, or teacher policies. The key idea of uding differentiable simulation is that, instead of learning only from success or failure, the drone receives informative gradients that guide it toward better alignment, timing, and collision avoidance. Our approach achieves 99% success in simulation and 80% in the real world, with training completed in just minutes on a single GPU. Compared to PPO, it requires 20× fewer training transitions and 40× less training time! Real-world experiments use only onboard visual and inertial sensing. The motion capture visible in the video is used only for ground-truth measurements.
Kudos to Riccardo Bianco, Yunfan REN, and Jiaxu Xing!
@uzh_ifi@UZH_Science@UZH_en@UZHspacehub@ERC_Research@UZH_ai@AUTOASSESS_EU
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
Gliding through the waves, the Stonking Stingray emerges with agility and grace.🌊
See it up close: https://t.co/DuzJWhZxXh
Ubuntu 26.10 lands on October 15, 2026.
🎓📸Ya puedes descargar la fotografía de generación 2025–2026 del Cinvestav.
Una imagen que reúne el esfuerzo, la dedicación y el compromiso de quienes hoy se suman a una comunidad científica con 65 años de historia.
👇
https://t.co/pB1OX8OQ1O