Ismail Geles
Lead author and primary implementer of the multi-agent drone racing RL system at University of Zurich.
“Geles, Bauersfeld, Wulfmeier, Scaramuzza — University of Zurich / Google DeepMind, 2026”
Source→“league-play training reduces collision rates by over 50% compared to single-agent baselines (23.30% average crash rate vs. 77.99%) while sacrificing only 0.03 seconds per lap in speed.”
Source→“In poker, AI first achieved superhuman play in two-player zero-sum settings... The transition to six-player poker proved categorically harder... Just as game-playing AI had to advance from two-player contests to multiplayer settings, robotic control must advance from optimizing solo trajectories to dynamic co-existence.”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.