Sebastian Scherer

Extracting long-range navigational insights from satellite imagery significantly boosts off-road navigation performance, reducing reliance on human input.

Robots can now autonomously learn to avoid failures by transforming recovery experiences into abstract planning skills, achieving over 50% better performance on unseen tasks.

Existing robotic methods falter in tackling fundamental physical reasoning challenges, as evidenced by KinDER's rigorous benchmark evaluation.

Coordinating robot teams to explore and relay information just got smarter: PRoID predicts future information gain to decide when robots should relay data, outperforming fixed strategies, especially when robots are at risk.