Jon Arrizabalaga - Egocentric Decision-Making for Robot Autonomy

The podcast centers on embracing egocentricity in autonomous decision-making for robotics. Jon Arrizabalaga from the Technical University of Munich, currently at CMU, discusses how biological systems make decisions from a subjective, first-person viewpoint, contrasting this with the allocentric, world-map-reliant approach common in robotics. He uses examples like the DARPA Subterranean Challenge, autonomous driving, and the Curiosity rover to illustrate the gap between desired robotic agility and current limitations. Arrizabalaga proposes a framework with three key ingredients: generalizing spatial projections, choosing suitable adaptive frames (specifically egocentric Parallel Transport Frames), and representing environments through differentiable parametric corridors, emphasizing the importance of safety guarantees and strategic decision-making.