Imitation Learning with Indirect Oracles for Vision-based Navigation - Fall 2019 Robotics Colloquium

Explore a cutting-edge lecture on Vision-based Navigation with Language-based Assistance (VNLA) presented by Debadeepta Dey from Microsoft Research. Dive into the world of grounded vision-language tasks where agents with visual perception are guided by language to locate objects in photorealistic indoor environments. Learn about the Imitation Learning with Indirect Intervention (I3L) framework and its application to the VNLA task. Discover how this approach significantly improves success rates for learning agents in both familiar and unfamiliar settings. Gain insights into recent developments in stabilizing imitation learning using the IL-as-f-divergence framework. Delivered as part of the Fall 2019 Robotics Colloquium at the Paul G. Allen School, this 56-minute presentation offers a deep dive into machine learning, control, and computer vision applications for autonomous agents and robotics.