Fall 2019 Robotics Colloquium: Debadeepta Dey ( Microsoft Research)

Lecture title: ImitationLearning with Indirect Oracles We present Visionbased Navigation with Languagebased Assistance (VNLA), a grounded visionlanguage task where an agent with visual perception is guided via language to find objects in photorealistic indoor environments. The task emulates a realworld scenario in that (a) the requester may not know how to navigate to the target objects and thus makes requests by only specifying highlevel endgoals, and (b) the agent is capable of sensing when it is l