Task 010/011: Self-flying Droves/Personalized Robots

Abstract:

In this talk, we discuss how we make use of data-driven approaches to autonomously fly a quadrotor under Size, Weight and Power (SWAP) constraints. Due to the small form of the platform, the payloads and computational power are limited. We discuss scenarios where the limitation of sensor capacity affects state estimation, the fundamental building block to a robotics system, presenting significant challenges to autonomous flight. We further present our recent work using data-driven approaches to address these challenges from three different perspectives: flying under collision (obstacle detection failure), flying with only an IMU (vision failure), and flying with a robust controller (modeling VIO errors).

Bio: Laura Jarin-Lipschitz is a 2nd year PhD student at the University of Pennsylvania, working on planning and control for robots.

Bio: Wenxin Liu is a 4th year PhD student working with Vijay Kumar and Kostas Daniilidis on Visual-Inertial state estimation for constrained platforms and the application of deep learning in this domain.