Antonio Loquercio on Agile Autonomy: Learning High-Speed Flight | Toronto AIR Seminar

Abstract:
Quadrotors are among the most agile and dynamic machines ever created. However, developing fully autonomous quadrotors that can approach or even outperform the agility of birds or human drone pilots with only onboard sensing and computing is very challenging and still unsolved. Current state-of-the-art works tackled this problem by splitting the task into a series of consecutive blocks: perception, map building, and planning. Although simple and effective, such an approach typically discards interactions among the different blocks and requires each block to make over-simplifying assumptions. Additionally, due to the presence of sequential processing blocks between sensors and actuators, the time to go from observation to action increases at the cost of agility. In this talk, I’ll present the astonishing potential of deep sensorimotor policies for agile vision-based quadrotor flight. Learning sensorimotor controllers represents a holistic approach that is more resilient to noisy sensory observations and imperfect world models. Such deep sensorimotor policies, generally trained in simulation, enable autonomous quadrotors to fly faster and more agile than what was possible before with only onboard sensing and computation. While humans and birds still have the advantage over drones, I will show the current research gaps and discuss possible future solutions. Teaser videos of the results I will present are Deep Drone Acrobatics (   • Deep Drone Acrobatics (RSS 2020)  ) and Learning High-Speed Flight in the Wild (   • Learning High-Speed Flight in the Wild (Sc...  ).


Paper:
https://doi.org/10.1126/scirobotics.a... (https://rpg.ifi.uzh.ch/docs/Loquercio...)
https://doi.org/10.48550/arXiv.2006.0... (accepted to RSS)
https://arxiv.org/abs/2009.00563 (accepted to CoRL)


Bio:
Antonio Loquercio is a PostDoc at the University of California at Berkeley. He works at the intersection of robotics, computer vision and neuroscience at BAIR under the supervision of Prof. Jitendra Malik. Before that, he got his doctoral degree at ETH and University of Zurich with the Robotics and Perception Group under the supervision of Prof. Davide Scaramuzza.
https://antonilo.github.io/


Toronto AIR Seminar:
The Toronto AI Robotics Seminar Series is a set of events featuring young robotics and AI experts. The talks are given by local as well as global speakers and organized by the Faculty and Students at University of Toronto’s Department of Computer Science. We welcome students, researchers and robotics enthusiasts from around the world to join us and interact with the Toronto Robotics Community.
Find out more at: https://robotics.cs.toronto.edu/toron...