Learning Agile, Vision-based Drone Flight: from Simulation to Reality

TL;DR

This paper discusses methods for training deep sensorimotor policies for agile, vision-based drone flight and successfully transferring these policies from simulation to real-world environments.

Contribution

It introduces transfer methodologies for simulation-trained policies to real drones and highlights open challenges for achieving human-pilot level agility and robustness.

Findings

Successful transfer of policies from simulation to real drones

Identification of key open research questions

Enhanced understanding of vision-based drone control

Abstract

We present our latest research in learning deep sensorimotor policies for agile, vision-based quadrotor flight. We show methodologies for the successful transfer of such policies from simulation to the real world. In addition, we discuss the open research questions that still need to be answered to improve the agility and robustness of autonomous drones toward human-pilot performance.