Vision-based Control of a Quadrotor in User Proximity: Mediated vs End-to-End Learning Approaches

TL;DR

This paper compares mediated and end-to-end learning methods for controlling a quadrotor to hover in front of a moving user using onboard camera data, finding both approaches perform similarly.

Contribution

It provides a direct comparison between mediated and end-to-end learning paradigms for quadrotor control in a user-proximity task, highlighting their equivalent effectiveness.

Findings

Both approaches achieve similar performance levels.

Qualitative analysis of quadrotor behaviors was conducted.

End-to-end learning can bypass high-level state estimation without loss of performance.

Abstract

We consider the task of controlling a quadrotor to hover in front of a freely moving user, using input data from an onboard camera. On this specific task we compare two widespread learning paradigms: a mediated approach, which learns an high-level state from the input and then uses it for deriving control signals; and an end-to-end approach, which skips high-level state estimation altogether. We show that despite their fundamental difference, both approaches yield equivalent performance on this task. We finally qualitatively analyze the behavior of a quadrotor implementing such approaches.