Event-based Navigation for Autonomous Drone Racing with Sparse Gated Recurrent Network

TL;DR

This paper introduces a novel deep learning approach using event-based vision and gated recurrent units with sparse convolutions for real-time autonomous drone racing, improving gate detection accuracy and efficiency.

Contribution

It presents a new perception pipeline combining event-based vision and gated recurrent units, with pretrained models on simulated data, for improved drone racing navigation.

Findings

Enhanced gate detection precision with event-based vision and recurrent units

Real-time navigation demonstrated on a physical drone platform

Public release of a new drone racing dataset

Abstract

Event-based vision has already revolutionized the perception task for robots by promising faster response, lower energy consumption, and lower bandwidth without introducing motion blur. In this work, a novel deep learning method based on gated recurrent units utilizing sparse convolutions for detecting gates in a race track is proposed using event-based vision for the autonomous drone racing problem. We demonstrate the efficiency and efficacy of the perception pipeline on a real robot platform that can safely navigate a typical autonomous drone racing track in real-time. Throughout the experiments, we show that the event-based vision with the proposed gated recurrent unit and pretrained models on simulated event data significantly improve the gate detection precision. Furthermore, an event-based drone racing dataset consisting of both simulated and real data sequences is publicly released.