# Autonomous Drone Racing with Deep Reinforcement Learning

**Authors:** Yunlong Song, Mats Steinweg, Elia Kaufmann, Davide Scaramuzza  
**arXiv:** [2103.08624](https://arxiv.org/abs/2103.08624)  
**Date:** 2021-03-15  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2103.08624v2/

## Key points

- Authors Yunlong Song, Mats Steinweg, Elia Kaufmann, and Davide Scaramuzza introduce a groundbreaking approach to near-time-optimal trajectory generation for quadrotors in autonomous drone racing.
- The work titled "Autonomous Drone Racing with Deep Reinforcement Learning" addresses the challenge of planning time-optimal trajectories through waypoints with limited prior knowledge.
- The authors leverage deep reinforcement learning and relative gate observations to compute near-time-optimal trajectories that can adapt dynamically to changes in the environment.
- This novel method offers significant computational advantages over traditional trajectory optimization methods, especially for complex track configurations.
- The proposed technique is evaluated on various race tracks in both simulation and real-world scenarios, achieving speeds of up to 60 km/h under physical conditions.
- The research has been accepted for publication at the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) in Prague 2021, highlighting its relevance and contribution to advancements in autonomous drone racing technology.

## Summary

Authors Yunlong Song, Mats Steinweg, Elia Kaufmann, and Davide Scaramuzza present a groundbreaking approach to near-time-optimal trajectory generation for quadrotors in the field of autonomous drone racing. Their work titled "Autonomous Drone Racing with Deep Reinforcement Learning" addresses the challenge of planning time-optimal trajectories through a set of waypoints with limited prior knowledge. Traditionally, solutions have been specialized for specific track layouts or compromised by simplified assumptions about platform dynamics. By leveraging deep reinforcement learning and relative gate observations, the authors introduce a novel method that can compute near-time-optimal trajectories and dynamically adapt to changes in the environment. This approach offers significant computational advantages over traditional trajectory optimization methods, particularly for complex track configurations. The proposed technique is evaluated on a variety of race tracks both in simulation and in real-world scenarios. Remarkably, the quadrotor achieved speeds of up to 60 km/h under physical conditions, demonstrating the effectiveness and practicality of the developed approach. This research has been accepted for publication at the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) in Prague 2021, showcasing its relevance and contribution to advancements in autonomous drone racing technology.