Dream to Fly: Model-Based Reinforcement Learning for Vision-Based Drone Flight
Angel Romero, Ashwin Shenai, Ismail Geles, Elie Aljalbout, Davide Scaramuzza
AI summary
Problem
Autonomous drone racing typically relies on explicit state estimation, simplified visual masks, or imitation learning bootstrapping, making it difficult to learn agile flight directly from raw pixels efficiently.
Approach
The authors apply the DreamerV3 model-based reinforcement learning algorithm to train a visuomotor policy that maps raw RGB camera images directly to quadrotor control commands by learning a predictive world model.
Key results
- Trains agile pixel-to-command policies from scratch without intermediate representations or imitation learning
- Emerges perception-aware camera steering toward texture-rich gates without handcrafted viewing rewards
- Successfully transfers learned policies to real-world quadrotors via hardware-in-the-loop simulation
- Achieves real-world flight speeds up to 9 m/s on a Figure 8 track
Why it matters
Demonstrates that model-based RL can efficiently bridge the perception-control gap for real-world mobile robotics, reducing reliance on privileged data and simplifying deployment pipelines.
Abstract
Autonomous drone racing has risen as a chal- lenging robotic benchmark for testing the limits of learning, perception, planning, and control. Expert human pilots are able to fly a drone through a race track by mapping pixels from a single camera directly to control commands. Recent works in autonomous drone racing attempting direct pixel- to-commands control policies have relied on either interme- diate representations that simplify the observation space or performed extensive bootstrapping using Imitation Learning (IL). This paper leverages DreamerV3 to train visuomotor policies capable of agile flight through a racetrack using only pixels as observations. In contrast to model-free methods like PPO or SAC, which are sample-inefficient and struggle in this setting, our approach acquires drone racing skills from pixels. Notably, a perception-aware behaviour of actively steering the camera toward texture-rich gate regions emerges without the need of handcrafted reward terms for the viewing direction. Our experiments show in both, simulation and real-world flight using a hardware-in-the-loop setup with rendered image observations, how the proposed approach can be deployed on real quadrotors at speeds of up to 9 m/s. These results advance the state of pixel-based autonomous flight and demonstrate that MBRL offers a promising path for real-world robotics research. Video: https://www.youtube.com/watch?v=nctQ2rxZnIc