Reproduce “Build Your Own Visual-Inertial Drone: A Cost-Effective and Open-Source Autonomous Drone” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Inkyu Sa, Mina Kamel, Michael Burri, Michael Bloesch, Raghav Khanna, Marija Popovic, Juan Nieto, Roland Siegwart
Researchers at ETH Zurich's Autonomous Systems Lab developed a cost-effective and open-source autonomous drone platform by integrating a commercial DJI Matrice 100 with an Intel ZR300 visual-inertial sensor and a custom software stack. The system achieved robust control performance with an outdoor hovering pose error of 0.049 m in 11-11.5 m/s wind and demonstrated precise state estimation with a 0.82% drift over an 180 m outdoor flight.
Key insights
- Overcoming commercial drone limitations for research requires direct modification of SDKs to reduce control latency and compensation for proprietary autopilot features like auto-trim and dead zones.
- Combining a state-of-the-art VIO framework (ROVIO) with multi-sensor fusion effectively provides robust and accurate ego-motion estimation suitable for complex autonomous tasks on affordable hardware.
- Batch-based system identification and non-linear Model Predictive Control can effectively enable high-performance control on platforms with undisclosed internal dynamics, even under significant external disturbances.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.