Caltech engineers have developed Neural-Fly, a tornado-resistant drone that can survive just about any kind of weather. More specifically, Neural-Fly includes a deep-learning method capable of helping drones cope with new and unknown wind conditions in real-time just by simply modifying a few key parameters. It was tested at Caltech’s Center for Autonomous Systems and Technologies (CAST) using its Real Weather Wind Tunnel.
Caltech’s Real Wind Tunnel is essentially a custom 10-foot-by-10-foot array consisting of over 1,200 small computer-controlled fans that enable engineers to simulate everything from a light gust to a tornado. Neural-Fly is able to navigate these challenges by using a separation strategy that only requires one to update a few parameters of the neural network in real-time. Remember that time the Blue Waters supercomputer was used to recreate the deadly EF-5 ‘El Reno’ tornado?
DJI FPV Combo – First-Person View Drone UAV Quadcopter with 4K Camera, S Flight Mode, Super-Wide 150° FOV, HD Low-Latency Transmission, Emergency Brake and Hover, Gray
- IMMERSIVE FLIGHT EXPERIENCE: Feel the thrill of immersive flight provided by the DJI FPV Goggles V2 and DJI FPV’s super-wide 150° FOV, giving you an ultra-smooth, real-time view of your flight.
- 4K/60FPS VIDEO: The DJI FPV aircraft can record 4K/60fps video at up to 120 Mbps, capturing crisp details that make footage look as exhilarating as the flight.
- BRAND-NEW S MODE: Get the dynamic look of FPV footage easily in S mode. This hybrid flight mode combines the freedom of flying manually with the simplified controls of previous DJI drones.
- ADVANCED SAFETY FEATURES: An auxiliary bottom light, Smart Return to Home (RTH), Low Battery RTH, and forward and downward obstacle sensing are all integrated into DJI FPV to help ensure a safe flight, even at high speeds.
- OCUSYNC 3.0 TRANSMISSION SYSTEM: Enjoy crystal-clear real-time and HD Low-Latency Transmission video, even at distances of up to 6.2 miles.
We have many different models derived from fluid mechanics, but achieving the right model fidelity and tuning that model for each vehicle, wind condition, and operating mode is challenging. On the other hand, existing machine learning methods require huge amounts of data to train yet do not match state-of-the-art flight performance achieved using classical physics-based methods. Moreover, adapting an entire deep neural network in real time is a huge, if not currently impossible task,” said Michael O’Connell, Caltech Graduate Student.