The design brief, operating requirements, and research behind HARD.
The problem
Emergency crews often have to make decisions before they can see the full scene. Dispatch reports cannot show blocked routes, hazards, or where people may be. A single live camera feed gives a useful view but little sense of depth or the surrounding layout. Responders need information that arrives quickly and helps them understand the space.
Proposed approach
HARD is planned as a high-speed FPV drone with two synchronized cameras and an IMU. An FPGA would timestamp and package matched camera frames and IMU readings for a ground computer to run stereo SLAM. The pilot would retain a low-latency live view, while the map could give responders a better sense of the scene. Simulation, bench tests, and flight tests will guide the design.
Design requirements
The design targets 120 mph capability and around 30 minutes of flight. It would travel to the site in a horizontal fast-flight state, then transition upright to hover and move slowly while stereo cameras capture the surroundings.
The FPGA would synchronize stereo exposures where supported, timestamp images and IMU samples at capture, and package matched pairs with frame IDs for ground-computer SLAM.
Precedent programs
The first drone-as-first-responder program in the United States, launched in 2018. Drones fly from fixed sites to emergency calls and stream live video to officers and dispatchers.
A purpose-built 911 response drone that launches from a robotic charging station and streams live video as it arrives.
Skydio X10 uses multiple navigation cameras and visual-inertial odometry to understand its surroundings. Its vision system offers a reference for HARD’s stereo mapping goal.