A group of scientists from the Massachusetts University of Technology (MIT) created a system called Flight Goggles for virtual drone exercises in orienteering. The essence of the system is that the drone “represents” obstacles and learns to bypass them while being in an empty room. Thanks to her, the number of device failures during test runs is reduced by several times.
The developers were inspired by the idea of creating automated copters that can compete with human-driven drone races. The following is a demonstration of how Flight Goggles works:
Key system features
The hangar for testing became a gym in one of the MIT buildings. It houses the camera capture motion to determine the position of the drone.
Flight Goggles includes a motion capture system, a program for photo-realistic image rendering and the hardware developed by the creators for quick processing and transferring data to a drone. Thanks to the powerful integrated supercomputer and gyrostabilizer, the image is transmitted at a frequency of 90 frames per second.
What do the tests show?
First, the development team conducted a series of experiments in a virtual environment. The goal is to fly through a window twice the size of a drone surrounded by a living room. After each attempt, the researchers improved the navigation algorithms. For 10 flights, the UAV successfully coped with the test 361 times, “crashed” three times. Drone scores were transmitted to scientists through motion capture cameras.
Then, the already trained drone was given a new task – to fly through the real window installed in the test room. The drone fulfilled the condition 119 times in 8 flights, 6 times “crashing” or demanding human intervention.
The developers claim that the next test will be navigation in the MIT Scala Center and flying around a person. In order to fulfill the last task, they propose to split the training ground into two parts: the first will be the drone, and the second – the person with the motion capture system. The data from the person will be transmitted to Flight Goggles, and the drone will “think” that the person is in the room with it.
This is not the first development of MIT scientists in the field of automated drones. Previously, they created a system for the effective movement of drones at high speeds.