Just like AI-driven automobiles, autonomous machines need to learn and navigate complex obstacles. MIT engineers designed a VR training ground specifically for AI drones.
Risk And Crash Reduction
When scientists and engineers started working on self-driving automobiles, they were ideally tested first in a virtual space. This removes the risks involved with road travel such as pedestrians and other factors. To reduce the rate wherein drones are repaired or replaced, the same principle was considered during its testing phase.
In turn, it keeps the research cost down and helps researchers with their machine-learning algorithm. The money saved could be used to upgrade components and more.
The Importance Of Virtual Reality
Unlike before, VR training programs are mostly used by humans to learn specific skills. Several modules for activities such as driving, flying, sailing, and others made it easier for instructors to teach students minus the risks and expenses.
Now, it seems that AI drones and other artificial intelligence-based machines can also benefit from these applications. The team at MIT calls the new project "Flight Goggles" and proclaim that the system is intended to lower the instances in which the autonomous drone crashes. The engineers just require a large empty space and program virtual objects into the 3D space scanned by the system.
"We think this is a game-changer in the development of drone technology, for drones that go fast," noted Sertac Karaman, MIT's associate professor of aeronautics and astronautics. "If anything, the system can make autonomous vehicles more responsive, faster, and more efficient."
New System, New Approach
According to the researchers, the Flight Googles system is a combination of an advanced image rendering program, a motion capture system, and custom-built circuit boards that house advanced processors to analyze all of the data to control the drone's flight.
Everything is mounted on a 3D-printed nylon and carbon fiber frame. It has a speed capability of up to 15 mph and can fly through a virtual obstacle course transmitted by the control team at an estimated 90 frames per second.
The project is supported by the MIT Lincoln Laboratory, NVIDIA, and the U.S. Office of Naval Research. It is expected to make its public debut at the IEEE International Conference on Robotics and Automation in Brisbane, Australia this week.
Advanced Navigation System
The MIT research team boasts that the VR training ground for its AI drone project is versatile. Its artificial intelligence system is apparently not limited to static objects only. Karaman claims that it will eventually learn to fly around moving obstacles such as humans.