Researchers at the Massachusetts Institute of Technology are using virtual reality technology to train drones in safer, cheaper environments.
The use of commercial drones has grown rapidly over the past four years. Breakthroughs in drone technology have reduced the price of key components, enabling filmmakers, researchers, and even amateurs to join the competition, leading to a thriving new market and offering A large number of use case scenarios.
Now, with the new dawn of package transportation (see Amazon Prime Air), engineers and developers have begun to work hard to develop more efficient automatic four-axis aircraft. Unfortunately, training high-tech, fast-moving machines to automatically detect and avoid physical objects is safe or cost-effective. Fortunately, several prominent individuals at the Massachusetts InsTItute of Technology have begun developing a new vr-based test program that can save thousands of hardware for manufacturers.
Researchers at the Massachusetts Institute of Technology (MIT) call it Flight Goggles, which uses virtual reality technology to let autonomous vehicles "see" a virtual world filled with navigation and dodge objects. In fact, they only fly in an empty space. This allows engineers to properly calibrate vehicle tracking components, adjust specific hardware and more without the risk of costly crashes.
Karaman Sertac, associate professor of aerospace at the Massachusetts Institute of Technology, said: "We believe that the development of drone technology will change the rules of the game, and the development of drones is fast. If anything, that is, this system can make autonomous cars more Sensitive, faster and more efficient."
Karaman continued: "When you want to do high-throughput calculations, even if you make the most subtle changes to the environment, it will cause the drone to crash. You can't learn in that environment. If you want to break through For speed and speed, you need some virtual reality environment."
In the early testing phase of the new system, the results were promising. In 10 flights, 361 flights passed through a narrow virtual window, and the drone only "crashed" three times. Once the camera on the plane is turned off and the vehicle passes real-world testing, the drone can walk through the open window for 119 times.
Karaman added: "It also has the same effect in reality, this is the program we programmed in the virtual environment, learning by making mistakes." But in the process, we did not break any real window.
Karaman and her team plan to reveal more information about their projects at the IEEE International Conference on Robotics and Automation next week.
The project was supported to some extent by the US Naval Research Office, the Massachusetts Institute of Technology Lincoln Laboratory, and NVIDIA.
Dongguan Tuojun Electronic Technology Co., Ltd , https://www.fibercablessupplier.com