Two algorithms being developed at MIT could soon enable unmanned aerial vehicles to negotiate their way around obstacles automatically.
In December the United States Federal Aviation Administration (FAA) made it obligatory for all US residents to register their unmanned air systems (UACs), a move intended as a first step towards the enactment of legislation that will enable widespread use of small drones. While drones have recently been in the spotlight as potential delivery vehicles, with Amazon, Google and Walmart announcing plans to use them in this way, their potential goes much further, extending to mining and other resource exploration, agricultural crop monitoring, rescue services and assistance to the police, as well as for transporting people. In addition, solar powered drones could replace satellites, and drones have even been used to create artistic performances.
However, a number of hurdles still need to be cleared, especially the risk of accidents. Whether they are steered by human beings – who are not always exactly air aces – or automatically by software, unmanned aerial vehicles do frequently crash, with the risk of injury to passers-by, destruction of property, not to mention the losses to their owners.
Now two students at the Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed two complementary algorithms which could soon enable unmanned drones to fly with perfect safety. The first enables the drone to analyse its environment and plan a route that avoids all obstacles. Unlike most algorithms of this type, it does not pinpoint the obstacles but conversely identifies empty spaces, a technique which its creator, 3D Robotics control system expert Benoit Landry, claims is a more efficient approach for small crowded spaces.
Nevertheless, in order to make its way around, the drone requires the second algorithm, developed by Anirudha Majumdar, a PhD student working on Robot Locomotion at MIT CSAIL. It combines a number of flying routines that the drone can put together, pulling off manoeuvres worthy of aeronautical display experts. The algorithm selects the manoeuvre according to the drone’s current position and destination, taking into account such variables as the environment and the wind speed. The two algorithms are not yet ready for the market: the first is still not able to perform the necessary calculations fast enough while the second only offers a limited range of manoeuvres at the present time. Nevertheless, in the same vein as the drone route-mapping project on which L’Atelier reported recently, the work of Landry and Majumdar seems likely to help make drones safer and so promote more widespread use of these aerial vehicles.