It was only a matter of time. By rethinking AI algorithms, researchers at the University of Zurich have enabled a racing drone to beat its human-piloted competitors.
You will also be interested
[EN VIDÉO] An emergency medical drone to save lives Rwanda is one of the most rugged countries in Africa. Getting around quickly is difficult, especially when it comes to delivering medical supplies. Faced with the absence of a road, the best solution is certainly to go by air. This is why the Zipline company offers a drone delivery service. Here it is presented in video.
Identify obstacles, anticipate maneuvers, all in real time … Until now, humans still surpassed machines within the framework of drone races simulated or not. And as Futura clarified, it was only a matter of time for Artificial Intelligence to take over. It’s now done. In Switzerland, researchers fromUniversity of Zurich (UZH) developed an algorithm that allowed a drone Autonomous racing, powered by an AI, to beat not one, but two human remote pilots on an experimental racing circuit.
To achieve this, researchers have changed the way they think. Previous approaches were based on models making it possible to calculate the optimal trajectories to gain the upper hand over humans. But these paths did not take into account the constraints physical quadcopter drone and its positioning in space. In the end, this one could loseenergy and velocity to stay on these optimized axes and found himself slower than his opponents.
External cameras in support
With their HE, scientists always ask the drone to go through waypoints, but without giving it instructions on when to turn or accelerate, or to define its position in space. On these last points, it is the AI of the drone who manages his flight according to his own physical constraints.
To verify the relevance of their model, the researchers compared it to the maneuvers performed by the remote pilots on the Race circuit. Cameras were also placed on the circuit to indicate to the AI the exact position of the drone during its evolutions. By training the AI and comparing it to human maneuvers, the AI managed to beat the remote pilots every turn. The researchers’ idea is not so much to be faster than humans but to become more efficient in carrying out rescue operations or even deliveries.
Drone race: who won AI or humans?
Article by Louis Neveu, p
On the starting line, a racing drone piloted by the reigning world champion. Its competitor: a fully autonomous drone with sensors and boosted with artificial intelligence. For the loser of this race to become an unbeatable champion, it is only a matter of time.
Finally, Man still knows how to maintain his superiority over the machine. At least, for the moment … Last Friday, in Austin, Texas (United States), during the Drone Racing League, was held the first competition Man versus machine, it was the drone racing world champion who won the round.
The remote pilot, the Canadian Gabriel Kocher, known by the nickname Gab707, won the race hands down in six seconds with a course strewn with curves and obstacles. Its competitor, an autonomous drone doped with AI and developed by the Team team MAVLab from the Netherlands, took 11 seconds to complete the same course.
The autonomous drone, almost twice as slow
Achieving this result is in itself a feat for an autonomous drone. He must identify and avoid obstacles to great speed and in real time, and anticipate maneuvers. the drone from MAVLab was equipped with two Nvidia processors and four cameras. This drone had won the final of a race called Artificial Intelligence Robotic Racing (Airr) which only included autonomous drones with the same equipment.
As the winner, the team then faced off against the reigning human champion. If the MAVLab team lost the race, they still got a prize of $ 1 million for creating the fastest autonomous drone. If the Man still has a little leeway, the league expects that in 2023, the autonomous drones be able to beat the best human remote pilots.
Artificial intelligence comes to drone races
Article by Louis Neveu published on 02/19/2019
Unbeatable in the game of Go, chess and even Starcraft, the AI will soon embark on racing drones to fully independently face a human champion in a competition called Artificial intelligence robotic racing. The closing date for registrations for autonomous drone development teams is approaching, tension is mounting …
Confront drones totally autonomous to the champions of racing drones, this is the objective ofArtificial intelligence robotic racing (AIRR) organized by the Drone Racing League (DRL). As Futura had already specified, the project supported by the aircraft manufacturer Lockheed Martin and Nvidia door the name of AlphaPilot, which is reminiscent of the name given to Google’s AI, AlphaZero, AlphaGo… It is open to all drone enthusiasts or specialists in artificial intelligence (AI).
In the coming months, teams will have to develop an AI associated with machine learning algorithms so that the drone can evolve completely autonomously. To date, more than 300 teams have already registered. Registrations and qualifications take place from the platform HeroX. The deadline to participate is set for March 8. The selections will be made throughout the month of April. In any case, the common hardware base for designing the drone is Nivdia’s Jetson AGX platform which is dedicated to autonomous machines. The subtlety is that it is not possible to pre-program the navigation. The drones are therefore left to themselves during their race.
From virtual to real
First, the teams will fly their autonomous drones virtually on circuits boosted by AI and inspired by video games. It is also a team from MIT that developed the simulator that will be used for qualifying. This season of races will start on 1is August. At the end of it, the drone fastest autonomous will face the 2019 Allianz World Champion of the DRL league in real conditions.
The difficulty for the AlphaPilot teams will remain to successfully transpose their virtual inventions into reality. At the end of the adventure, there is a pretty carrot with over $ 2 million in cash for the most advanced and fastest AI. It remains to be seen whether on a real circuit, the AI will be more efficient than the best driver of racing drone from the moment.
AlphaPilot: drone pilots and AI will compete in a real race
Article by Marc Zaffagni published on 09/11/2018
Launched by Lockheed Martin, the Drone Racing League and NVidia, AlphaPilot is a drone racing project that will pit human pilots against artificial intelligences. For the defense and aerospace giant, the challenge is to prepare the technologies of tomorrow.
Starting next year, humans will face artificial intelligences (IA) during races drones. Baptized AlphaPilot, this project is the result of a rather surprising association between Lockheed Martin, the Drone Racing League (professional drone league) and NVidia. The competition is endowed with two million dollars of prizes, with in particular a bonus of 250,000 dollars to the first team whose autonomous drone will manage to beat a human pilot.
The races will take place under a new format launched by the Drone Racing League appointed Artificial intelligence robotic racing (AIRR). AlphaPilot is open to all drone enthusiasts or specialists in artificial intelligence, whether they are students, researchers or employees of a company. They will have to create an AI and a machine learning algorithm (machine learning) capable of fly a drone on the various routes offered without prior navigation programming.
Participants will be provided with all the necessary equipment so that everyone has the same basic equipment. In particular, they will have to use NVidia’s Jetson computing platform, which allows them to create applications of on-board artificial intelligence forautomobile, robotics or drones. In its press release, Lockheed Martin specifies that its engineers will play the role of ” mentors »For the teams involved.
The presentation video of the AlphaPilot competition. © Lockheed Martin Videos
A way for Lockheed Martin to do R&D
« By entering this competition, your knowledge and ideas can directly contribute to the future of autonomous transportation, delivery, disaster relief and even space exploration. », Explains the company specializing in defense systems and aerospace. In short, AlphaPilot is a way for Lockheed Martin to do R&D, find future talents and advance on the technologies that will equip its products of tomorrow.
Creating an autonomous drone capable of competing with the reflexes and dexterity of human pilots is obviously no easy task. But the experiment has already been tried with some success. Last year the Jet Propulsion Laboratory from Nasa had organized a race between autonomous drones and a human pilot. If the latter had won with several seconds of difference, the drones driven by an AI had demonstrated great consistency in their trajectories. However, as with the game of Go, chess and video games, human unpredictability and the ability to adapt will be the major obstacles that the machine will have to overcome in order to be successful.
Interested in what you just read?
.
