After chess, the Jeopardy, the Go game, researchers are now interested in a new sport, the FPV Racing, i. e. the drones race. Researchers at the JPL (Jet Propulsion Laboratory) in Pasadena, a NASA laboratory, have unveiled the results of more than two years of research and the result is stunning. Confronted with an experienced pilot, the IA doesn’t have the weight, but it does perform well on a rather complex course, and she rides this competition drone probably more quickly than many of us.
Those who were able to attend Roborace’s first demonstrations at the grand opening of the Formula E Grand Prix were able to see the work that remains to be done by the designers of these robot cars before offering a show worthy of a grand prize.
In the field of FPV Racing, things seem to go much faster (in every sense of the word). The work unveiled by JPL researchers is quite spectacular; they confronted their AI with a seasoned pilot, Ken Loo. if he defeated the AI, the drone performed the course at high speed without a hitch. The AI had an average time of 13.9 seconds on the course, compared to 11.1 seconds for the human pilot.
This Google-funded search uses Google Tango technology, the drone that uses the on-board camera to locate itself on the course, in the same way as the human pilot. Three UAVs have been assembled (Batman, Joker and Nightwing), quadrotors capable of flying at speeds of up to 129 km/h, but limited here at 48/64 km/h. Researchers believe that such high-performance real-time positioning via the camera and mapping of locations in place of GPS could have applications in other areas, such as navigation in warehouses, disaster sites or the corridors of the international space station.
Source: “Drone Race: Human Versus Artificial Intelligence”, JPL, 21 November 2017