Sandrine Ceurstemont, editor, New Scientist TV
A pro skier is likely to have practised for years, and endured a few falls, before winning a slalom race. Now a minicopter is taking inspiration from this approach by learning from its mistakes to perfect a route through an obstacle course.
Developed by Angela Schoellig and colleagues from the Swiss Federal Institute of Technology Zurich, the drone initially flies between a series of poles by following a pre-computed path. But because of real-world inconsistencies, such as unpredictable aerodynamics, it will deviate slightly from the prescribed trajectory. To help it zigzag more accurately, cameras track its motion and provide feedback. "It's similar to what humans do intuitively," says Schoellig. "If a tennis player plays a ball too short, he knows that he must hit harder the next time."
By using this approach, a minicopter typically masters any obstacle course within 10 tries. Since its full trajectory is updated after each trial, several previously unknown effects can be incorporated instead of only considering a single type of error. In future versions of the system, the drone may not need to start each flight from scratch. "It would be interesting to investigate if the quadrocopter can use information obtained from learning one slalom course to start off better when learning a different course," says Schoellig.
As drones become more sophisticated, we'll be able to trust them to perform a range tasks, from flying next door to borrow a cup of sugar to decommissioning a nuclear reactor. For more about how flying robot servants will change our lives, check out our full-length feature, "Welcome to the personal drone revolution".
If you enjoyed this post, watch flying robots team up to juggle or see a swarm of minicopters perform the James Bond theme.
Slaloming robot minicopter learns from its mistakes
This article
Slaloming robot minicopter learns from its mistakes
can be opened in url
http://newsstonefish.blogspot.com/2012/12/slaloming-robot-minicopter-learns-from.html
Slaloming robot minicopter learns from its mistakes
