eduzhai > Applied Sciences > Engineering >

Learning-to-Fly Learning-based Collision Avoidance for Scalable Urban Air Mobility

  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: With increasing urban population, there is global interest in Urban AirMobility (UAM), where hundreds of autonomous Unmanned Aircraft Systems (UAS)execute missions in the airspace above cities. Unlike traditionalhuman-in-the-loop air traffic management, UAM requires decentralized autonomousapproaches that scale for an order of magnitude higher aircraft densities andare applicable to urban settings. We present Learning-to-Fly (L2F), adecentralized on-demand airborne collision avoidance framework for multiple UASthat allows them to independently plan and safely execute missions withspatial, temporal and reactive objectives expressed using Signal TemporalLogic. We formulate the problem of predictively avoiding collisions between twoUAS without violating mission objectives as a Mixed Integer Linear Program(MILP).This however is intractable to solve online. Instead, we develop L2F, atwo-stage collision avoidance method that consists of: 1) a learning-baseddecision-making scheme and 2) a distributed, linear programming-based UAScontrol algorithm. Through extensive simulations, we show the real-timeapplicability of our method which is $ approx !6000 times$ faster than the MILPapproach and can resolve $100 $ of collisions when there is ample room tomaneuver, and shows graceful degradation in performance otherwise. We alsocompare L2F to two other methods and demonstrate an implementation onquad-rotor robots.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×