Specification-Based Maneuvering of Quadcopters Through Hoops

Motivated by creating systems that can achieve satisfying multiple objectives through user input, this paper approaches the use of quadcopters to manuever through suspended hoops. Often in scenarios like automated farming or surveillance, autonomous robots will need to satisfy a number of objectives over a finite amount of time with constraints on its state space. We can represent any number of desired objectives with these suspended hoops and define high level specifications that ensure the quadcopter will execute trajectories to sasify our constraints.

In this work, we propose two methods of input to the end-to-end planner we have developed. Users can either specify a sequence of hoop/direction pairs where hoops are approximated as ellipsoids with fixed dimensions and paths are planned through spline interpolation through pre-defined waypoints around the hoops or users can  provide specifications in linear temporal logic (LTL), a logic formalism for specifying linear temporal tasks. Both methods allow users to give desired high-level instructions to a quadcopter. Given an LTL specification, we convert the specification into a Buchi automaton and search through the automaton for a satisfying sequence. If the trajectory satisfies the specification this trajectory is accepted. We also apply cross-entropy optimization techniques where the alogirthm then samples a set number of trajectories that satisfy the given specifications and finds an acceptable subset of trajectories that are below a threshold cost function. This process is iterative until the desired cost has been met by a single trajectory and the trajectory is executed by the quadcopter.



Comments are closed.