We introduce a new version for the quantum version of the standard reinforcement learning benchmarking problem: the cartpole. The goal is to keep a wavefunction stable on top of an inverted potential, by applying weak measurements and feeding the results to a controller. Our approach shows, that even with minimal measurement information, it is possible to extend the lifetime of the wavefunction, by a significant margin. Our approach works for arbitrary potentials and due to weak measurements makes it feasible for experimental implementation.

Download the slides for this talk.Download ( PDF, 1362.42 MB)