Deep Reinforcement Learning Control of Quantum Cartpoles

TL;DR

This paper demonstrates that deep reinforcement learning can effectively control quantum systems, specifically stabilizing a quantum particle in an unstable potential, and performs comparably or better than traditional control methods.

Contribution

It introduces a deep learning-based approach for quantum control, extending classical reinforcement learning benchmarks to the quantum domain with successful stabilization and cooling applications.

Findings

Deep reinforcement learning stabilizes a quantum particle in an unstable potential.

The approach performs on par or better than classical control strategies.

It is applicable to measurement-feedback cooling of quantum oscillators.

Abstract

We generalize a standard benchmark of reinforcement learning, the classical cartpole balancing problem, to the quantum regime by stabilizing a particle in an unstable potential through measurement and feedback. We use state-of-the-art deep reinforcement learning to stabilize a quantum cartpole and find that our deep learning approach performs comparably to or better than other strategies in standard control theory. Our approach also applies to measurement-feedback cooling of quantum oscillators, showing the applicability of deep learning to general continuous-space quantum control.