Quantum control, particularly in the analog regime, is distinctively hard. The ability to control parameters on specific quantum hardware to achieve any logical operation or gate with minimal information leakage or fidelity loss has the potential to directly push quantum computing into the fault tolerant regime. Near-perfect control eliminates one of the largest sources of coherent errors in any quantum system and reduces the need for the additional gates arising from error correction needs.
However, due to dimensionality constraints, efficiently simulating combinations of controllable parameters for hardware-specific Hamiltonians becomes increasingly difficult. The number of potential trajectories grows exponentially and traversing this solution space becomes near impossible. We attempt to solve this problem by turning to novel machine learning techniques to learn these controllable system parameters at the pulse-level to achieve target gates on specific systems. We use a reinforcement learning policy model to learn these parameters robustly in the presence of system noise, while attempting to mitigate leakage and maximize gate fidelity. The output parameters from the trained policy are to be used to directly guide gate implementations on hardware.