Dicke quantum batteries harness the collective effects of high-power charging, yet whether robust, near-perfect superabsorption can persist under mesoscopic disorder and limited global control in realistic solid-state platforms remains unclear. Using reinforcement learning (RL), we identify a time-dependent cavity frequency that enables efficient and robust charging of disordered Tavis-Cummings (Dicke-type) batteries, yielding an ensemble-averaged normalized ergotropy exceeding 0.97 with fluctuations below 0.02. We demonstrate superlinear scaling of the charging power for charging times scaling as $T\propto N^{-1/2}$. Notably, the learned controls are well described by simple few-segment, single-sweep ramps reminiscent of multilevel Landau-Zener driving, under which the dynamics remains concentrated in the collective bright manifold. Exploiting this feature, we introduce a truncated total-spin ($J$-irreducible-representation) description that faithfully captures the charging dynamics for system sizes up to $N\sim 30$. These findings demonstrate a scalable framework for designing high-performance Dicke batteries and highlight the versatility of RL in disordered many-body systems.

Close

PrevNext

View figure in article

• Cavity quantum electrodynamics
• Quantum description of light-matter interaction
• Machine learning