Direct DC motor speed control using Soft Actor–Critic reinforcement learning

Accurate and robust speed regulation of direct-current (DC) motors is essential in industrial drive systems operating under load disturbances and parameter uncertainty. Conventional proportional–integral (PI) control remains widely adopted; however, its performance depends strongly on gain tuning and may degrade when operating conditions change. This paper proposes a Soft Actor–Critic (SAC) reinforcement learning controller for direct DC motor speed regulation, where the learned policy generates the armature voltage command without relying on a predefined PI structure. For benchmarking, a Twin Delayed Deep Deterministic Policy Gradient (TD3) agent is implemented to adaptively tune PI gains under identical simulation conditions, including the same motor model, sampling time, reference profile, disturbance scenarios, and reward formulation. Simulation results show that SAC achieves substantially faster transient performance, reducing peak time from 0.2648 s to 0.0790 s (70.2%) and settling time from 2.3376 s to 0.5698 s (75.6%), while reducing steady-state error from 0.04% to 0.0001% (99.75%). Robustness tests under severe load disturbance (T_L=80 N m), increased inertia (+30%), and increased armature resistance (+20%) confirm that SAC maintains closed-loop stability and rejects disturbances faster than the TD3-tuned PI baseline. These results highlight the effectiveness of entropy-regularized direct policy learning for high-performance DC motor speed regulation under uncertainty.