Piezoelectric transducers are used within smart structures to create functions such as energy harvesting, wave propagation or vibration control to prevent human discomfort, material fatigue, and instability. The design of the structure becomes more complex with shape optimization and the integration of multiple transducers. Most active vibration control strategies require the tuning of multiple parameters. In addition, the optimization of control methods has to consider experimental uncertainties and the global effect of local actuation. This paper presents the use of a Deep Reinforcement Learning (DRL) algorithm to tune a pseudo lead-lag controller on an experimental smart cantilever beam. The algorithm is trained to maximize a reward function that represents the objective of vibration mitigation. An experimental model is estimated from measurements to accelerate the DRL’s interaction with the environment. The paper compares DRL tuning strategies with [Formula: see text] and [Formula: see text] norm minimization approaches. It demonstrates the efficiency of DRL tuning by comparing the control performance of the different tuning methods on the model and experimental setup.