Reinforcement Learning and Hardware in the Loop for Localized Vibrotactile Feedback in Haptic Surfaces

Touch is commonly used to mediate human-machine interactions, notably in the setting of Digital Musical Instruments (DMIs), where touch screens are prevalent. The lack of rich haptic feedback has an impact on the richness and quality of the interaction. Piezoelectric transducers can be integrated to induce localized vibrations over a stiff surface to reestablish the correct haptic exchange. For instance, the time-reversal method has been successfully used to create a localized peak on a thin-rigid surface by using one or several actuators. This paper presents a deep learning approach to optimize the time-reversed signals on a single-actuator setup. With only one transducer, the amplitude and contrast ratio are increased and the desired position of the localized peak is assured to be exact. It is shown that Reinforcement Learning can be applied to optimize a pre-trained Neural Network to achieve similar or even better results when compared with the state-of-the-art approach using time-reversed impulse responses.