The goal of this report is to present you my semester project on signal generation for haptic interfaces using Reinforcement Learning algorithm. The aim of this project is to improve the signal generated by state of the art methods. The vibration are generated in an one dimensionnal aluminum plat of 250mm long. We use supervised learning with a database of those signals . Then we start the Reinforcement learning from pretrained networks and not from randomly initialized networks. We worked first in the time only domain with high dimension (5000 values) ouptut for our networks. Next we used fourier transform for dimension reduction. The signal generated by our process didn’t beat the signal from the database, they had similar performances. Then the Reinforcement learning approach, with the set of parameters we tried, can be used to generate new kind of signals as a GAN or an Auto encoder would do, but nothing more.