Can you provide me ways to fast forward the simulations? Since I am using the MuJoCo model for reinforcement learning, I need to know ways so that the training can be done faster ?
You could use larger time steps, but make sure the system doesn't go unstable. Making the constraints a bit softer tends to allow larger time steps to be used without instability. I am assuming you are running simulations in parallel on all the cores of your processor? If not, you really should -- that is the best way to speed up RL. The idea is to create one mjModel and many mjData (one per thread). Then you can call mj_step in parallel, with a shared mjModel and thread-specific mjData.