I tried to construct a simulation env following fetch_pick_and_place. I noticed that the following code is used to initialize the env:

    for _ in range(10):
        self._mujoco.mj_step(self.model, self.data, nstep=self.n_substeps)

Similarly, I followed the above code to initialize my own env but got inaccurate configurations. As I reduced the number of loops, I got configurations closer to the desired configuration.

So, my confusion is why more mj_steps lead to inaccurate simulation results.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.