Non linear path determinism

[alesspalma]

Hello, I am having an issue with reproducibility. Even when fixing all the python (and libraries) related seeds, I get different results between different runs of the same seed in RLBench/PyRep. This is apparently due to the intrinsic randomness in the PyRep's Arm get_nonlinear_path method, which every time is computing a different path even if the python's seeds are the same, and this is causing the objects in the scene to end in slightly different positions. Do you know how I can avoid this randomness by any chance? Lowering the distance_threshold parameter is not solving the problem. Thanks!