I'm running a simulation of an arm robot traversing and eventually colliding with a body on a frictionless table. I have force torque sensors located at two interfaces (one near the end of the robot arm and one near the point of contact on the body sliding on the frictionless table). I will simulate the motion and contact dynamics and log the joint angles, link locations, contact sensor messages, and force torque sensor readings. When I run the simulation twice I get concurrence in the link, joint, and contact data logs but the noise/bias-free force torque sensor loggings are inconsistent between simulations under identical conditions. I was wondering if anyone had observed this previously or if anyone had any ideas on where the non-repeatability could be coming into play.