Hi Emo. Thanks for all your help in the past. I am a reinforcement learning researcher with only a very limited background in robotics. I have a robot that is learning to grasp a block. My suspicion is that my model is currently very inefficient. In particular I am trying to improve the settings on the actuators/joints. Here are the things (by order of priority) that I would like to improve: 1. Stability: if nothing else, the model must not throw a Mujoco error. 2. Performance: each call to mj_step should take as little time as possible. 3. Speed: each call to mj_step should move the robot as much as possible -- the fewer steps to grasp the block, the better. 4. Control: the agent should have tight control over the robot's movements by setting the ctrl parameter (for example, I would like to minimize oscillations for position servos). 5. Realism: this is not too great a concern, but I would, for example, like the floor to behave more like linoleum than ice. I know this is kind of a broad, unfocused question so I especially appreciate your help.