Hello, I am trying to simulate 18 DOF quadrupedal robot (the model is as the following image) using MuJoCo. (GUI was developed by myself.) In order to check energy conservation, I applied the z-direction force for certain amount of time and then dropped the robot and check if the energy is conserved or not during the free drop. You can see the video in the link: There's no internal collision, and the timestep is 0.005 second. Since there's no collision, I expected the total energy would be a constant, but it reduces as the following plot. I tested both Euler and RK4 integrator, but the results are almost same. Am I missing something here? Could you give me any suggestion to make the total energy of simulation is conserved?