Arbitrary order of actuator dynamics

Discussion in 'Feature Requests' started by Jongwoo Lee, Jan 10, 2018.

  1. Hello,

    I've informed that MuJoCo is capable of simulating actuators upto 1st order dynamics with one actuator state. If necessary, would it be possible to increase the actuator system order to n-th order? To do so, I think I need to have access to a numerical integrator but it seems not.

    Do you have any plan to implement the feature? or could you guide me to customizing MuJoCo to implement that feature?
     
  2. Emo Todorov

    Emo Todorov Administrator Staff Member

    Yes you can use the "act" variables to model things like air dynamics in pneumatic cylinders, or activation states of biological muscles. If you want higher order dynamics you have to implement it yourself. The field mjData.userdata can be used to store user state variables (it is not touched by MuJoCo).

    Note that MuJoCo computes the rigid-body dynamics in continuous time: the function mj_forward() computes the acceleration in mjData.qacc. This is used by the built-in integrators (semi-implicit Euler and RK4), but you can bypass those integrators and implement your own.
     
  3. I understand. Again, thank you for your quick response.