jmichaux / gym-underactuated Goto Github PK

Should I be solving the discrete-time or continuous-time Ricatti equation?

Right now, the environments only allow linearizing the dynamics around a fixed point. I need to add figure out a way to linearize the dynamics around an arbitrary point as this will be useful from testing trajectory optimization algorithms.

It looks like there is a need to implement runge-kutta 45 for the acrobot/pendubot. I've noticed 3 things:

1. The solution will diverge from the fixed point, even in the absence of torque commands or perturbations.

2. From some initial states the pendula appear to gain kinetic energy.

3. From some initial states the dynamics seem to dampen, as if there were friction in the system.

Right now the pendulum does not adhere to the standard manipulator equation format. This needs to be changed.