search

pantor / ruckig

Motion Generation for Robots and Machines. Real-time. Jerk-constrained. Time-optimal.
https://ruckig.com
MIT License
634 stars 155 forks source link

Flaky oscillations when Ruckig timestep is not constant #104

Closed AndyZe closed 2 years ago

AndyZe commented 2 years ago

With a constant timestep, we get nice behavior out of Ruckig. But if the input has a different timestep at every iteration, we occasionally see oscillations. Got any insight on this?

Constant timestep on the left, dynamic timestep on the right: ruckig_dynamic_timestep

Zoomed on the unexpected oscillation: zoomed

Here's the code used for the dynamic timesteps. It just constructs a new Ruckig instance with a different timestep at every waypoint:

      timestep = trajectory.getWayPointDurationFromPrevious(waypoint_idx + 1);
      ruckig_ptr = std::make_unique<ruckig::Ruckig<RUCKIG_DYNAMIC_DOF>>(num_dof, timestep);
AndyZe commented 2 years ago

For the nice constant timestep graph, the timestep was 0.1s. For the wacky dynamic timestep graph, it ranged from 0.03-0.25s.

pantor commented 2 years ago

Hey @AndyZe! Can you provide me a bit more information here, maybe even a simple code example? Are the figures showing the position over time? Ruckig is not really meant to be used with dynamic time steps, so I'm not exactly sure what's going on here.

pantor commented 2 years ago

Feel free to reopen, but I think I need more information to help here.