Skip to main content
This page is a stub. A human will curate and expand it before publishing.
The Sim tab’s Advanced Settings section sets how a simulation runs. Every field is optional — leave one empty and the server picks, using the value shown in the box. If your environment has an RL controller assigned, the blanks are filled with the values that policy was trained at.

​
The fields

  • Physics timestep (dt) — how far the simulator advances per step, in seconds. Smaller is more accurate and proportionally more expensive. Match your training dt if you care about matching a policy’s behaviour.
  • Pacing — whether one second of simulated time takes one second of real time. Leave on Auto unless you know you want otherwise; navigation and mapping stacks plan against the clock and need real time. Custom multiplier lets a headless run go faster than real time.
  • Stream frequency (Hz) — how often the simulation publishes robot state and camera frames to controllers. Keep it at or above your policy’s control rate.
  • Range sensor rate (Hz) — how often simulated LiDAR, radar and pointcloud sensors scan.
  • Range self-hit fallback — what a beam does when the first thing it meets is the robot’s own body, or something closer than the sensor’s minimum range. Off by default: the beam reports no return, which is what a physical scanner does when its own chassis is in the way. Turn it on and the beam continues past the obstruction and reports whatever is behind it — useful if you specifically want the older behaviour, but those are returns a real robot never produces.

​
If a simulation runs slow or stutters

Lower the range sensor rate first. Simulated range sensors are usually the most expensive thing in a scene: every scan traces a ray for each beam, and a high-resolution lidar scanning at full rate can consume most of a simulation’s budget on its own. Dropping it to a few Hz often restores smooth motion with no visible change to navigation. If that isn’t enough, in rough order:
  1. Check the range self-hit fallback is off. It is off by default, and on a robot whose scanner can see its own legs or wheels, resolving those beams costs far more than every other beam combined.
  2. Raise the physics timestep — this is the cheapest large gain. Watch for robots jittering against the floor or sinking into it, which means you have gone too far.
  3. Lower the stream frequency, as long as it stays at or above your controller’s rate.
  4. Reduce the sensor’s own resolution on the asset (its points-per-second and update rate), which lowers the cost of every individual scan rather than the number of scans.
Setting the range sensor rate to 0 removes the separate budget and scans on every stream pass — the older behaviour, and rarely what you want on a scene with a real lidar on it.

​
When you can change them

These settings are fixed for the life of a run. Set them before pressing Start; the fields are disabled while a simulation is active. Values you type are kept even if you collapse the section.